TruIntel Platform Documentation

TruIntel is an AI Engine Optimization (AEO) platform for the era of AI-assisted search. As buyers shift from typing keywords into Google to asking ChatGPT, Claude, Gemini, Perplexity, and Google AI Overview, TruIntel measures and improves how those AI engines perceive, rank, and cite your brand. It pairs that AEO core with a full SEO toolkit, an AI-optimized CMS, traffic and lead intelligence, and automated weekly reporting.

Where traditional SEO optimizes for one ranking algorithm, AEO optimizes for the way large language models recommend brands inside conversational answers. TruIntel runs your real customer questions through all five AI engines on a schedule, parses every answer for whether you are recognized, where you rank, the sentiment around you, and whether a source links back to you, then turns that into a single Visibility Score and a queue of fixes.

ChatGPT

OpenAI gpt-5.4-mini via the Responses API with the web_search tool.

Claude

Anthropic claude-haiku-4-5 with server-side web search enabled.

Gemini

Google gemini-3.1-flash-lite with Google Search grounding.

Perplexity

Perplexity sonar, routed as perplexity/sonar through OpenRouter.

Google AI Overview

Google's own AI Overview, captured via SearchAPI.io (not a model TruIntel runs).

You can get your first AI visibility check running in a single sitting. The flow moves from account creation, through choosing a plan, into a guided brand-setup wizard that ends by firing your first full check across all five AI engines.

TruIntel uses JWT-based authentication with a short-lived access token and a rotating refresh token. You can sign up with email and password or with Google, and invited teammates join automatically when they verify or sign in for the first time.

Password requirements

• 8 to 100 characters long.
• At least one uppercase letter, one lowercase letter, and one digit (no symbol is required).
• Stored hashed with bcrypt; passwords are never persisted in plain text.

OTP lifecycle

• Six numeric digits, delivered by email via Resend.
• Valid for 10 minutes; a new request resets the attempt counter.
• Up to 5 wrong attempts — on the fifth the code is invalidated and you must request a new one.
• Used for both email verification at signup and password reset.

An organization is the top-level workspace that owns your brands, team members, plan, and billing. Understanding how organizations resolve and switch is key for anyone who belongs to more than one.

• Each user can OWN exactly one organization — creating a second returns an error. The creator becomes the OWNER.
• A user can BELONG to many organizations through invitations, even though they can only own one.
• A newly created organization starts with no plan, so plan selection is the immediate next step.
• Switching organizations clears the selected-brand and query caches so you always see the right workspace data.

How your current organization is resolved

When you load the app, TruIntel picks your active workspace using a single deterministic rule (never an arbitrary first-membership pick), which prevents the intermittent plan-bounce issues common to multi-org accounts.

Onboarding is a guided wizard that takes you from a brand-new account to a live, monitored brand. The organization step is skipped automatically if you already have a workspace, and progress is saved so you can resume where you left off.

A brand is the unit TruIntel monitors: it carries the queries, competitors, sources, SEO data, content, and reports tracked for one website. Your plan determines how many brands an organization can run at once.

• Brand fields include name, domain, description, industry, country, a default country code (used to localize AI prompts), logo, and the optional Brand Voice profile.
• The active-brand count is checked against your plan limit on every create — exceeding it returns a brand-limit error prompting an upgrade.
• A brand stays in onboarding until you complete it; until then the app routes you back into the wizard so setup is never left half-done.
• Completing onboarding on a paid plan also generates a one-time onboarding brand report.

TruIntel offers a permanent Free tier plus four paid plans. Pricing is monthly and billed in USD through Razorpay. The Free plan is a real, hard-limited tier — not a time-limited trial.

Enterprise is a custom plan (contact sales) with 4 brands, 50 queries per brand, 5 competitors, up to 10 team members, 1000 AI credits per month, and the highest daily-priority check allowance.

Organizations support multiple users with three roles: owner, admin, and member. Role checks gate only organization-management actions — they do not restrict which brand features a teammate can use. Feature access is gated by your plan, not by role.

Inviting teammates

AI Visibility Monitoring is TruIntel's core feature. For every brand it runs your tracking queries (prompts) against 5 AI answer engines — ChatGPT, Claude, Gemini, Perplexity, and Google AI Overview — and measures whether, where, and how each engine surfaces your brand. Results are stored as time-series so you can watch trends, get alerts, and prove progress.

Each check produces a composite AI Visibility Score (0–100), a query x platform performance matrix, per-platform scores, brand perception (attributes + sentiment), Share of Voice vs competitors, and cited-source tracking. The score is computed by deterministic Python text-parsing — there is no LLM at scoring time, so the same responses always produce the same number.

TruIntel queries 5 AI answer engines concurrently on every check. The exact production models are listed below and are visible in each mention's "Run details" panel. The first three (ChatGPT / Claude / Gemini) are environment-overridable; Perplexity and Google AI Overview routing are fixed. The same models run on every plan, including Free — models are global, not plan-tiered.

ChatGPT

gpt-5.4-mini via the OpenAI Responses API with the web_search tool (default search context "low"). The most widely used AI assistant.

Claude

claude-haiku-4-5 (Haiku 4.5) via the Anthropic Messages API with the web_search_20250305 server tool (max 5 searches per query).

Gemini

gemini-3.1-flash-lite via the google-genai SDK with Google Search grounding; falls back to non-grounded mode on a billing or quota error.

Perplexity

perplexity/sonar routed through OpenRouter (not Perplexity's own API), returning built-in source citations from real-time web access.

Google AI Overview

Google's own AI-generated search summary, scraped via SearchAPI.io in a two-step google -> AI-Overview flow. Many queries simply have no AI Overview — a normal, non-failing result.

How each platform is queried

For every query x platform pair, the pipeline parses the AI response with pure text/regex analysis (no LLM, under ~50 ms) and stores a structured mention record. Competitors are parsed first, so they are captured even when your brand is not mentioned.

Query intent taxonomy

Every query is auto-classified into one of 5 intents at creation time (via Kimi K2.5, defaulting to informational if unavailable). The intent automatically sets the query's business value (1–5) — this is derived, not manually assigned.

The Query Performance Matrix shows how each tracking query performs across all 5 platforms. Rows are queries, columns are the 5 platforms plus an "Avg #" column, and each cell is a status pill (not just a shaded square). The AI Visibility page shows the first 15 active queries (newest first) with a "View all N queries" link to Prompt Tracking.

Hovering a mentioned cell shows position, mention type, sentiment, and Share of Voice. Clicking a row opens the Query Detail view — platform tabs, the full AI response with your brand highlighted, the citation list, competitors found on that platform, and run telemetry (model, latency, tokens, Live/Cached).

TruIntel surfaces the attributes and sentiment that AI engines associate with your brand — effectively the keywords AI uses to describe you. The Brand Perception table is built from BrandAttribute rows from the latest full visibility check only, grouped by attribute + sentiment.

• Brand attributes — up to 3 per response (from Kimi K2.5), each with its sentiment, an occurrence count, and which platforms produced it.
• A per-platform dropdown (All / ChatGPT / Claude / Gemini / Perplexity / Google AIO) filters the table client-side.
• Sentiment distribution — positive / neutral / negative mention counts, available both overall and per platform.
• Mention types — recommended / alternative / compared / mentioned / not_mentioned, refined by the Kimi enrichment pass.

Share of Voice (SoV) measures your brand's slice of total brand attention within a single AI response, relative to all tracked competitors. It is computed deterministically (no LLM): for each response, SoV% = your brand-sentences / total brand-sentences (yours + competitors) x 100, using sentence-level whole-word matching.

• SoV is computed per mention at store time and saved on each QueryMention.
• The Share of Voice card shows the average SoV per platform across the visible queries (a per-platform bar).
• Per-mention SoV also appears in the query-detail tooltip and sidebar.
• Competitor mention counts and head-to-head detail live on the Competitors page.

Every URL in every response is extracted as a CitedSource and aggregated into a SourceDomain with citation count, usage %, and first/last-seen dates. This shows which websites AI relies on in your industry and whether your own site is among them.

Source domain types

Each cited domain is classified into one of 8 types. Classification runs in order: your own domain -> corporate; a tracked competitor -> competitor; hardcoded lists and institutional suffixes (.gov/.edu) for editorial / reference / ugc / marketplace / institutional; otherwise "other", which a background Kimi K2.5 pass later reclassifies. Domain types are cached for 7 days.

Gap analysis

Source Gap Analysis finds domains cited in responses where your brand was NOT mentioned but a competitor WAS — high-value outreach targets. If none are found, it falls back to your own low-citation domains. These results feed Off-Page Outreach.

How many queries, brands, and competitors you can track — and how often checks run automatically — depends on your plan. All 5 platforms are available on every plan, including Free. Free runs exactly one onboarding visibility check and has no recurring schedule.

Each visibility check runs a fixed processing pipeline. The score itself is pure deterministic text-parsing; the Kimi K2.5 enrichment and domain reclassification are non-blocking add-ons that never change the number.

Alert types

Five alert types are generated after a check. Only HIGH/CRITICAL alerts trigger an email to members.

Platform clients make a single attempt; all persistent retry is owned by a dedicated retry service so that "failed" means the same thing everywhere — in the matrix pills, query detail, and retry counts. Silent empty-200 responses are caught and surfaced honestly instead of showing an infinite spinner.

