What an AI answers audit can and can't tell a client
What an AI search audit from a one-person AI agency measures for a local business, what it can't promise, how to run one fairly, and how to explain limits.
A dentist asks her web designer a question she's been turning over for weeks. Her nephew typed "best family dentist near me" into ChatGPT and got three practices. Hers wasn't one of them. Can the designer fix that? And how would anyone know if it worked?
There's a fair service in that question for a one-person AI agency, and an unfair one. The unfair one promises she'll "show up in AI" and sends a colourful score each month. The fair one checks what AI assistants actually say about her practice, writes down exactly how it checked and when, fixes what's wrong at the source, and checks again.
That fair version is an AI answers audit. It tells a client what AI assistants said about their business, on a date, for a set of questions, with the sources they cited. It can find wrong details and gaps you can fix. It can't tell them how many customers came from AI, and it can't promise what tomorrow's answers will say. Explaining both halves plainly is what makes it a service worth paying for.
What does an AI visibility audit measure?
A dated snapshot of answers. Nothing more mysterious than that.
You pick a list of questions a real customer might ask, ask each one in the assistants that matter, and record what came back. For each answer, you note:
- Was the business mentioned? By name, and where in the answer.
- Were the details right? Name, address, phone, hours, services, prices if given.
- Which sources were cited? The business's own site, its Google listing, directories, review sites, news.
- What else was said? Competitors named, anything outdated or odd.
For the dentist, a question list might be: "family dentist in [suburb]", "dentist open Saturday near [suburb]", "does [practice name] do emergency appointments", "[practice name] phone number". Some questions are about finding her; some are about whether what's said is true.
The value is in the record. A client can see, for a given day, what customers asking those questions were told, and where those answers came from.
What can't an AI answers audit tell a client?
Four things, and saying them out loud at the start is the difference between a trusted service and a disappointed client.
How many customers came from AI answers. Google's guidance on AI features says that sites appearing in AI Overviews and AI Mode are counted in the overall web search traffic in Search Console, in the normal web search report. There's no separate AI line to quote. Other assistants may send visits that look like any other link. An audit shows what the answers said, not how many people read them or acted on them.
What everyone sees. Answers vary. Google's help page for the Search Console Performance report notes that search results are specific to the time, place, device and recent history of the person searching. OpenAI's help page for ChatGPT search says it may use an approximate location and, if memory is turned on, saved memories when it searches. Two people asking the same question can get different answers. An audit records what you saw, with how you asked, and that's all it can honestly claim.
That the answers are right. OpenAI's own help page says search results and citations can be incomplete, outdated or incorrect. An audit can catch that, but it can't stop it.
A ranking. There's no public "position one" in an AI answer to track, and any score that turns answers into a single number is the auditor's invention. Report what you observed, not rankings.
How do you run an audit fairly?
Write down the method before you start, and use the same one every time.
| What to record | Why it matters |
|---|---|
| The date and time of each check | Answers change; a result without a date is useless |
| The assistant and mode used | Different tools and settings give different answers |
| Signed in or not, memory on or off | Personal history can change answers |
| The location you searched from | Local questions depend on where you are |
| The exact wording of each question | Small wording changes can change the answer |
| The full answer, saved as text or a screenshot | So the client can see it, not just your summary |
| The sources cited | This is where fixes happen |
Use a clean, signed-out session where the tool allows it, or say plainly that you didn't. Ask each question more than once if you have time, and record whether the answers matched.
Write the method into your service recipe so every audit for every client follows it. When a client compares this quarter with last quarter, the comparison only means something if the method stayed the same.
What should the audit report say?
Keep it short and concrete. A useful report has four parts:
- What we asked, where, and when. The question list and the method, in two or three lines.
- What the answers said. Per question: mentioned or not, details right or wrong, sources cited. A simple table works.
- What's wrong and where it comes from. For example: "Two answers gave your old phone number. Both cited a directory listing that still has it."
- What we'll fix and when we'll check again.
What the report should never say: that the business "ranks" in an assistant, that visibility went up by a percentage you can't trace, or that more AI mentions will bring more patients.
What can a business actually improve?
Mostly the sources the answers draw on. Google's guidance says there are no extra technical requirements for appearing in its AI features beyond being eligible for normal search, and that ordinary good practice applies: pages Google can crawl, important information in text, structured data that matches the visible page, and an up-to-date Business Profile.
For a local business, that usually means:
- Correcting wrong facts at the source. The website, the Google Business Profile, and the directories that were cited. If ChatGPT repeats wrong business information, the fix is almost always an outdated listing or page, not the assistant; there's a step-by-step guide to fixing wrong business details AI assistants repeat.
- Saying clearly what the business does. A plain services page, with the suburb, the hours and the questions customers ask, in text.
- Keeping local search basics in order. Much of what helps a business in AI answers is the same work that helps in ordinary local search. The difference between GEO vs SEO is smaller than the names suggest.
Then check again, with the same method, and record what changed. Sometimes nothing changes for a while. The report should say that too.
How do you explain the limits to a client?
In the first conversation, before they pay. Something like:
"I'll check what ChatGPT and Google's AI answers say about your practice for the questions your patients are likely to ask, write down exactly how and when, and fix anything that's wrong at the source. I can't promise you'll be mentioned, and I can't tell you how many patients come from AI answers, because nobody can measure that reliably. What you'll get is a dated record of what's being said about you, and a list of fixes."
A client who hears that and still wants the service knows exactly what they're buying. A client who wants a guarantee will find someone who'll sell them one, and that's fine.
If you're deciding whether to offer this at all, it sits well alongside care plans and local search work; weighing which service to add after websites is a good place to start. If you already sell local services as a marketer, the same rule holds: selling services honestly to local businesses is easier when the report shows exactly what you did.
Where Volant fits
Volant's AI visibility service records what AI assistants say about a client, with the date and method, and reports what it observes rather than rankings. The questions, answers and fixes sit on the client's record next to the rest of their work, so each audit starts from the last one. The services page shows it alongside the other services you can offer after the website.
Whatever you use, sell the record, not the promise. A dated, honest account of what's being said about a business is worth paying for. A score nobody can check isn't.