Short answer: an AI visibility audit means asking ChatGPT, Perplexity and Gemini the questions your customers are already asking them, scoring every answer with a numeric grid, then fixing the errors at their source. It has become a basic hygiene check: according to a study published in July 2026, 93% of companies have at least one basic fact false or missing in AI assistants' answers, and small businesses are hit significantly harder than big brands. Here is the full method, with the scoring grid and realistic correction timelines.
AIs are already talking about you, and they are often wrong
In July 2026, the platform Searchable queried ChatGPT, Perplexity and Gemini more than 13,000 times about real companies, then checked every answer against those companies' verified sources (official registries, LinkedIn profiles). The results, reported by SME Magazine, deserve a second read.
- 93% of companies had at least one basic fact hallucinated or missing entirely from the answers.
- 50% of SMEs received at least one fabricated fact about them, versus 32% of large companies.
- AIs confused an SME's brand name with another company's five times more often than for large brands (4% versus 0.7%).
- An SME's services were described correctly only 87% of the time, versus 93% for large companies.
- The facts most often wrong or missing: company size, contact details and founding year.
The most troubling detail: these errors are delivered with confidence. The study measures a confidently-fabricated-facts rate of 5% for SMEs, versus 2% for large brands. A prospective customer reading a false but confident answer has no way of knowing. And while your website displays the right information, it may be an outdated or invented version of your business that is answering in your place.
Why you audit each platform, not "AI" in general
First reflex to correct: checking ChatGPT and assuming you are fine. Answer engines do not draw from the same sources. Profound's analysis of 680 million citations (2024-2025) shows that only 11% of the domains cited by ChatGPT are also cited by Perplexity. ChatGPT leans heavily on Wikipedia, which accounts for nearly half of its ten most-cited sources; Perplexity favours Reddit and data-rich content; Google's AI Overviews favour yet other sources.
The direct, counter-intuitive consequence: being well represented on one platform tells you almost nothing about what the others are saying. A serious audit therefore queries at least three engines separately, and your scores will probably differ widely from one platform to the next. That is normal, and it is precisely what the audit is meant to reveal.
The five-step method
- Build a list of 12 questions, in three categories. Four branded questions ("What do you know about [your business]?", "What services does [your business] offer?"), four unbranded questions phrased the way your customers phrase them ("Recommend a bilingual [your profession] for a small business"), and four comparison or objection questions ("[Your business] or [competitor], which should I choose?", "Can [your business] be trusted?"). Freeze this list: it is what will make your audits comparable over time.
- Query at least three platforms, under neutral conditions. ChatGPT, Perplexity, and Gemini or Google's AI Overviews, in private browsing, not logged into an account that already knows you, and without rephrasing when you dislike the answer. Copy each answer as is into a spreadsheet, with the date.
- Score each answer out of 6 points. Presence: 0 if you are absent, 1 if you are mentioned, 2 if you are recommended. Accuracy: 0 for a serious error (invented services, confusion with another company), 1 for an approximation, 2 if everything is correct. Sources: 0 if no source concerns you, 1 for generic sources, 2 if the AI cites your website or profiles. Twelve questions at 6 points each: a score out of 72 per platform.
- Sort the errors by type and fix them at the source. That is the table in the next section: each error type has a different correction channel and a different realistic timeline.
- Repeat the audit every quarter, with the same questions. Below 40% (under 29 points out of 72), you are invisible or misrepresented: basic corrections come first. Between 40 and 70%, your presence is inconsistent: work on third-party sources. Above 70%, your foundation is solid: monitor and move to fine-tuning.
To give a realistic order of magnitude: a typical small professional-services firm with an up-to-date website but little media presence often scores decently on branded questions and collapses on unbranded ones, which is exactly where new clients are. A 20-to-30-point gap between two platforms is nothing unusual: it mirrors the 11% overlap measured by Profound.
Fixing it: the right channel for each error type
| Error type | Where to fix it first | Realistic timeline |
|---|---|---|
| Missing fact (size, contact details, founding year) | Your website (About page), Google Business profile, LinkedIn, industry directories | 2 to 6 weeks for engines that read the live web |
| Outdated fact (old services, old address) | Update every source that contradicts the others, including old press releases and forgotten profiles | 2 to 6 weeks after harmonization |
| Entity confusion (mixed up with a similarly named company) | Reinforce your full legal name and structured data, differentiate your descriptions everywhere | 1 to 3 months, depending on how deep the confusion runs |
| Error baked into the model (persists despite an up-to-date web) | Third-party sources: local press, professional associations, recognized directories, customer reviews | Several months, at the pace of model updates |
The last row is the most important one. Muck Rack's Generative Pulse report (May 2026), which analyzed more than 25 million links cited by ChatGPT, Claude and Gemini across 17 industries, establishes that 84% of AI citations come from earned media: press coverage, third-party analysis, independent mentions. Your own website carries surprisingly little weight. Fixing your About page is necessary, but it is the external validation layer that tips the answers: an article in a local outlet, a listing in your professional association's directory or a quoted interview is often worth more than ten pages on your own site.
Two natural companions to this audit: checking what your customer reviews are telling the AIs, since they feed the engines' judgment directly, and measuring the traffic AIs already send you to track the effect of your corrections in your analytics. The audit tells you what the AIs are saying; the measurement tells you what it earns you.
Frequently asked questions
What is an AI visibility audit?
An AI visibility audit means asking assistants like ChatGPT, Perplexity and Gemini the questions your customers actually ask them, then scoring every answer against a grid: is your business mentioned, are the facts accurate, are you recommended, and which sources does the AI cite. The result is a numeric score per platform that becomes your baseline for fixing errors and measuring progress. It is the answer-engine equivalent of an SEO audit for Google.
How often should you repeat an AI visibility audit?
Once a quarter is enough for most small businesses, as long as you ask exactly the same questions under the same conditions so the scores stay comparable. Auditing more often adds little, because corrections take two to six weeks to show up in the answers of engines that read the live web, and several months for those that depend on the models' training data.
How long does it take to fix an error in AI answers?
It depends on where the error comes from. If the AI is misreading the live web, fixing your website, your Google Business profile and your directories produces an effect in roughly two to six weeks. If the error is baked into the model's training data, it can persist for months: the workaround is to multiply accurate, recent third-party sources, because according to Muck Rack, 84% of AI citations come from earned media, not from brand websites.
Want to know what ChatGPT answers when a customer asks who to recommend in your field?
Book a discovery call