Auto-retry fires immediately after a check for transient failures, then backs off across 30s / 2m / 5m / 15m / 1h for up to 6 attempts before marking the cell "exhausted". A 60-second beat sweeper re-runs any brand whose next retry is due. Each retry inserts a new mention row, so history is preserved and the original aggregate score is not mutated.

• Per-cell retry — click the amber retry pill on a matrix cell to re-run that one (query, platform) inline; a manual single-cell retry resets the chain so 6 fresh auto-retries can follow.
• "Retry all failed" button — sweeps every manually-recoverable failure (exhausted, in-backoff, credits-depleted, silent-empty); only configuration and invalid_request are excluded because a retry cannot fix them.
• The "Retry N failed" count equals exactly what the button re-runs (pending + stuck), so it never lies.
• An in-flight Redis lock collapses double-clicks and overlapping sweeps into a single provider call, and the matrix spinner is server-derived so it survives page navigation and device switches.

The AI Crawlability / Site Audit view (at /visibility/site-audit) checks how well AI crawlers can reach and read your site — a prerequisite for being cited at all. It combines site-health signals with per-vendor LLM-bot access derived from your robots.txt.

AI Visibility spans several pages in the AI Visibility navigation section. Free can reach AI Visibility, AI Crawlability/Site Audit, Prompt Tracking, Prompt Research, Sources, and Competitors; richer surfaces such as Prompt Volumes are paid-only.

Key API endpoints

# Run and watch a check
POST /api/v1/visibility/check
GET  /api/v1/visibility/check/{task_id}/status
GET  /api/v1/visibility/check/{task_id}/stream   # SSE live progress

# Scores, history, mentions
GET  /api/v1/visibility/score/{brand_id}
GET  /api/v1/visibility/history/{brand_id}
GET  /api/v1/visibility/checks/{brand_id}
GET  /api/v1/visibility/checks/{check_id}/details
GET  /api/v1/visibility/mentions/{brand_id}
GET  /api/v1/visibility/mentions/{brand_id}/{query_id}
GET  /api/v1/visibility/response/{mention_id}
GET  /api/v1/visibility/{brand_id}/site-audit

# Matrix, retries, sources (dashboard + sources routers)
GET  /api/v1/dashboard/{brand_id}/queries          # performance matrix
POST /api/v1/dashboard/{brand_id}/queries/{query_id}/regenerate  # single cell
POST /api/v1/dashboard/{brand_id}/queries/retry-failed           # manual sweep
GET  /api/v1/dashboard/{brand_id}/retry-status                   # badge polling
GET  /api/v1/sources/{brand_id}/type-stats
GET  /api/v1/sources/{brand_id}/gap

A tracking query (also called a "prompt") is a buyer-style question that TruIntel runs across the five AI platforms on a schedule to measure your brand visibility. Each query represents something a potential customer might genuinely ask an AI assistant about your category — for example "What is the best CRM for small teams?" rather than a keyword.

Every query stores its text, country/locale (default India), tags, status, intent, an auto-derived business value, and an optional topic. Queries are managed on the Prompt Tracking page at /queries; each query has a detail view at /queries/:id showing its visibility trend over time.

Every query has one of three statuses, and the status plus the priority flag together decide when (and whether) the query is checked across the AI platforms.

Your plan limit counts only non-inactive queries (active + suggested), so pausing a query frees a slot. The check cadence then depends on whether the query is flagged as priority:

When a query is created, TruIntel classifies its intent exactly once and caches the result on the row. Business value is then auto-derived from that intent — it is never a number you set yourself. The classifier runs on Kimi K2.5 (moonshotai/kimi-k2.5) via OpenRouter and falls back to informational / value 2 if classification ever fails.

The effective business-value range is therefore 2-5 (never a user-assigned 1-5). In the UI, a QueryIntentBadge shows the intent and a "$" / "$$" badge marks high-value queries (value >= 4 shows $, value 5 shows $$).

Bulk-created queries (from onboarding) are classified in the background a few seconds after they are saved, so their intent badges fill in shortly after the queries appear.

The "Add Prompt" modal has two tabs: Manual (type a query yourself) and AI (generate suggestions). There is no CSV upload — bulk creation happens through the AI-generation and onboarding flows.

Topics group related queries so you can slice visibility by theme across the dashboard. Each query belongs to exactly one topic, and topics carry a denormalized query count.

• Topics are unlimited on every plan — there is no per-plan topic cap.
• Create, rename, and delete topics from the Topics Management modal; duplicate names are blocked case-insensitively.
• Deleting a topic does not delete its queries — their topic is simply set to none (they keep all history, they just lose the grouping).
• Filter visibility data by topic anywhere the topic filter appears to focus on one area of your business.
• Topics also help the AI Insights engine group related findings.

TruIntel can generate topics and prompts for you using its provider-agnostic utility-LLM chain (Anthropic-first, falling back to Gemini then OpenAI). These are an on-demand onboarding flow — every "suggest" call is preview-only and saves nothing until you accept selections through the bulk endpoints. They are not a continuously refreshed feed.

• Topics-with-prompts: one shot that returns N topics each pre-filled with K prompts (the current onboarding step).
• Topics only, then prompts-per-topic: the legacy two-step flow for expanding one topic at a time.
• Generate: produces queries for topics you have already saved, distributed across your plan's per-brand query budget.
• Generated prompts are deliberately brand-agnostic (your brand name is forbidden inside the query text) and localized to your brand's country.

How many topics and prompts you can generate depends on your plan:

Prompt Research (/prompt-research) is an ideation tool that turns keywords into ready-to-track prompt ideas. You enter a list of keywords and TruIntel returns up to 15 natural-language prompt ideas, each with suggested tags and an intent label, that you can add straight into tracking with one click.

• Input is a set of keywords (not a single seed topic).
• Output is 15 flat prompt ideas — there are no topic clusters.
• Each idea carries up to 3 tags and an intent of transactional, comparison, recommendation, or informational.
• Generation runs on the utility-LLM chain (Anthropic-first fallback) and is stateless apart from saving a history row.
• Past runs are saved: browse the research history and open any past run to see its full prompt list.

Query capacity scales with your plan. Limits below are per brand — both the per-brand query cap and the daily priority slots are counted per brand.

When an AI platform answers one of your tracked queries, it often cites source domains and URLs. Source Tracking (/sources) records every one of those citations so you can see exactly which sites AI engines lean on when they describe your category — and which ones mention you versus your competitors.

TruIntel classifies every cited domain into one of eight stored source types. A domain is typed once and the result is cached, so each domain is classified only a single time.

Classification follows a fixed order: (1) if the domain matches your brand it becomes corporate, and if it matches a tracked competitor it becomes competitor; (2) hardcoded lists / first-seen typing handle well-known hosts; (3) anything still "other" is sent to a Kimi K2.5 (via OpenRouter) classifier, whose output is cached. The LLM can return any type except competitor (that one is reserved for domain matching).

The Sources page opens with a five-cell stat strip, then a distribution donut, a gap-opportunity table, and the full citation table.

• Stat strip: Total sources, Earned Media, Corporate-Owned, Competitor, and Gap Opportunities.
• Source Type Distribution donut: citations broken down by type (interactive donut).
• "Sources You Should Target": the gap table, with a per-row "Draft outreach" button that hands the domain to the Outreach workflow.
• "All Citation Sources": the full table with domain search, a type dropdown (All / Earned Media / Owned-Corporate / Editorial / UGC / Reference / Institutional / Marketplace / Competitor), a platform filter, and sorting by Most cited, Last seen, First seen, or Highest usage %.

Click any row to open the Source Detail modal, which lists that domain's top cited URLs, which AI platforms cite each one, and the first/last seen dates.

Gap Analysis surfaces high-authority domains that AI cites in responses where your competitors appear but your brand does not. These are your highest-leverage outreach targets — places AI already trusts in your category that have not yet picked you up.

Competitor Tracking (/competitors) measures how your rivals show up in AI answers alongside you. You add competitors (manually or from AI discovery) and TruIntel records every time one of them is mentioned in a response to your tracked queries — position, sentiment, snippet, and platform.

Competitors move through a suggested → active → rejected lifecycle. AI discovery proposes candidates; you decide which to track.

The Visibility Comparison table shows you side by side with each tracked competitor across the latest full check window. A status PillTabs filter lets you view suggested, active, or rejected competitors.

Beyond the comparison table, the Competitors page offers four deeper analysis surfaces.

How many competitors you can actively track depends on your plan. Unknown plans fall back to Starter limits.

Prompt Research is an AI prompt-idea generator. You type 1-20 seed keywords and an LLM returns up to 15 natural-language search prompts — the kinds of questions real people ask AI assistants like ChatGPT, Claude, Gemini and Perplexity in your industry. Each prompt comes with topic tags and an intent label, and any prompt can be promoted into a tracked AI Visibility query with one click.

The flow is a single round-trip: you supply keywords, TruIntel injects your brand context, an LLM drafts the prompts, and the result is sanitized and saved. No AI platforms are queried at this stage.

Each generated prompt is a small object: the prompt text, up to 3 topic tags, and exactly one intent category. Intents render as colored pills so you can scan the mix of question types at a glance.

The 4 intent categories

Add to Queries is the real payoff. It converts a generated prompt idea into a tracked AI Visibility query, so TruIntel begins measuring whether your brand appears in AI answers to that question on every platform.

