The capability surface
Eight parts of one automatic system, all reading from the same weekly record — so a missed recommendation becomes a shipped fix with nothing falling through the cracks.
Short answer
Eight automatic advantages: weekly monitoring across six AI assistants, alerts when something changes, a 100-point check of your own site, five ranked fixes drafted for you, new buyer questions you were missing, an exposure estimate, board-ready reports with a private link, and API access so your own tools stay in sync.
Everything reads from one weekly record. The site check reveals what is costing you recommendations, the tracker shows what buyers were told, the alerts catch what changed, and the impact view proves whether your fix won the recommendation back. One version of the truth, working while you don't have to.
How it is built
Measuring, spotting problems, writing fixes and proving results are separate stages on purpose, so improving one never throws doubt on another.
Part 01 · Tracking
Each of the sixty weekly checks records how warmly you were described, where you placed, which websites the AI used, and the reason behind the outcome. Everything else in the product reads from that record.
Five stances are distinguished — first choice, recommended, alternative, mentioned only and cautioned against — and each carries a multiplier applied on top of position. A tiered list and a flat list are different outcomes, and a mention count cannot tell them apart.
Every answer also yields the reason behind the outcome, so a finding is never a dead end.
Sources
Every citation is normalised to a canonical domain, classified by type — review hub, community thread, editorial, documentation, vendor — and tracked for brand and competitor presence. Source gap analysis is derived from this, not guessed.
Grounding
Reconnaissance reads your pricing page, feature set and competitor positioning before a single prompt is written. The result is constraint-driven buyer language — team size, budget pressure, integration requirements, migration pain — instead of keyword strings that no human would type.
Coverage
Part 02 · Alerts
Detection runs on every processed answer and ranks urgency, so one severe incident never competes with five minor ones. Each signal produces an incident with the evidence attached and a correction kit generated from it. How the radar and kits work.
| Severity | Signal | What triggers it |
|---|---|---|
| Critical | Cautioned-against stance | An engine now warns buyers away from you. |
| Critical | Negative sentiment | A claim about your product reads negative — the hallucination signal. |
| Warning | Stance drop | First choice or recommended this cycle becomes alternative or mentioned-only. |
| Warning | Decision-stage conquest | A competitor takes first choice on a buying-stage query where you are absent. |
| Warning | 5.0-point weekly drop | Weighted visibility falls five points or more between two complete weeks. |
Factual Correction Kit
Delivery
In-app radar
A severity-sorted incident list with the evidence, the correction kit and a resolution control. Always on, no quota, no delivery dependency.
Email digest
One batched digest for all un-notified incidents, severity-ordered, with a twenty-four-hour per-brand cooldown and an idempotency key so a retry never double-sends.
Optional webhook
A Slack webhook fires after a successful digest, and independently when a done-for-you implementation request is submitted.
Detection is tied to the weekly cycle. This is drift detection, not real-time monitoring.
Part 03 · Your site
Before arguing about recommendations, find out whether the AI can reach your website at all. If a crawler is blocked, every hour you spend on content is invisible.
Parsed per user-agent group with longest-prefix matching, falling back to the wildcard rule. Reports allowed, blocked or unspecified for GPTBot, ClaudeBot, PerplexityBot and Google-Extended. A blocked retrieval crawler is a critical finding with a ready-to-paste allow block.
Presence and substance of the proposed convention at your domain root, worth 10 and 5 points. Missing llms.txt is a warning; a missing full corpus is informational, since that extension is optional.
Walks every JSON-LD block on your homepage and pricing page, including @graph nesting. Awards points for an Organization entity, a sameAs cluster of at least two authoritative profiles, a Product or SoftwareApplication, transparent pricing in offers, and a FAQPage.
Exactly one H1, at least three H2 sections, a table or definition list, a visible-text floor with a text-density ratio, and a recognisable landmark region. This is what separates a page an engine can parse from one it abandons.
Illustrative composite score
Score bands
What you get back
The ring above illustrates how the four areas compose. Scores, stances and citations are computed from each brand's own weekly cycle — see the methodology.
Part 04 · The fixes
Each fix only appears when your own data calls for it. If nothing needs fixing, the list stays empty rather than inventing work to look busy — which costs us upsell and earns your trust.
The no-fabrication rule
Generated assets use only facts observed in your crawl and citation data. Anything unverified is emitted as an explicit placeholder marker for you to complete, and a verification script fails the build if a zero-price offer template or a missing marker ever reappears.
A ≤45-word direct answer to the exact objection the engine raised, plus FAQPage JSON-LD.
Trigger · A forensic gap reason was extracted from an engine answer.
