What is an AI perception report?
An AI perception report documents what ChatGPT, Claude, Gemini and other AI assistants say about a brand: what they think it is, whether they name it when buyers ask for recommendations, and who they name instead. Each question runs multiple times; every answer is captured verbatim, dated, and turned into a prioritised fix list.
The same category trades under other names, including AI visibility audit and AI brand audit. This page defines what a proper report contains, how it differs from free checkers, monthly tracking tools and agency audits, what no report can promise, and how to read one without fooling yourself.
By Fred Loo · Rational Magic, Melbourne · Published 6 July 2026
Why AI perception reports exist
AI perception reports exist because buyers now ask AI assistants who to use before they ever search, and the assistants answer from a short set of brands they already know and can describe. A brand outside that set is not ranked lower. It is absent, and absence is invisible from the inside: you never see the conversations where a model named someone else.
The clearest evidence on how assistants choose names comes from SparkToro's Rand Fishkin, who examined 2,961 AI answers in January 2026: individual runs were wildly inconsistent, yet prominent, clearly describable brands kept resurfacing while unknown ones stayed out entirely. Models recommend from a consideration set. Entry comes from being genuinely known and cleanly described; placement cannot be bought, and it cannot be gamed.
Absence is one failure mode. Confusion is the other: if your name collides with something better documented, the models confidently describe that instead. We measured this on ourselves before selling anything. On 12 June 2026, zero of the five models we probed knew Rational Magic was a brand-strategy studio; the name split across five other identities, including a 2023 magazine essay and a tabletop role-playing game. That before-state is published and dated, and it is why this page exists.
Underneath both failure modes sits plumbing most brands have never checked. Most AI crawlers do not execute JavaScript, so content that only appears after scripts run is invisible to them; Google's Gemini, which rides Googlebot's rendering, is the notable exception (Vercel and MERJ, The Rise of the AI Crawler, December 2024). ChatGPT's live search retrieves through Bing, so a site absent from Bing's index is absent from those answers; Seer Interactive matched roughly 87% of SearchGPT citations to Bing top results in February 2025. And page structure moves answers: the GEO study presented at KDD 2024 measured visibility lifts in generative-engine answers of roughly 41% from adding authoritative quotations, 32% from statistics and 30% from citing sources (Aggarwal et al., arXiv:2311.09735).
A perception report is the instrument that turns all of that from anxiety into a measurement: what the models say today, on the record, with the causes tested.
What a proper AI perception report contains
A proper AI perception report contains five sections: what the models think you are, how often you are named when buyers ask for recommendations, who gets named instead, whether your site passes the machine gates, and a prioritised fix list. Behind all five sits an appendix holding every model output, verbatim and dated.
What the models think you are
Entity questions, "what is X" and "who is behind X", put to several models. This catches the quiet failure: a confident description of the wrong company, a merged identity, or a name-collision with something better documented than you.
Whether you are in the answer
Buyer-intent questions, "best X for Y" and "alternatives to Z", each asked multiple times per model and reported as counts: named in N of M runs. One run is a sample, not a verdict, and should never be presented as your standing.
Who gets recommended instead
The same runs read the other way: which names keep surfacing in your category's answers, and how often. This is the competitive half most quick checks skip, and it is usually where the sting is.
The gates, tested with evidence
Whether machines can read you at all: content that only renders through JavaScript, AI crawler access in robots.txt, presence in Bing's index, entity ambiguity, and whether key pages open with self-contained answers. Every check ships with its evidence, not just a verdict.
The fix list, prioritised
Plain-language actions ordered by likely impact, each tagged for whoever should do it: a developer task, a content task, or a strategy decision. A report that stops at measurement is a scoreboard, not a diagnosis.
Then the appendix: every model output behind every number, verbatim and dated. A report that presents conclusions without its raw outputs is asking you to take its word about perception, which is the exact problem you were trying to solve.
Ours looks like this. The AI Perception Report runs roughly twenty buyer-intent questions across five models, asks the key ones multiple times, covers the five sections above, and attaches the full verbatim appendix. Delivered as a private report page within two business days of a five-minute written intake. USD $290, one-off, with a 14-day money-back guarantee; the full $290 is credited toward the USD $1,950 strategy engagement if you upgrade within 30 days.