Every generation is persisted per brand so you can revisit and reload past ideas without spending another credit. The idle screen surfaces recent searches as clickable pills.

• Each Recent Search stores the keywords used, the full prompt list, and the prompt count.
• Clicking a recent-search pill reloads its keywords and prompts instantly — no new generation, no credit spent.
• History is scoped to the brand and organization and is paginated (default 20 per page, up to 100).
• Reloading a past entry lets you Add to Queries from older ideas at any time.

There is no route-level plan gate on Prompt Research — access is governed purely by your AI-credit balance. Each generation costs exactly 1 AI credit, drawn from your monthly plan_credits pool first, then any never-expiring extra_credits top-ups. Because the Free plan gets 0 credits, Prompt Research is effectively a paid-plan feature.

• plan_credits reset on the 1st of each month; extra_credits top-up packs never expire.
• Top-up packs (one-time): Starter Pack 50 credits $10, Growth Pack 150 credits $25, Pro Pack 400 credits $50.
• The page itself unlocks only after the brand's first AI Visibility check completes — until then it shows a loading gate.

Prompt Research does not use a single fixed model. It runs through TruIntel's provider-agnostic utility-LLM chain: the first provider that returns valid JSON wins, and the chain falls through on any error, a non-JSON response, or a 60-second timeout. This keeps generation resilient when one provider is rate-limited or down.

Prompt Research lives under the AI Visibility section in the navigation, and is backed by three endpoints under the /api/v1/prompt-research prefix.

Backend endpoints

Prompt Volumes is a single-keyword "AI search volume" research tool. You type one keyword or topic and it estimates how many times people are asking AI engines about it, broken down across five AI platforms, plus a 12-month trend, the kinds of prompts users actually ask, the top cited source domains, co-mentioned brands, and topic categories. Think of it as keyword research for the AI-answer era.

It lives under the AI Visibility group at the route /prompt-volumes and replaced the retired "AI Mentions" feature (the old /ai-mentions route permanently redirects here). The page is a single scrolling layout — there are no tabs.

The page always shows five platforms, but they are not all measured the same way. This is the single most important nuance to understand: only ChatGPT and Google AI Overview carry real measured data. Gemini, Claude, and Perplexity volumes are statistically modelled from the ChatGPT number using fixed market-share multipliers.

ChatGPT

REAL data. Volume is derived from Google "People Also Ask" — DataForSEO's official proxy for ChatGPT prompt volume, since no public ChatGPT-usage API exists.

Google AI Overview

REAL data. Sourced from DataForSEO Google AI Overview impressions and Google Search Volume data, using your actual selected region.

Gemini

ESTIMATED. Modelled as ChatGPT volume x 0.171 market-share multiplier, with a confidence band.

Claude

ESTIMATED. Modelled as ChatGPT volume x 0.043 market-share multiplier, with a confidence band.

Perplexity

ESTIMATED. Modelled as ChatGPT volume x 0.071 market-share multiplier, with a confidence band.

Each estimated platform row carries an "[Estimated]" badge with a tooltip explaining the methodology, and a confidence range (low / high) derived from the per-platform band multipliers above. The Total Volume in the hero is the sum of all five platform values (two real plus three estimated).

ChatGPT is US-only at the source

DataForSEO's ChatGPT dataset is hard-pinned to the US (location code 2840, language English) regardless of the region you select. For non-US regions, TruIntel applies a silent regional multiplier to the US ChatGPT number so the displayed value reflects that region's relative scale. Google AI Overview, by contrast, uses your actual selected region directly with no weighting.

A standard page lookup fans out four parallel DataForSEO calls plus one Gemini call, then assembles the trend, deltas, intent, sources, brands, and categories from the merged results.

The onboarding variant (used during brand onboarding) skips the two search/live calls, so it returns only the headline volume — relevant prompts, sources, brands, categories, and hierarchy come back empty by design.

24-hour caching

Every lookup is cached for 24 hours, keyed by a hash of the keyword, region, language, and purpose. An identical lookup within 24 hours is served free and marked as cached — it does not cost money and does not consume your lookup quota.

• A "force refresh" action bypasses the cache and re-bills (exposed as a secondary action in the UI).
• Page lookups and onboarding lookups never share a cache row (different purpose in the key).
• Self-heal: if a cached row has volume but an empty prompts list (e.g. Gemini was down at capture time), only the cheap Gemini call re-runs — not the full ~$0.42 bundle.
• Stale-cache fallback: if every upstream DataForSEO call fails but a past row exists for the exact keyword, the stale row is served rather than an error.

Lookup outcome statuses

The results page renders as a single scroll. Here is what each panel shows and how to read it.

Trend and period filter

The 12-month trend is built from monthly search history summed across all returned items so the line matches the headline magnitude. The two real series (ChatGPT and Google AI Overview) are trimmed to a common end month because Google publishes AI Overview data on a slower cadence than ChatGPT. The estimated series are the ChatGPT spine scaled by each platform's weight.

Intent badge

A rule-based classifier (no LLM) labels the keyword by matching keyword prefixes. Transactional terms are checked first, then commercial, then informational; anything unmatched defaults to informational at low confidence.

Relevant Prompts (AI-generated, not captured)

A page lookup returns five relevant prompts by default. The volume column for each is filled in asynchronously: the frontend sends each prompt's search keyword to a follow-up endpoint that runs DataForSEO cross-aggregated calls (Google AI Overview + regionally weighted ChatGPT) and returns a per-keyword AI search volume, cached for 24 hours. Onboarding lookups produce no relevant prompts.

Prompt Volumes is a paid feature — the Free plan has no access and sees an upgrade gate. There are three separate, independently metered quotas related to Prompt Volumes, plus a hard monthly spend wall that sits above all of them.

(a) Page lookups — daily vs monthly cadence by plan

The main page-lookup cap uses a monthly cadence on the lower tiers and a daily cadence on Pro and Enterprise. Monthly counters reset on the 1st of the next month; daily counters reset at midnight.

(b) Onboarding AI Search Volume — separate daily budget

The lighter onboarding lookup (used in brand onboarding Step 3) has its own per-day budget so onboarding bursts never drain your page-lookup allowance.

(c) Topics Performance "AI Volume" column — Pro+ only

The AI Overview page's Topics Performance card has an "AI Volume" column that batches up to 10 keywords per call. It is locked below Pro and metered as batches per day. A batch only counts when at least one keyword in it is uncached.

Spending credits past the cap

When a paid-plan user exhausts the page-lookup cap, they can spend AI credits to keep going at a rate of 5 credits = 1 lookup.

• Available only on the page surface, never on onboarding.
• Credits are checked up front but deducted only after a confirmed fresh DataForSEO charge — cache hits and upstream failures never burn credits.
• Deducts from your monthly plan_credits first, then from never-expiring extra_credits; you need at least 5 available.
• The Free plan can never redeem (no Prompt Volumes access at all).

Save the keywords you care about so you can re-run them later without retyping. The watchlist is a per-brand list of saved keywords, each remembering its region and language.

A "recent searches" list also surfaces your most recent lookups (newest first) so you can jump back to a previous result. The page URL (carrying the keyword, region, and period) is the source of truth, which makes a result link shareable and refresh-safe.

Page states

Prompt Volumes is reached from the AI Visibility group in the sidebar. Below is the page route and the backend endpoints that power it (all under /api/v1, all requiring an active subscription).

API endpoints

Error codes

TruIntel's SEO module pairs traditional search optimization with the AEO engine. It runs on two engines: free homepage-level checks powered entirely by free Google APIs and self-hosted parsing, and paid domain-level metrics from DataForSEO plus a self-hosted site crawler. Keyword data is Google Search Console first, with DataForSEO only enriching volume and difficulty.

The SEO Dashboard aggregates PageSpeed, keyword rankings, backlink analysis and on-page audit data into a single view headed by the Site Health gauge. It is the landing page for the SEO module and is route-accessible on every plan, though Free brands only ever see data from their single onboarding audit.

• Site Health stats bar — pages crawled, issues found, average response time, average page score and broken links
• Search Performance card — GSC clicks, impressions, CTR and average position over time on an interactive line chart
• Top Pages by Inbound Links — pages with the most internal links pointing to them
• Quick Wins section — dynamic alerts for PageSpeed under 50, SSL expiration, broken links and image issues
• Images preview — table of images missing alt text with an inline fix button
• Competitor SEO card — top competitors with DA, trust score, backlinks and referring domains
• DA & Trust vs Competitors — horizontal bar chart comparing your domain against tracked competitors

The Site Health Score is computed entirely on the client as a simple average of whichever of four components are present. It is the single source of truth shared by the Overview "SEO Health Score" card and the SEO Dashboard "Site Health" gauge. There is no separate backend aggregate, and there is no "technical health" term.

• SSL is scored valid = 100, expiring = 60, any other status = 0
• Only the components that are present are averaged, then rounded; the score is null when none are present
• This matches the backend Site-Audit "Site Health" definition used by the visibility site-audit service

Every brand gets a suite of free, homepage-level SEO checks that require no paid vendor. Nine checks run in parallel, with mobile-friendliness derived from the PageSpeed mobile score and an HTTP security-headers analysis added from already-captured response headers. Each check has its own timeout and fails gracefully without blocking the others.

