
Visibility You can start this yourself today.
Tracking AI answers, scored by rules you can read.
Your buyers’ questions go to ChatGPT, Perplexity, Claude, Gemini, Grok and Google AI Overviews, and a rules pass scores every answer the same way each time.
At a Glance
Not a rank tracker with an AI badge.
Jinn asks the AI engines the questions your buyers actually ask, scores the answers with a rules pass you can read, and builds the content that closes the gaps.
- What it measures
How prominently the AI engines name your brand when asked real buyer questions, blended into one Brand IQ from zero to a hundred.
- How it scores
A fixed rules pass reads where your name lands in each answer and scores it from zero to five. No AI grades another AI, so the same answers always produce the same number.
- What it fixes
A page-by-page Playbook of rewrite blocks grounded in your verified brand record. Every block passes a hard compliance gate and a voice check before you see it.
- What it refuses
Shortcuts. Confirmed corrections do not move the score, an accusation needs a quality source, and no accuracy number publishes until the method clears its own bar.
Every mechanism on this page is checked against the code that runs it. Where a part is still in build, the copy says so.
The loop turns the same way every time: ask the engines, check what they said, draft the fixes, and wait for your yes.
One answer, one date, one fix

The question
Wire asks the engine.
Wire puts a buyer’s question to ChatGPT and saves the answer word for word, dated.

The date check
Wire checks the date.
The answer put an airline’s worst week a year late. Wire checks it against the public record and flags the line.

The fix
Murmur writes the fix.
Murmur writes the correction in the brand’s own voice, and it waits for a yes.
The loop
Measure, fix, measure again.
Four stages, and every one of them can be inspected. The loop only works because the measuring half is fixed: when the number moves, the answers changed, not the ruler.
- Ask
The questions your buyers really ask go to the engines they use: ChatGPT, Perplexity, Claude, Gemini, Grok and Google AI Overviews. The scored question set is the same on every plan.
- Score
A rules-based reader scores how prominently each answer names you, from zero (not mentioned at all) to five (named first and called a top pick). The engine scores blend into one Brand IQ, and a run with a missing engine flags itself as degraded instead of pretending to full coverage.
- Fix
Jinn picks the pages behind your gaps and drafts rewrite blocks grounded in your verified brand record and the facts you confirmed: FAQ blocks, About copy, structured data. Nothing reaches you until it clears the gates below.
- Prove
Re-run the audit and the same questions go back to the same engines, scored by the same rules. You publish better content, and the number answers honestly, either way.
A separate fact-audit lane grades what the engines say about you for accuracy. Its machinery is described below, and it is in build: it publishes nothing to customers yet.
Sampling
Two engine rosters, on purpose.
Being seen and being right are different questions, so Jinn runs them on different rosters and does not bend one to match the other.
- The visibility lane
This lane answers how prominently the engines mention you, and rolls up to Brand IQ. Google AI Overviews counts here. Engine weights rebalance over the engines that actually answered, and the result carries a degraded flag whenever coverage was short.
- The fact-audit lane
This lane answers whether the engines get your facts right, on the engines proven to give factual claims. Google AI Overviews is left out of this lane by design, and the roster itself says so.
- Gaps are recorded
When an engine gives no answer to a question, the fact-audit run records which engine it was.
- Every engine, every question
Every question goes to every engine in the lane, twice per run. The grid covers the whole roster, not whichever engines were easy that day.
The fact-audit lane is in build and publishes nothing yet. These rules are how it is engineered, not a live customer surface.
Fact audit
Adjudication, not one model’s opinion.

When the fact-audit lane grades a claim an engine makes about you, it grades in two stages. Stage one checks each claim against your brand record where the record covers it, or against what the other engines agree on where it does not. Anything suspicious is flagged.
Stage two checks every flagged claim again with a live web search. A source counts only if the search tool itself found it: a link the model typed from memory does not count, so an invented citation can’t prop up a verdict.
Calling a claim wrong or invented takes a source from an approved list: your own site, government or education sites, or named news wires and encyclopedias. With anything less, the verdict drops to an unverified disagreement. Jinn does not accuse without a quality source, and the verifier enforces it.
A claim that can’t be sourced either way is left out of the accuracy count entirely, so you are not scored down for something nobody could check. Every suspicious claim is checked one by one, plus a sampled check on the rest, sized to catch the failure agreement can’t see: every engine repeating the same wrong fact.
When a score does publish, it will carry a ninety-five percent confidence range built from that sample. With too few confident answers, the code gives no estimate at all rather than a shaky one.
No accuracy percentage appears on this page, on purpose. The method must first clear its own agreement bar against a set of answers people have labeled. Until then you get the discipline, not the number.
Fixed gates
The gates are code, not another AI.
Every Playbook block passes checks that are plain text rules. They cost nothing to run, they can’t be talked round, and the same text always gets the same verdict.
Compliance hard-fails
Any banned word or phrase from your brand record fails the block: whole word, any case, with spacing evened out first so a line break can’t sneak a banned phrase through. A block that fails compliance can’t be published.
Voice is scored, honestly
A score from zero to a hundred on hedging, sentence-length mix and generic marketing filler, graded pass, warn or fail. The code calls it a rule of thumb, because that is what it is.
Regulated brands get a backstop
For brands in regulated categories, disease and efficacy claims are blocked before you see a fix. A draft that still breaks the rule after its retries is held back.
Structure, not promises
The readiness score on a rewrite grades citable structure: specific claims, figures, the answer first, evidence. It scores the content you ship, and it does not predict an AI citation or a Brand IQ change.
Honest machinery
Built so you can’t cheat it. Neither can we.
You can’t lobby the score
Brand IQ takes only the engines’ raw answers. Your confirmed corrections are not an input, and a pinned test runs the score with and without them and checks that the number is identical.
Corrections are yours to confirm
Each fact correction is yours to confirm or reject, in one step that fails safe if anything goes wrong. A confirmed correction grounds every later rewrite, and it does not touch the score.
Crawl checks tell three truths
When Jinn reads your site’s rules for AI crawlers, each crawler reports allowed, blocked or honestly unknown. An error page dressed up as a success never becomes a list of false “allowed” answers. These checks run on request inside an engagement; nothing crawls your site on a timer.
The record underneath
Every rewrite grounds in the brand record behind all of Jinn: 700+ signals drawn from your own site and evidence. Jinn reads the record. It does not invent one.
Questions
What people ask about the method.
Does an AI grade my Brand IQ?
No. The score is a fixed rules pass over where your name lands in each engine’s answer. No AI judge sits in the scoring path, so the same answers always produce the same number.
Can I pay, or optimize, to move the score?
You can publish better content and re-run the audit. You can’t lobby the number: the scored questions are the same on every plan, and confirmed corrections have no way in.
Does Jinn watch my site around the clock?
Monitoring is off by default; it is switched on when you want the trend line. Crawl checks are different: they run on request, inside an engagement. You get measurement you can repeat, not always-on theater.
Where does the fix content come from?
From your verified record and the corrections you have confirmed, drafted into page-level rewrite blocks and gated before you see them. Nothing is copied from competitors, and nothing goes out unchecked.
Why is there no accuracy percentage here?
Because the fact-audit method has not yet cleared its own bar against labeled answers. The bar is code, and the number stays unpublished until the method earns it.

Visibility: want to see it run on your brand? Let's talk.
- Tell us about your brand.
- A person on our team reads it.
- We read every application and reply within two business days.
Already use Claude or ChatGPT? Connect your AI, free