AI perception report vs free checkers, tracking tools and AI visibility audits
An AI perception report sits in the gap the rest of the market leaves open: deeper than a free checker's single run, cheaper and faster than an agency's AI visibility audit, and unlike a monthly tracking tool it diagnoses and prescribes rather than monitors. The other three are honest products; they answer different questions.
Against free instant checkers
A free checker answers one question: does AI mention you at all. HubSpot's AI Search Grader takes a short form and returns a graded snapshot; Ahrefs' free checker skips the signup and previews a partial result. Both run once and are partial by design, because the free answer exists to open a paid relationship. Treat them as a smoke alarm, not a diagnosis.
Against monthly tracking tools
Tracking tools measure your share of AI answers continuously, and the good ones measure precisely. What they do not do is diagnose or prescribe. An independent hands-on test by Graph Digital in April 2026 found the tools report visibility without telling you why you are absent or what to change, and that a scripted prompt battery captures about 80% of a paid tool's value for a few dollars a week. Published tiers at the tools we verified run from $29 to $989 a month. The sensible order: diagnose once, fix, then track if the stakes justify it.
Against agency AI visibility audits
An AI visibility audit is usually the agency-tier product: wider scope, human hours, usually sold through a sales conversation, and priced accordingly. The agency audit we verified in detail costs USD $4,500, credited in full against the agency's monthly retainer, which runs about USD $8,000 a month, if taken up within 60 days. That tells you what the audit is for: it is the front door to recurring work. Nothing wrong with that if an agency relationship is what you want. A perception report is the fixed-scope alternative: the measurement and the fix list, no retainer attached.
| Free instant checker | Monthly tracking tool | Agency visibility audit | AI perception report | |
|---|---|---|---|---|
| Price | $0 | $29 to $989 a month, published tiers | USD $4,500 at the agency we verified | USD $290 one-off |
| What it answers | Does AI mention you at all | How your share of answers moves over time | Broad AI search visibility, scoped per engagement | What models say, who wins instead, what to fix |
| Runs per question | One | Scheduled re-runs over time | Depends on the agency | Multiple runs, reported as counts |
| Diagnosis and fix list | No | No; measurement only | Yes | Yes; prioritised, plain language, tagged by owner |
| Verbatim model outputs | Partial by design | Depends on the tool | Depends on the agency | Every output, dated, in the appendix |
| Exists to sell you | A paid subscription | Its own monthly renewal | A monthly retainer | Itself; an upgrade only if the result stings |
Category-level view. Pricing and funnel shapes observed 6 July 2026 on public provider pages; the named free checkers are HubSpot's and Ahrefs'.
What an AI perception report cannot do
No report, tool or agency can make AI assistants recommend your brand, and none can promise a ranking. Answers change between runs, days and model versions. What a good report can do is measure honestly, date everything, and aim the fixes at the parts you control: being findable, readable and clearly describable.
The volatility is not a footnote; it is the headline finding of the measurement research. Don't Measure Once, a University of St Gallen paper published in April 2026 (Schulte, Bleeker and Kaufmann, arXiv:2604.07585), found that the sources cited in AI search answers change roughly 65% day to day, and nearly as much between runs minutes apart. When a major model version changed in June 2026, SISTRIX recorded about 47% citation volatility within 48 hours across 3.8 million responses. A number measured once is weather. A distribution measured across many runs starts to look like climate.
The same honesty cuts both ways. Three levers are widely sold and poorly supported; if a report leans on them, ask harder questions.
llms.txt as a lever. Across 137,000 domains studied by Ahrefs in June 2026, 97% of published llms.txt files were never fetched by the AI crawlers they were written for. Keep one as hygiene; do not pay for it as a strategy.
Schema as citation magic. A controlled Ahrefs experiment in May 2026, 1,885 pages against a 4,000-page control group, found adding JSON-LD did not lift AI citations. Schema earns its keep at a different job: telling machines which entity you are, which matters when your name is confusable.