1. PageSpeed Insights (Google PSI v5) — mobile and desktop, with performance / SEO / accessibility / best-practices category scores (30s timeout, retries on 429/500/502/503)
2. Core Web Vitals — LCP, INP and CLS field data from the Chrome UX Report (CrUX)
3. SSL certificate — validity, issuer and expiry via an HTTPS HEAD request
4. Structured data — JSON-LD schema detection with types and details
5. Meta-tag audit — title, description, Open Graph, Twitter cards, hreflang and canonical
6. Robots.txt & Sitemap — validity, sitemap URL count and blocking rules
7. Readability — Flesch Reading Ease and Gunning Fog index
8. Content quality — heading hierarchy, links, image alt text and paragraphs (0-100 buckets)
9. Technology detection — CMS, frameworks, analytics and CDN identification
10. Mobile-friendly (derived) — from the PSI mobile score: 90+ good, 50+ needs improvement, under 50 not friendly
11. HTTP headers (derived) — security-header analysis from the captured response headers

PageSpeed combines lab data (Lighthouse via PSI v5, mobile and desktop) with field data (Chrome UX Report). Lab data yields detailed metrics and category scores plus prioritized performance opportunities; field data reflects real users at the 75th percentile.

• Category scores — performance, SEO, accessibility and best-practices are captured for both mobile and desktop
• Performance opportunities — estimated ms/byte savings for render-blocking resources, unused CSS/JS, modern image formats, responsive images, text compression, image encoding and long cache TTL
• Diagnostics — server response time, DOM size, total byte weight, main-thread work breakdown, bootup time and third-party count
• Per-page PageSpeed scanning — scan multiple pages, track progress and rescan individual pages from the PageSpeed page

Paid checks pull domain-level metrics from DataForSEO, running sequentially over a shared session: Domain Authority, then the backlink profile, then competitor comparison. If Domain Authority or backlinks return insufficient funds, the remaining calls are skipped and the report is marked PARTIAL.

Domain Authority & Trust

• DA is derived from the DataForSEO raw rank (0-1000): DA = min(100, round(rank / 10)), scaled to 0-100 for the UI while the raw domain_rank is preserved
• Trust Score = round(dofollow referring domains / referring domains x 100) — the share of referring domains that link dofollow
• Also returns total backlinks, referring domains, referring IPs and referring subnets
• da_change is the current DA minus the previous report DA
• When the Backlinks subscription is inactive the source degrades to estimated_fallback (a whois-based estimate) and the UI warns

Backlink profile & competitor DA

• Rich backlink data: broken links, new/lost counts, referring IPs/subnets and the dofollow/nofollow split
• New and lost backlinks are tracked by comparing referring domains against the previous report
• Anchor-text distribution (non-critical)
• Competitor DA comparison fetches DA, rank, trust, backlinks and referring domains for each active competitor with a domain, capped at your competitor plan limit
• Optional top-organic keywords pull the brand domain ranking keywords from DataForSEO (not run inside the main paid checks — keywords stay GSC-first)

Keyword tracking is Google Search Console first. Connecting GSC gives you real positions, clicks, impressions, CTR and average position per query, with DataForSEO only enriching volume and difficulty. Without GSC, ranking coverage is limited to enriched and top-organic data.

• Position distribution — breakdown of keywords in Top 3, Top 10, Page 2 and below
• Position changes — which keywords improved or declined, with delta indicators
• Search volume & difficulty — estimated monthly searches and competition level
• Filter tabs — Improved, Declined, Top 10 and All keywords
• Keyword history chart — position trends for individual keywords over time
• Position alerts — notifications when important keywords move significantly

Monitor your full backlink profile to understand site authority. TruIntel tracks totals and changes over time, flags potentially toxic links and provides a complete disavow workflow that exports a Google-compatible file.

• Total backlinks and referring domains with trends over time
• New and lost backlinks tracked against the previous report
• Dofollow vs nofollow distribution
• Anchor-text distribution and referring IPs/subnets
• Toxic backlinks — spam score, toxic count, broken backlinks and nofollow breakdown (Pro+)

Disavow workflow (Pro+)

TruIntel's self-hosted breadth-first crawler (httpx + BeautifulSoup, concurrency 3, with optional JS-link recovery) scans your whole site and stores a detailed record per page. It runs on a dedicated crawl queue with a 30-minute time limit, separate from the inline paid checks.

Per-page score

Each page starts at 100 and loses points for each issue by severity: critical -20, high -15, warning -10, medium -8, low -3, info -2, floored at 0.

• Crawl summary carries the site graph (nodes + edges), crawl budget, sitemap stats, security headers and computed site warnings
• Per-brand page cap: Free 500, Starter 10,000, Lite 50,000, Pro 100,000, Enterprise 200,000 — enforced by a post-run truncation
• Image optimization is a byproduct of the crawl: every <img> is extracted and HEAD-fetched for file size, alt text, format, dimensions, indexability and loading attribute

The On-Page SEO Checker is an on-demand tool that audits any single public URL in under 45 seconds and returns roughly 90 checks across 10 weighted categories, each with recommendations and quick wins. It is distinct from the site-wide crawler audit — use the Checker for a fast deep-dive on one page.

• Each category score = (passed x 1.0 + warning x 0.5) / checks x 100; the overall is the weighted average
• Letter grade: A >= 90, B >= 80, C >= 70, D >= 50, F < 50
• Render modes: auto, browser (Playwright fallback for SPAs) or http
• Works on any public URL — the brand only owns the scan for quota and history purposes

Compare your SEO metrics against tracked competitors to see how your domain authority, backlinks and keyword footprint stack up. Competitor metrics come from the paid-checks competitor comparison run.

• Domain Authority, trust score, backlinks and referring domains per competitor
• Competitor top keywords with a dropdown selector
• Keyword overlap analysis — shared vs unique keywords
• DA & Trust vs Competitors comparison chart
• Track multiple competitors simultaneously, up to your plan limit

Keyword Suggestions returns up to 50 related keywords from the DataForSEO Labs keyword-suggestions API for a seed keyword you enter, scoped to your country. Each suggestion includes search volume and difficulty so you can prioritize, and history is saved for reuse.

• Enter a seed keyword to get up to 50 country-scoped suggestions
• Each suggestion shows search volume and keyword difficulty
• One-click add to your tracking keyword list
• Suggestion history is saved per brand

The image audit is a byproduct of the site crawl. Every image on crawled pages is extracted, HEAD-fetched for its file size, and analyzed for SEO best practices including alt text, format, dimensions and indexability.

• Detect oversized images that slow down page load (real file sizes via HEAD fetch)
• Flag missing or poor alt text and decorative images
• Recommend modern format conversions (e.g. PNG to WebP)
• Identify images without specified dimensions or proper loading attributes
• Track indexability per image alongside the page it appears on

Gap analyses surface competitive opportunities by comparing your footprint against competitors. Both are Pro and Enterprise only and power content and link-building prioritization, including outreach targets.

Keyword Gap buckets

Keyword Gap supports filters, sort and search with an opportunity-volume summary. Backlink Gap (a DataForSEO domain intersection) lists domains that link to your competitors but not to you, feeding the outreach link-building pipeline.

Crawl Visualization renders your site's internal link structure as an interactive directed graph built from the crawler's forward links. It uses React Flow to draw pages as nodes and internal links as edges, helping you spot orphan pages, overly deep pages and crawl issues.

• Interactive graph with zoom, pan and node selection
• Detail panel showing page title, path, status, depth, score, issues, response time, word count, noindex flag and link counts
• Orphan page detection — pages with no internal links pointing to them
• Crawl-depth view — how far each page sits from the homepage
• Graph metadata: total nodes/edges, max depth and root URL

SEO checks run on the Celery beat scheduler in Asia/Kolkata (IST). All times below are genuine IST. On-demand refreshes are also available, subject to per-plan quotas and a rate limit.

On-demand refresh

• Default refresh runs free checks only (homepage PageSpeed mobile + desktop, SSL, meta, robots/sitemap, structured data, readability, content quality, tech)
• A full refresh also queues paid checks (DA/backlinks/competitors) and a site crawl — paid plans only; on Free these are skipped
• Rate limit: 1 refresh per hour per brand (the first-time onboarding full check is exempt)
• Each refresh consumes 1 of the monthly PageSpeed quota

SEO capabilities scale with your plan. Dashboard, PageSpeed, Site Audit, On-Page Checker and Crawl Visualization pages are route-accessible on every plan, but Free brands only ever see data from their single onboarding audit and paid tiles show "—" or an upgrade prompt.

The SEO Analyze tools are four on-page and ranking utilities that turn your existing crawl, Google Search Console, and SERP data into specific, actionable fixes. They live under the SEO section nav in two groups: "On-Page & Indexing" (Internal Linking, Indexation Monitoring, CTR Optimization) and "Keywords & Rankings" (Rank Tracking).

Where each tool gets its data

Plan gates at a glance

Internal Linking runs an on-demand scan of any public website (defaulting to your brand domain) and returns three things: orphan pages with no inbound internal links, internal-linking opportunities (a source page that already talks about a target topic but does not yet link to it), and a PageRank-style link-authority distribution that flags important pages starved of internal links.

What a scan finds

How it works

Each scan runs a fresh same-site crawl (it does not read a cached content index), then builds the page corpus and link graph in memory. The only AI step validates candidate opportunities and writes the anchor text.

