Your brand has a second reputation. You didn't write it.
For twenty years the question was simple: where do you rank on Google? Now a buyer opens ChatGPT, Claude, or Gemini and types a question in plain language. The engine answers in a paragraph. It names some brands and skips others. It states facts with total confidence, and sometimes those facts are wrong. The AI Visibility Audit is the practice of reading that answer before your customer does - and doing something about it.
The method is deliberately unglamorous. You write a set of questions the way real buyers ask them. You run each question across every major engine. You record whether your brand shows up, where it sits, who appears instead of you, and which sources the machine trusted to build its reply. Then you sort the gaps by how much they cost you. The mental model is closer to financial reporting than to a product launch: measure, benchmark, act, repeat.
Ask three chatbots the same question and you get three brands back. The gaps between them are the most honest market research you'll ever read. The audit, in one sentence
Six steps, one afternoon
None of this requires a data team. It requires patience and honesty about what you find.
Write the questions buyers actually ask
Build 15 to 30 prompts across four intents: branded (your company and products), category ("best tools for X"), problem-solution (a pain point, no brand named), and comparison (you versus a named rival). Pull the phrasing from real sales calls and support tickets, not a brainstorm.
Run every prompt across every engine
Put the same question to ChatGPT, Claude, and Gemini - add Perplexity if your buyers use it. For each answer, record four things: do you appear at all, where in the reply, which competitors take your place, and which sources the engine cited.
Measure consistency
Line the answers up side by side. A brand that shows up in ChatGPT and vanishes in Gemini has a distribution problem, not a messaging one. Track appearance rate and position over time, and re-run on a schedule so you see the trend, not a snapshot.
Hunt for hallucinations
Read each answer against what you know is true. Flag invented features, wrong founding dates, stale pricing, capabilities you don't offer. These are the highest-priority items: a confident error is repeated identically to every buyer who asks.
Find the missing knowledge and use cases
Note where the engine is simply thin - a product it can't describe, a use case where a competitor owns the answer. Every use case the AI can't name is a recommendation you'll never receive. This gap list is your content brief.
Prioritize the fixes
Rank by reach and risk. Correct hallucinations first, then fill missing product knowledge and contested use cases, then strengthen the high-authority sources the engines cite - so the true, corrected story is the one they learn next time.
Four machines, four foundations
They disagree because they're built on different ground. Understanding the source explains the answer.
The scoreboard changed and most people didn't look up
Share of voice used to be an advertising metric. In the age of answer engines it is closer to a survival metric, because the reply a buyer reads is often the only research they do. If the model never names you, you were never in the running - and you won't see a rejection email to tell you so.
This is why hallucination detection now sits at the center of the practice. A wrong ranking on a search page is annoying. A wrong fact stated with authority inside an answer is a small, repeatable act of misinformation about your company, delivered on request, at scale. Catching it early is brand safety, not vanity.
A hallucination about your brand is a confident lie told to every buyer who asks. Why hallucinations rank first
The good news is that the fix loop is short. Because the engines learn from the open web and the sources they cite, improving the record - clearer product pages, stronger third-party mentions, accurate structured facts - eventually changes the answer. The audit tells you exactly which record to fix, and in what order.
The five most common ones
What is an AI Visibility Audit?
A structured check that runs realistic buyer-intent prompts across ChatGPT, Claude, and Gemini, records whether and where your brand appears, compares you to competitors, and flags any inaccurate or missing information.
Which questions should I ask the engines?
Build 15 to 30 prompts that mirror real buyers: branded questions about your company and products, category questions like "best tools for X", problem-solution prompts with no brand named, and head-to-head comparisons with named competitors.
How do I measure consistency?
Run the same prompts across each engine and compare. Track whether you appear, where you sit in the answer, and whether the facts match. Repeat on a schedule - quarterly or more often - to see the trend rather than a single snapshot.
How do I spot hallucinations?
Read each answer against known facts about your brand. Flag invented features, wrong founding dates, outdated pricing, or capabilities you don't offer. These confident errors are the highest-priority fixes because every buyer sees the same wrong answer.
How do I prioritize improvements?
Rank gaps by reach and risk. Fix hallucinations first, then fill missing product knowledge and use cases where competitors dominate, then strengthen the high-authority sources the engines cite - so the corrected story is the one they learn.