Guaranteed percentage lifts. The widely quoted 40% visibility gain came from a simulated engine inside the GEO research, not from live assistants. Real measured gains are smaller and less stable, which is exactly why results belong in distributions, not promises.
That standard applies to everyone selling in this category, including us. A report tells you where you stand and what to fix. It does not promise where you will land.
How to read an AI perception report
Read an AI perception report as a distribution, not a verdict. "Named in two of fifteen runs" is information; a single run dressed up as a ranking is noise. Five checks tell you whether a report can be trusted: multi-run counts, per-model breakdowns, dates, a stated market, and verbatim evidence.
Counts, not single samples
Any one run can name you or miss you; across SparkToro's 2,961 examined answers, individual runs were wildly inconsistent even where the long-run pattern was stable. A trustworthy report states how many runs it made and reports mentions as N of M.
Per-model breakdowns
Five models are five different readers with different sources and habits. A blended overall score is convenient, but the per-model view is where the diagnosis lives. Insist on both.
Dates on everything
Answers reshuffle when models update, so a report is a dated baseline, not a permanent state. Use it as the before-state, ship the fixes, then re-measure against it.
A stated market
The St Gallen research also found answers shift with locale: a probe from Australia will not match one from the United States. A proper report says which market it measured for.
The appendix test
Every number in the report should trace to a verbatim, dated model output you can read yourself. No appendix, no trust.
See what the models say about you
The cheapest honest start is free: three standard questions about your brand, put to five AI models, with the verbatim answers in your inbox within two business days. No card, no call, no dashboard. If the models already describe you accurately, you are done; you never need us.
If they do not, the AI Perception Report is the deep look: the five sections above, roughly twenty buyer-intent questions across five models, multiple runs, and the verbatim appendix, delivered within two business days of a five-minute intake.
14-day money-back on the report · the full $290 credited toward the USD $1,950 strategy engagement if you upgrade within 30 days · async, no calls
Answered first, then explained
How do I check what ChatGPT says about my company?
Ask it the way a buyer would, in a fresh chat each time: "what is [your company]", "best [your category] for [your buyer's situation]", "compare [you] with [a competitor]". Ask each question several times, because answers change run to run; one answer is one sample, not your standing. Also check whether your site appears in Bing, since ChatGPT's live search retrieves through Bing's index. An AI perception report does the same thing systematically: roughly twenty questions, five models, multiple runs, every answer captured verbatim.
What does it mean if AI says nothing about my brand, or describes it wrongly?
Silence usually means the gates: your content is missing from the indexes models search (Bing above all), only exists behind JavaScript, or is blocked to AI crawlers in robots.txt. A wrong description usually means entity confusion: your name collides with something better documented than you. Both are testable, and both can be worked on directly because they sit on the parts of the system you control.
Is an AI perception report the same as an AI visibility audit?
Same family, different tier. "AI visibility audit" is usually the agency product: wider scope, sales-led, USD $4,500 at the agency we verified, with the fee credited toward a monthly retainer. An AI perception report is the fixed-scope version: the measurement, the competitive picture and the fix list at a fixed one-off price, with no retainer attached.
Can anyone guarantee my brand gets recommended by ChatGPT or other AI assistants?
No. In the best measurement study to date, the sources cited in AI answers changed roughly 65% day to day, and answers reshuffle again when models update. Anyone selling a guaranteed ranking is selling something the evidence does not support. What can be sold honestly is a dated measurement, fixes to the gates you control, and a re-measurement to show what moved.
How much does an AI perception report cost?
Free checkers cost nothing, run once, and are partial by design. Monthly tracking tools we verified publish tiers from $29 to $989 a month. Agency AI visibility audits run to USD $4,500 and up. Our AI Perception Report is USD $290 one-off: five sections, roughly twenty buyer-intent questions across five models, multiple runs, a verbatim appendix, delivered within two business days.
What is the free look?
Three standard questions about your brand, asked of five AI models, with the verbatim answers emailed to you within two business days. No card, no call, no dashboard. It is the same battery the paid work starts from, at surface depth. If the models already describe you accurately, you do not need anything else we sell.