Applying a suggestion

The page

The landing hero invites a scan; once a scan completes you get a stat strip (Orphan pages, Opportunities, Pages analysed, Avg links / page) and sticky pill tabs.

• Opportunities tab: a table of "Source to Target" rows with the italic context sentence, a suggested-anchor pill, a relevance score, and Apply / Copy / Dismiss actions.
• Orphan Pages tab: the list of pages with no inbound links.
• Authority tab: the link-authority distribution plus the under-linked important pages table.
• States covered: empty ("No linking opportunities found" / "No orphan pages"), loading ("Discovering internal links..."), error, scan-incomplete, and the paid-feature upgrade gate.

Indexation Monitoring tracks whether Google has indexed each of your pages using the Google Search Console URL Inspection API. It buckets every URL as indexed, not-indexed, excluded, or error, detects deindexation transitions (was indexed, now not), emails an alert when that happens, and charts coverage over time.

The coverage buckets

Daily budget and pacing

Google caps URL Inspection at roughly 2,000 calls/day and 600/min per property. TruIntel enforces a per-brand, per-property daily counter capped at min(plan limit, 2000) and paces requests at about 6/second to stay under the per-minute ceiling.

The list of URLs to inspect is built on demand by crawling the chosen property (sitemap + links, hard-capped at 1,000 pages, cached 6h). Within the daily budget it prioritises never-checked URLs first, then the least-recently-checked, rotating the long tail across days so coverage stays fresh.

Schedule and alerts

• A daily scheduled indexation check runs at 05:30 AM IST (staggered just after the 05:00 GSC sync) for Pro+ brands with a connected GSC.
• You can also run a check on demand ("Check now") or re-check a single URL.
• A single concise deindexation email is sent after a run commits, only for confirmed indexed-to-not transitions, never for a transient error.
• A daily IndexationSnapshot row records the indexed / not-indexed / excluded / error counts that power the trend chart.

The page

• Stat cards: Indexed, Not indexed, Excluded, Errors.
• A coverage donut (Indexed / Not indexed / Excluded / Errors) plus an "Indexed over time" trend line.
• A URL table with columns URL, Status, "Google's reason" (coverage state), Last crawled, and a per-row Re-check action.
• A Recent checks history with expandable check detail.
• States: no-brand, error, the "Indexation Monitoring is a Pro feature" upgrade gate, "Connect Google Search Console", "This site isn't in your Search Console account" (reconnect), and "Run first check" / "Try again".

CTR Optimization finds pages that already earn impressions in Google but get clicked far less than their position should deliver, then writes query-aware AI title and meta-description rewrites for the biggest opportunities, with a before/after Google preview and one-click apply (CMS pages) or copy (anything else).

How an opportunity is found

TruIntel pulls GSC Search Analytics on the page dimension over a default 28-day lookback that ends 3 days back (to dodge GSC data lag), then compares each page's actual CTR against an expected-CTR-by-position curve.

Two ways to run it

Applying a suggestion mirrors Internal Linking: a CMS-managed page is updated and republished to the edge, while any other page falls back to manual copy buttons.

The page

• Landing hero "Find your CTR wins" with Property + Lookback + Min impressions inputs and an "Analyze CTR" button, plus Recent Analyses pills.
• Scan detail: a stats header (Opportunities count, Potential clicks +N, Window) and an Opportunities table (Page, impressions, CTR, position, potential clicks).
• Optimize modal: a before/after Google SERP preview with the AI title/meta and apply or copy actions.
• States: GSC-required, the Free-plan "Upgrade to Connect" gate, and empty "No CTR opportunities".

Rank Tracking monitors your keywords' Google organic SERP positions via DataForSEO. You group a domain or URL plus a set of keywords into a "project" (with a chosen country, device, and language), and each on-demand run checks every active keyword's position, diffs it against the previous run, and stores the movement.

Project configuration

The triple spend guard

A run is blocked if the estimated spend would push the brand over its monthly USD budget. The run-count and budget counters both reset on the 1st of the month.

What a run records

• Each keyword's current organic position and the change versus the previous run.
• Movement summary: Avg position, Moved up, Moved down, Unchanged.
• A distribution donut bucketed Top 3 / 4-10 / 11-100 / Not ranking.
• An average-position-over-time line and a per-keyword history sparkline.
• Per-keyword SERP competitors for the tracked query.

The page

• An optional brand-wide spend strip, then a 2-column split: selectable project cards on the left (name, domain, keyword count, last run, latest avg position) and the ProjectPanel on the right.
• ProjectPanel has a movement StatRow, the distribution donut + avg-position line, and Overview / Keywords / History tabs.
• The Keywords table shows Keyword, Position, Change, Volume, and URL, with expandable per-keyword history sparkline and SERP competitors.
• The "Run check" button shows a live cost estimate ("Run check ~ $X") before you spend.
• Run status pills: Complete / Running / Queued / Failed, plus a run history list.

A consolidated reference for every plan limit, monthly reset, and route across the four Analyze tools.

Per-plan limits

Routes

Connecting Google Search Console (GSC) links TruIntel directly to Google’s own Search Analytics and URL Inspection APIs, so every clicks / impressions / CTR / position number you see is pulled straight from Google — no third-party estimation or enrichment. The connection powers three distinct capabilities for each brand.

Authorization uses a popup-based OAuth flow with a read-only scope, and the resulting tokens are encrypted at rest. Because the Google token is account-wide, one connected Google account can read every verified property in that account.

GSC connection and the Search Performance dashboard are available on every paid plan (Starter and up). The Free plan is explicitly blocked — the connect modal and the Search Performance page both render an upgrade prompt for Free accounts. Indexation Monitoring is gated higher, to Pro and Enterprise only.

You can start the connection from two places: Settings → Integrations and the empty state of the Search Performance page. Both open the same shared connect modal.

The connection is read-only and hardened against CSRF and token theft. TruIntel never requests write access to your Search Console.

• Scope is webmasters.readonly — read-only access only, no ability to change anything in your Search Console.
• The OAuth request uses access_type=offline and prompt=consent so Google always returns a refresh token for long-lived sync.
• The OAuth state parameter is a signed JWT containing brand and user IDs plus a nonce, with a 10-minute expiry, protecting against CSRF.
• The callback returns a tiny popup-close HTML page that postMessages the result back to the opener; the frontend validates the message origin before acting on it.
• Both access and refresh tokens are Fernet-encrypted at rest (AES-128-CBC + HMAC-SHA256) using a server-side key.
• Tokens auto-refresh proactively whenever they are within 5 minutes of expiry.

The Search Performance page lives at /seo/keywords (the URL says "keywords" but the page is titled Search Performance). It is a live mirror of your GSC Search Analytics data, queried per request with a 15-minute Redis cache, not a read of the daily sync.

Controls

Data table tabs

The results table has six dimension tabs, each sortable by clicks, impressions, CTR, or position, with page sizes of 10 / 25 / 50.

• Queries — the search terms surfacing your site
• Pages — per-URL performance
• Countries — performance by geography
• Devices — desktop / mobile / tablet split
• Search Appearance — rich-result and feature breakdown
• Days — the date-dimension time series

Separate from the live dashboard, the daily keyword sync writes ranking history that other SEO surfaces read. Understanding its windows explains the position-change deltas you see elsewhere.

• First sync (never synced before) pulls the last 90 days; every subsequent daily sync re-pulls the trailing 28 days.
• Rows are fetched on the query and page dimensions, paginated up to 25,000 rows per page.
• Results store one KeywordRanking row per query, aggregated across pages: the best page (lowest position) wins, while clicks and impressions are summed. Queries longer than 500 characters are skipped.
• Position change is computed against the most recent prior GSC record that is at least 7 days old, so the overlapping 28-day windows still yield a meaningful delta.
• Clicks, impressions, CTR, and average position come directly from Google — there is no DataForSEO or other third-party volume enrichment (a previously stored search_volume is simply carried over if it already exists).
• Every sync writes an SEOReport record and is guarded by a Redis lock so global and per-brand syncs never overlap.
• All GSC calls use a 3-try retry ladder with exponential backoff that respects Retry-After on 429s and retries 5xx / timeout errors.

Connection status

The status surface exposes whether the integration is connected and active, whether a sync is currently running, the connected property URL, last sync time and status, any sync error, and the available data date range. The UI polls this every 10 seconds while a sync is in progress.

Indexation Monitoring is a Pro and Enterprise feature that rides the exact same read-only OAuth scope — no re-consent needed. It calls Google’s URL Inspection API to report the real index status of your pages and alerts you when pages fall out of the index.

• Classifies each page as indexed, not_indexed, excluded, error, or neutral, using Google’s own inspection verdict.
• The URL list is built on demand by a site crawler (sitemap + links); never-checked and least-recently-checked URLs are prioritized so the long tail rotates daily.
• Google caps URL Inspection at ~2,000 inspections/day and 600/min per property. TruIntel enforces a per-(brand, property) daily budget of min(plan limit, 2,000) and paces requests at roughly 6/sec.
• When a previously indexed page becomes not_indexed or excluded, a deindexation alert is raised.
• The daily indexation check runs at 05:30 IST, staggered just after the 05:00 keyword sync.

All GSC-related background jobs run on the Asia/Kolkata (IST) timezone. Times below are genuine IST.