A complete /llms.txt file: proposition, capability matrix, pricing, key pages, contact.
Trigger · The AI Health Check finds /llms.txt missing.
Organization JSON-LD with sameAs slots and step-by-step placement instructions.
Trigger · Your brand is only mentioned in passing, or cautioned against.
An honest comparison page with a capability matrix, when-to-choose-each, and migration friction.
Trigger · A tracked competitor took first choice on a commercial query.
A non-promotional, affiliation-disclosing contribution draft for the source that decided the answer.
Trigger · A Reddit thread or review site was the deciding citation.
Part 05 · More questions
We look at where competitors won, where you were absent, and which sites never mention you — then write the questions a real buyer would ask next, in five recognisable buyer moods.
Suggestions are de-duplicated against everything you already track, gated so trend-derived queries refresh on a thirty-day window, and added to the current cycle in one click so they begin reporting in the same week.
Part 06 · What it costs you
This is an estimate, not a measurement, and we label it that way everywhere. It turns “we are not being recommended” into a number you can put in front of a finance team.
The model
missed shortlists = decision-stage queries × 4–10 monthly buyer evaluations × omission rate
A model, not a measurement. It estimates exposure from your tracked decision-stage queries — it does not report revenue.
Part 07 · The report
A private link that opens without an account, an export that survives a board pack, and a measured before-and-after attached to every fix you completed.
Executive assessment
Share of voice, stance strength, competitor citation delta, top forensic vulnerabilities, crawler health, a ninety-day roadmap.
Confidential share link
An unguessable capability URL, no login required, excluded from search indexing.
Data export
CSV and JSON with brand identity and metrics only — never user identity or access tokens.
Print-ready PDF
A dedicated print pipeline with exact colour preservation and clean page breaks.
Share links do not expire and cannot currently be revoked from the interface. Deactivating the brand disables the link. Stated here because a capability URL with no expiry is a decision you should make deliberately.
Part 08 · For your own tools
Six read tools over a standard, brand-scoped API. Create a token, paste it into Cursor or Claude Desktop, and ask your own assistant what changed this week. Full MCP setup guide.
get_citation_reportCurrent and previous snapshot, per-engine breakdown, delta.list_promptsTracked buyer queries with stance, sentiment and position per engine.list_sourcesSource domains, brand presence, competitor presence, opportunities.list_actionsPending remediation queue with draft copy and priority.mark_action_doneLocks the current snapshot as the impact baseline.dismiss_actionRemoves an action from the queue.Bearer token · SHA-256 hashed at rest · 60 requests per minute per token · brand-isolated
The rate limit is enforced in memory per server instance, so it is a per-instance ceiling rather than a global quota. A shared limiter is on the roadmap; we would rather state the constraint than imply a guarantee we do not hold.
Product walkthrough
A recorded walkthrough of the remediation loop: reading a forensic gap, accepting a tier-one generation, copying the answer and its JSON-LD, and watching the impact surface pick it up on the next cycle.
Recording in production. The player below activates the moment the Loom URL is set — no third-party script or iframe is loaded before that.
Feature questions
The same answers are published as FAQPage structured data so an answer engine reads exactly what you read.
Four areas scored out of 100: crawler access for GPTBot, ClaudeBot, PerplexityBot and Google-Extended worth 25 points; llms.txt and llms-full.txt worth 15; Schema.org coverage worth 35 including Organization, sameAs, Product, transparent offers and FAQPage; and semantic structure worth 25 for headings, tables, text density and landmarks.
Five drift signals are detected: a cautioned-against stance, negative sentiment, a stance drop from first choice or recommended to alternative or mentioned-only, a competitor taking first choice on a decision-stage query, and a weighted visibility fall of five points or more between two complete weeks.
Five deliverables generated from the detected incident: the cleaned objection, a forty-five-word refutation that opens by naming the claim as inaccurate, a FAQPage JSON-LD block, a verified-facts section you can append to llms.txt, and a five-step remediation plan ending in re-running the weekly cycle.
No. Asset generation is bound by a no-fabrication rule: facts come only from observed crawl and citation data, and anything unverified is emitted as an explicit placeholder marker for you to complete. There is a verification script that fails the build if a zero-price offer template or a missing placeholder marker appears.
No — it is a model, and it is labelled as one. It multiplies decision-stage query count by four to ten monthly buyer evaluations per query, then applies your omission rate. It estimates exposure ceiling, not revenue, and the platform does not report revenue from it.
Run the audit first
The AI Health Check scores crawler access, llms.txt, Schema.org and semantic structure out of one hundred — with prioritised fixes for each finding. Free for fourteen days, before your competitors pull further ahead.