Where GSC appears in the app, and the backend endpoints that power it.

AI Insights is the brain that turns a brand's week of raw visibility, SEO, traffic and lead data into a ranked set of strategic Findings and concrete Tasks. Each brand gets a weekly Insight Report produced by a 4-step Gemini pipeline, then a closed-loop layer measures which completed tasks actually moved the needle and feeds that memory back into the next report.

Every report is produced by a 4-step sequential Gemini pipeline where each step feeds the next. The pipeline first aggregates a rich week-over-week snapshot of the brand, then runs trend analysis, competitor analysis, strategy (findings) and task generation — finishing with a deterministic grounding pass that drops or hides anything the live data contradicts.

After the four steps, two final passes run before the report is marked complete:

• Grounding pass (no LLM): deterministically DROPS findings/tasks that the live data hard-contradicts (e.g. a "robots.txt blocks AI crawlers" claim when the live crawler check says all bots are allowed) and FLAGS unverifiable ones (grounded=false), which are hidden from the default view.
• Task-to-finding linking + query resolution: each task is matched to at most one source finding, and its affected queries are resolved to canonical tracked-query text so the attribution loop can join them later.

The differentiator is the breadth of input. Before any LLM call, the aggregator builds a week-over-week snapshot (current period vs the prior period) from full visibility checks plus every other connected data channel. The richer the input, the sharper and more grounded the findings.

Each report contains roughly 5-10 findings (the grounding pass may drop or hide some, so a report can show fewer). Findings are ordered by impact score, descending, so the most consequential observations sit at the top.

Every finding carries the following fields:

The Insights page (/insights) shows a stats row (Visibility Score and change, Findings count, Critical/High, Tasks Generated, Mentions) above a Key Findings table with columns for Finding, Priority, Category, Impact, Platforms, Tasks count and Date. Search and Priority/Category filters narrow the list, and a week selector lets you browse report history. By default the page scopes to the latest completed report and shows grounded findings only. Clicking a row opens /insights/:findingId for the full evidence breakdown.

Tasks are the concrete actions derived from findings. Each report generates 5-15 tasks, and every task maps to at most one source finding via a heuristic match (it may be unlinked). A task is either Owned or Earned — a distinction that defines how you act on it.

Tasks move through a simple three-state lifecycle:

Beyond status, each task exposes a rich set of fields you can use to plan and prioritise:

• Category — content / seo / pr / technical / social
• Priority — critical / high / medium / low
• Content type — listicle, how-to guide, article, comparison, category page, profile, UGC (reddit/youtube/quora/forum), outreach, technical SEO
• Targeting — target source and target platform (chatgpt / claude / gemini / perplexity / google_ai_overview / all)
• Planning — estimated effort hours, suggested due date, success metrics, expected impact
• Impact tracking — actual_impact and impact_measured_at, auto-stamped by attribution (prefixed "Auto:") or set manually on done tasks

The Tasks page (/tasks) offers two ways to work, plus summary analytics and quick filters. By default it scopes to the latest completed report, ordered by priority then date.

Three independent filter axes plus search let you slice the board:

A stat row across the top (Total, To Do, Urgent = critical+high, Done, Dismissed) doubles as a click-to-filter control. Summary cards show a status donut, priority bars and a progress card with completion rate, earned/owned counts and per-category breakdowns. Completing a task opens a modal for notes and optional impact; dismissing requires a reason. The completion rate is computed as done divided by (total minus dismissed).

Outcome Attribution is the closed loop that proves which completed tasks actually moved visibility. It runs period-over-period — primarily monthly, folded into the monthly brand-report task — rather than as a strict next-week check.

For each task completed in the period, attribution joins the task's affected queries to how those queries actually moved. A query only counts as a creditable win when one of these is true:

• It flipped from not-mentioned to mentioned ("query won")
• Its rank position improved
• It gained a citation

Each period writes a BrandPerformanceSnapshot (upserted per brand and period) capturing tasks completed/total, overall score and change, queries won, net position change, citations gained, attributed outcomes and a what_worked map (category to measured-impact weight). That what_worked memory then biases both the next report's strategy and task prompts and the Opportunity Agent's scoring — closing the loop. It surfaces at Content Studio's Accountability page (/content-studio/accountability) and via the performance-history and accountability endpoints.

The Opportunity Agent turns gaps into ranked, evidence-backed content opportunities — the rows that populate the content calendar (there is no separate calendar table). It fuses three signals into a single opportunity_value score from 0-100.

• AEO query gaps — tracked queries with one or fewer platform mentions
• Keyword/backlink gaps — DataForSEO Labs gaps vs top-3 competitors (budget-guarded)
• What-worked bias — a +/-0.20 nudge from the brand's proven-impact categories

Opportunities refresh weekly for paid brands (Sundays 09:00 IST) and surface in Content Studio at /content-studio/opportunities and the publishing /content-studio/calendar. Although these pages sit under the Insights nav section, they are Content Studio surfaces of the same closed loop.

Reports are generated automatically each week and can also be triggered manually. Manual generation runs asynchronously: the report row is pre-created in a "generating" state so the UI can poll for completion every few seconds.

The Insights and Tasks pages handle every report state explicitly:

AI Insights is effectively a Starter-and-above feature. The Insights and Tasks pages are gated to Starter+, and generating a report requires a paid plan. The AI models themselves are global, not plan-tiered.

The product surfaces for AI Insights and Tasks, plus the Content Studio surfaces of the closed loop:

Key backend endpoints under /api/v2/insights:

• GET /{brand_id}/reports, /reports/latest, /reports/{report_id} — list and fetch reports
• POST /{brand_id}/generate (async by default), POST /{brand_id}/generate-tasks — generate a report or regenerate tasks
• GET /{brand_id}/findings, /findings/{finding_id} — read findings (latest report + grounded-only by default)
• GET /{brand_id}/tasks, /tasks/summary, /tasks/{task_id} — read tasks and summary counts
• PATCH /{brand_id}/tasks/{task_id}/status, /tasks/{task_id}/impact — change status or set impact
• GET /{brand_id}/performance-history, /accountability?period=YYYY-MM — the closed-loop data

Brand Reports are pre-computed, point-in-time snapshots that consolidate your AI Visibility (AEO), SEO, traffic, and lead data into a single executive-ready document. Each report bundles headline metrics, an LLM-written executive summary, and linked findings and tasks, so a stakeholder can understand a brand's standing without opening five different dashboards.

Reports are computed once and stored as a record — they do not re-query live data when you open them. This makes them fast to load and gives you a stable historical archive you can compare period over period. You can browse them at /reports and open any single report at /reports/:reportId.

TruIntel produces three distinct report types. All three share the same underlying data model and detail layout — they differ in the window they cover, how they are triggered, and how their executive summary is framed.

Brand Reports are separate from AI Insight Reports. When an insight report overlaps a brand report's window, the brand report optionally links to it so its findings appear inside the report detail page.

The report detail page renders its data as a sequence of section cards (around 13 — the three brand-narrative sections, How AI Describes Your Brand / New This Period / Earned Media Opportunities, share a single three-card row), opening with a hero that shows the overall visibility score, recognition rate, total mentions, the five AI platforms, and a button to jump to the linked insight report. Underneath, the sections walk through every dimension of the brand's performance.

1. Executive Summary — the AI-written narrative for the period
2. AI Visibility Overview — a stat row of the headline AEO metrics
3. Platform Breakdown — per-platform performance across ChatGPT, Claude, Gemini, Perplexity, and Google AI Overview
4. Query Performance — how individual tracked queries surfaced the brand
5. Competitor Landscape — a comparison table against tracked competitors
6. Brand Perception — sentiment distribution for the period
7. How AI Describes Your Brand — extracted brand attributes
8. New This Period — what changed since the last snapshot
9. Earned Media Opportunities — citation and outreach openings
10. Citation Sources — the domains AI engines cited
11. SEO Health — domain authority, PageSpeed, backlinks, keywords, and crawl issues
12. Top Keywords — the brand's strongest keyword positions
13. Traffic Intelligence — a visitor-classification donut plus agent breakdown
14. Lead Verification — verified, review, and rejected lead counts
15. Findings & Actions — the task summary plus linked findings

Under the hood, a report stores four JSONB snapshots — AEO, SEO, traffic, and lead — built by reusing the same aggregation used elsewhere, so generating a report does not re-query the source tables. The SEO, traffic, and lead snapshots are null when a brand has no data for them, which is common on lower tiers where that underlying data is feature-gated. The AEO snapshot is the richest, carrying current metrics, changes, platforms, competitors, sources, attributes, queries by intent, query highlights, sentiment distribution, totals, a task summary, and (monthly only) accountability data.

Every report opens with a short executive summary written by gemini-3.1-flash-lite at a low temperature (0.3) for consistency. The model receives the aggregated metrics for the period and returns a tight, data-grounded narrative — never marketing fluff.

Reports are generated automatically on the platform's Celery beat schedule. All times are genuine India Standard Time (Asia/Kolkata) — the scheduler runs in IST, so no timezone conversion is needed.

Both scheduled generators only process paid brands that have a visibility check within the past 14 days. The monthly job does extra work: it also regenerates content opportunities and fans out a monthly accountability email to every org member.

Beyond the scheduled cadence, two automatic mechanisms keep a fresh report available without you asking.

Instant "State of Your Brand"

Every time a full visibility check completes, TruIntel auto-dispatches a weekly-type report scoped to a single day (today), about 20 seconds later. It is idempotent per day, so repeated checks do not pile up duplicate rows. This instant report is hidden from the historical list — it powers the "current state" view rather than the archive. The frontend polls for it every 4 seconds while it builds.

Onboarding report coordinator

When onboarding finishes, TruIntel creates a placeholder onboarding report and dispatches a coordinator task. Because AEO and SEO checks finish at different times, the coordinator polls readiness and re-dispatches itself every 45 seconds until both are ready, then generates the report.

Separately from the in-app brand reports, TruIntel emails a weekly visibility digest every Monday at 08:00 IST to each org member of an active paid brand. It is built directly from your two most recent VisibilityChecks — not from a stored brand report — so it can go out before the brand report is generated.

• A circular visibility-score gauge with the week-over-week delta
• A Week-over-Week Movement table
• Top Insights for the week (up to 3)
• A list of open tasks (top 5)
• A "Full Report" link back to /overview

Any completed report can be exported as a clean, multi-page A4 PDF for sharing with clients or leadership. The exporter uses your browser's native print pipeline rather than a screenshot library, so text stays crisp and selectable.

The Reports area lives under the REPORT group in the sidebar. The list page at /reports gives you a filterable archive, and each report opens at /reports/:reportId.

List page (/reports)

• Tabs to filter by All / Onboarding / Weekly / Monthly
• A table paginated at 20 rows per page
• Type badges: weekly (info), monthly (default), onboarding (success)
• An empty state that explains the automatic onboarding report and the weekly/monthly cadence
• Instant single-day weekly rows are hidden; onboarding rows do appear

Detail page (/reports/:reportId)

• A hero with the score, recognition rate, mentions, "5 AI platforms", and a linked-insight button
• The ~13 section cards described above
• A generating guard with a spinner (e.g. "Building your onboarding report") while a report is still computing
• A failed state with a "Retry generation" action
• A developer-only Raw JSON view

Report endpoints live under the /api/v1/reports prefix. The router requires an active subscription for all reads; manual generation additionally requires a paid plan (Starter or above).

Reports are entirely a paid capability. The table below summarizes what each plan can do with reports.

There is no further per-tier gating on reports themselves — a Starter report and a Pro report are structurally identical. However, lower tiers may see null SEO, traffic, or lead sections because that underlying data is feature-gated elsewhere in the product.

Content Studio is TruIntel's in-app editorial production hub. It turns your measurement signals — AI-answer gaps, keyword and backlink gaps, and SERP gaps — into a ranked plan of what to write, a research-grounded staged draft flow, a publishing calendar that auto-publishes real articles to your website, a unified drafts library, and a monthly accountability view that proves the work moved the needle.

It lives under the Insights section of the left navigation, and every page sits under the /content-studio/* route family. Content Studio is the v2.1 successor to the older standalone "Plus" express generator: the frontend feature key and plan gate are still labelled "plus" / "TruIntel Plus" internally, but the user-facing product is Content Studio.

The Opportunities page is the front door of Content Studio. It presents a ranked table of content ideas, each one carrying the evidence behind it — search volume, difficulty, the target AI query, ranking competitors, and the specific gap that was found. Rows are sorted by opportunity value (0-100) descending so the highest-impact ideas surface first.

Three signals fused into one ranked plan

Each opportunity is fused from up to three independent gap signals. A candidate that appears in more than one signal is merged into a single combined opportunity rather than duplicated.

When only one signal is present, that sub-score is used directly. The final value is then biased up to plus or minus 20% by the latest performance snapshot's "what worked" weighting — keyword-gap ideas lean to the "seo" category, everything else to "content" — so the plan adapts to what has actually been moving your numbers.

Status lifecycle

Page actions and states

• Per-row Create — opens the staged Create wizard seeded with that opportunity (?opportunity_id=).
• Per-row Schedule — a date picker that sets scheduled_date and flips status to scheduled. This action is hidden until a CMS domain is connected.
• Header Refresh — re-derives opportunities on demand.
• Filters — server-side status filter (suggested / scheduled / in_progress / published / dismissed) plus a client-side title search.
• Summary StatRow — total count, top value, high-impact count (value 75 or higher), and scheduled count.
• States — no-brand, loading skeleton, error with Retry, empty (a setup checklist prompting you to run your first scan when none exists, otherwise a Refresh CTA), and the ranked table.

You rarely have to ask for opportunities — TruIntel generates them for you. There are four paths that populate the list, ranging from a one-time back-fill when you onboard to a weekly scheduled refresh.

What powers the keyword signal

The keyword/backlink gap signal diffs your domain against your top three most-mentioned active competitors that have a domain, using DataForSEO Labs. It requires a connected domain, at least one qualifying competitor, and an available per-brand SEO budget.

Staged Create is a resumable five-step wizard that grounds every editorial decision in real research before you spend a single credit. You always enter it from an opportunity (or with an explicit topic and seed keyword); it is intentionally not a standalone nav item. Every checkpoint surfaces the evidence behind its recommendation.

What goes into a brief (free)

Building a brief is free — the credit charge only happens at the Draft step. Before any paid call, a budget gate runs; if the per-brand SEO budget is exhausted the request returns 429 PLAN_LIMIT_REACHED. The research that goes into a brief:

• Keyword research — DataForSEO Labs keyword suggestions (up to 50) plus search volume for the seed keyword.
• GSC demand — your top 25 Google Search Console query rows over a 90-day lookback, where impressions represent real demand (degrades to DataForSEO-only when GSC is not connected).
• Secondary cluster — Labs suggestions ranked by volume with GSC terms folded in, deduped by token-set, capped at 8 keywords, each evidence-tagged with its source (dataforseo or gsc).
• SERP analysis — one paid advanced live-SERP call (depth 10) returning organic URLs and PAA; the top pages are fetched and analyzed to compute an outlier-filtered median word count (with a 120-word floor).
• Merged outline — an LLM-merged table of contents (up to 14 sections) built from the analyzed pages, with a deterministic frequency-merge fallback if the LLM is unavailable.
• H1 synthesis — a small utility-LLM call proposes the H1 (opportunity titles are used as-is).

Editing a brief (free)

You can patch any field on a brief without charge — topic, H1, primary and secondary keywords, target location, audience, tone, word count, table of contents, PAA, and status. Briefs move through their own status lifecycle: draft, keywords_confirmed, toc_confirmed, drafted, archived.

The Draft step is the only part of the staged flow that costs credits. TruIntel deducts the credits up front (under a row lock to prevent double-charging), creates a placeholder generation, marks the brief drafted, and dispatches a background task that runs the full content-generation pipeline. If the dispatch fails, the credits are automatically refunded and the brief is reset.

Repurpose one brief into many surfaces

The primary blog draft comes from the Draft step. Repurpose takes the same brief and produces the other fast, grounded answer surfaces from it — each becomes its own generation that you poll until ready. The brief's status and its opportunity link are left untouched.

Credit pools and monthly allowances

Credits come from two live pools: plan_credits, your monthly allowance that resets on the 1st, and extra_credits, top-up packs that never expire. The discontinued legacy "Plus" credit pool is no longer spendable.

The Calendar page is a publishing calendar over your real CMS articles — not over opportunities. The month grid shows scheduled, publishing, published, and failed articles on the day they go live, and a side Planner drawer lets you queue new work.

The Planner drawer

How scheduling and auto-publishing work

Clicking any chip opens that article in the CMS editor. The calendar reads from GET /cms/content/calendar?month= and uses POST /cms/content/{id}/schedule, POST /cms/content/{id}/unschedule, and GET /cms/content/{id}/publish-status.

The Library is a single unified "my drafts" view that merges both content stores behind one table, each row carrying a source badge so you always know where a draft came from.

• Source and status filters plus a title search.
• A freshness stamp showing when the list was last loaded.
• States — no-brand, loading skeleton, error with Retry, empty (a generate CTA), and the merged table.

Accountability is the "did the work move the needle?" surface. It answers, month by month, whether the content and SEO work TruIntel recommended actually changed your numbers — and what to do next. It has two halves.

Looking back

Defaulting to the latest monthly snapshot (or any period you pick), the looking-back view proves impact with concrete, attributed evidence rather than vanity charts.

Looking forward

The second half lists that period's top-ranked content opportunities, each with evidence pills that link straight into the staged Create flow — closing the loop from proof back into the next round of work.

Data comes from GET /api/v2/insights/{brand_id}/accountability?period=YYYY-MM and GET /api/v2/insights/{brand_id}/performance-history.

All of Content Studio is a paid-plan feature. The capability-level gates and the full route reference are below.

Frontend routes

Backend endpoints

TruIntel Plus (branded "TruPlus" in the app) is a credit-metered AI content generator. It turns your brand's AEO and SEO visibility data into ready-to-use marketing assets — SEO blog posts, FAQ pages, structured answers, schema markup, and metadata/social tags — that are pre-optimized to win mentions and citations across AI engines.

Every generation deducts credits from your organization's credit ledger. Output can be copied as Markdown or HTML, downloaded as a .md file, marked Published or Archived, or sent straight to the CMS as an editable draft.

The create surface offers 5 grounded generators across 2 tiers. (The old "Backlink Strategy" tier — directory listings, outreach templates, citation plans — was retired from Plus and moved to the dedicated Outreach workspace at /outreach.)

Content tier

Technical SEO tier

You can generate from a dedicated single-type page (a single form) or from the generic 4-step wizard at /plus/generate. Both call the same backend and both cost 1 credit.

Which model runs it

Plus labels every generation gemini-3.1-flash-lite in its metadata, but execution actually routes through the provider-agnostic llm_json fallback chain, which is Anthropic-first (Claude Haiku 4.5) and only falls back to Gemini, OpenAI, or OpenRouter on failure. The displayed model name is nominal, not necessarily the provider that produced the text.

Once generated, content lives as a PlusGeneration record you can export, push to the CMS, or move through a simple publish lifecycle.

Export actions

• Copy Markdown — copy the raw Markdown to your clipboard.
• Copy HTML — copy the rendered HTML to your clipboard.
• Download .md — save the generation as a Markdown file.

Send to CMS

Eligible types (Blog Post, FAQ Page, Structured Answer) can be pushed into the CMS with one click. The bridge converts Markdown to HTML, parses frontmatter, runs a free Gemini SEO-optimization pass, and creates a CMSContent draft linked back to the generation. It is free, charges no credit, and is idempotent (re-sending updates the same draft). Structured Answers map to the CMS blog_post type.

Lifecycle & statuses

Delete permanently removes a generation row. Publish and Archive are reversible status changes via PATCH /plus/generations/{brand}/{id}/status.

The /plus/history page is a searchable, filterable record of everything you have generated for a brand.

Plus content consumes AI credits. Credits are drawn from two live pools in a fixed order, and unused express generations are refunded on failure.

Monthly plan allowance

What each action costs

When your plan credits run low, the TruPlus hub surfaces a low-credits banner and a "Buy Extra Credits" modal. The live UI sells flat top-up packs purchased via Razorpay — one-time, and the credits never expire.

Your live credit balance and usage are shown on the TruPlus hub (low-credits banner + balance widget) and the Buy-Credits modal, and are also surfaced in Settings → Billing. For India, 18% GST is added on top of the listed USD price.

These Plus-service capabilities share the same credit pool and PlusGeneration table but are surfaced through the Content Studio brief flow rather than the core Plus pages.

TruPlus pages and the dedicated single-type generators in the product app:

Key backend endpoints (all under /api/v1/plus, gated by require_paid_plan):

The TruIntel CMS is an in-app content module mounted under /cms/* in the product app (app.truintel.ai) — not a separate cms.truintel.ai portal. It lets a brand scan its own website, generate AEO-optimized articles with AI, edit them in a rich-text editor with a live AI SEO score, and publish them to Cloudflare’s global edge so the page looks like it lives on the brand’s own site.

The module is 16 pages / 16 routes, every one wrapped in a CMS protected route plus the Starter+ plan gate. It reuses your org-level AI credit pools rather than a separate CMS wallet.

Each new article is produced by a 5-step AI pipeline. Before step 1 runs, a pre-step gathers grounding data: your brand’s latest full AEO visibility check (overall/recognition/citation scores, per-platform scores, the queries where your brand is NOT mentioned, and your top 5 competitors), real product/pricing/feature snippets from the content index, relevant existing pages for internal linking, and your scanned brand tone. The pipeline grounds content in your actual AEO gaps and site facts — it does not crawl the live web.

Every step is a separate LLM call and has a deterministic fallback, so a transient model outage degrades quality rather than crashing the job. Output is stored as HTML plus a TipTap-compatible JSON wrapper. Generation progress is pushed live so the New page can show each step completing in real time.

There are two ways to kick off generation: a quick Express path from the CMS New page, and a staged, brief-driven path from Content Studio. They differ in how much research happens up front and in credit cost.

The premium brief tier biases the chain Anthropic-first and adds a premium editorial note. Despite an internal “Sonnet-first” label, premium still resolves to claude-haiku-4-5 — there is no Sonnet model wired in.

CMS spends from your org-level AI credit pools in the order plan_credits (monthly allowance, resets on the 1st) then extra_credits (top-ups that never expire). Monthly allowance by plan: Free 0, Starter 50, Lite 150, Pro 500, Enterprise 1000. One-time top-up packs never expire: 50 credits for $10, 150 credits for $25, 400 credits for $50.

Generated content opens in a full TipTap rich-text editor with rich text, images (with alt text), code blocks, links, an SEO sidebar, AI-assist, and editable schema.

AI Assist runs on selected text with three actions — rewrite, expand, or shorten (up to 5,000 characters of input). It is free (0 credits) and runs through the multi-provider LLM chain, so it survives a single provider outage.

AI SEO/AEO Score

The editor scores your draft against a 5-dimension rubric summing to 100. Scoring is LLM-judged (gemini-3.1-flash-lite via the chain), free, and cached on the content’s publish metadata keyed by a SHA-256 fingerprint of title, meta title, meta description, slug, schema, and content HTML — so re-running skips the LLM unless the content meaningfully changed.

Ratings: excellent at 85% or above, good 65–84%, needs work 40–64%, and poor below 40% of each dimension’s max. Note that the page title is rendered as the page H1 by the publish template, so the scorer counts the title as the H1 and will not penalize the body for a “missing H1.”

TruIntel publishes content to Cloudflare’s global edge (KV + Workers) using a multi-tier renderer that makes the published page blend into your existing site. The system picks the highest tier it can, falling back gracefully.

All HTML passes through a single sanitization choke point before serving. After render, the page is written to Cloudflare KV, sitemap.xml is regenerated, the published URL is built (preferring your real domain), search engines are pinged (Bing IndexNow plus a best-effort Google sitemap ping), and the edge cache is purged for the slug and sitemap. If the publish satisfied a linked task, that task is auto-marked done.

Content lifecycle statuses

• draft — created or generated, not yet published
• generating — the AI pipeline is running
• failed — generation failed (credits refunded)
• scheduled — queued for a future publish time
• publishing — actively being rendered and pushed to the edge
• published — live on the edge
• publish_failed — the publish step failed
• archived — retired from active management

You can unpublish a live piece at any time, and a publish-preview endpoint lets you see the rendered page before it goes out.

Instead of publishing immediately, you can schedule a piece for a future date and time. Scheduling is surfaced through the Content Studio publishing calendar, where you drag content onto a date to set its publish time.

• Content that is published, publishing, generating, or archived cannot be (re)scheduled.
• A piece must have HTML before it can be scheduled.
• The scheduled time must be in the future.

To publish on your own website you connect a custom domain. TruIntel supports two connection modes depending on how much of your DNS you want to delegate.

Helper endpoints validate and diagnose the connection: detect-origin, validate-origin, check-canonical, dns-status, kv-status, and disconnect-info. You can also re-extract the brand template at any time if your site’s design changes.

The site scanner BFS-crawls up to 200 pages (concurrency 3, max depth 10, with a 20-minute crawl cap) to build a content index and capture everything needed for on-brand publishing. Per-page AI analysis, content-gap analysis, and recommendations use gemini-3.1-flash-lite directly — so for the scan, Gemini Flash is correct.

A cleanup task marks any scan stuck for more than 30 minutes as failed so the scan UI never hangs indefinitely.

The Optimization page turns the crawler’s detected issues into actionable fixes. Each issue carries severity, a credit cost, and whether it is auto-fixable. Auto-fixable issues are applied as edge modifications to KV; non-auto-fixable issues come with manual guidance only.

TruIntel keeps published content current by refreshing it on a monthly schedule. Refresh rewrites only the outdated sections, updates meta, and adds new internal links — it is not a full rewrite. The refresh model is gemini-3.1-flash-lite.

• Runs monthly on the 1st at 03:00 IST.
• A piece is eligible when it is over 3 months old, has new internal-link opportunities (pages crawled after its last refresh), or was published over 30 days ago and never refreshed.
• Refresh updates only outdated sections plus meta and internal links.

Beyond generation, editing, and publishing, the CMS includes a set of live-site management pages for a connected domain. All 16 CMS pages share the single Starter+ gate.

Backend support comes from roughly 60 endpoints under /api/v1/cms (scan, content index, credits, content CRUD + calendar, SEO analysis, domains, publish + schedule, optimization, AI-assist, brand template, page editor, page SEO, robots.txt, and sitemap) plus the Content Studio brief flow under /api/v1/content (brief create/get/patch, paid draft, repurpose, opportunities, and calendar).

Off-Page Outreach is a human-in-the-loop backlink and earned-media workspace. It finds high-value link and AI-citation prospects, drafts one finished, ready-to-send email per prospect signed with your real identity, and tracks every prospect through an honest 5-stage CRM funnel. It does not auto-send email — sending is a one-click handoff to your own email client.

Off-Page Outreach is available on the Pro and Enterprise plans. Drafting and regenerating emails is completely free — it never spends AI credits. The only metered cost is the paid backlink-gap lookup during a refresh, which draws on your brand's SEO budget.

Outreach lives in the left sidebar under the Insights section, in the "Off-Page" group as "Off-Page Outreach". It is a two-page experience: a ranked prospect table and a single-prospect composer.

A refresh fuses three off-page signals into one ranked list. Each prospect carries an evidence object — there is no prospect without a concrete grounding signal, which keeps the list (and the emails written from it) free of generic filler.