AI visibility
What assistants say about you when nobody is watching.
There is no Search Console for ChatGPT. If you want to know what an assistant tells somebody who asks for a recommendation in your category, the only way to find out is to ask it, systematically, and record what comes back.
How do you find out what AI assistants say about your business?
By asking them, the way a customer would, on a fixed schedule, and recording the answers. A defined set of prompts, run across the assistants that matter in your category, repeated regularly because outputs vary between sessions for the same question.
What gets recorded is whether you were named, whether what was said about you was accurate, and which sources were cited instead of you. That third item is usually the most useful, because it shows what the model treats as authoritative in your field.
Nobody can promise a position in an AI answer. There is no placement mechanism, and any vendor offering one is describing something that does not exist.
Why this is a monitoring problem, not a ranking one
A search engine returns a list you can check. An assistant writes an answer that differs between sessions, between users and between phrasings of the same question. There is no position to track, which is why this is measurement rather than optimisation.
That does not make it unmeasurable. It makes it a sampling exercise. Ask the same question repeatedly and a pattern appears: which businesses get named, how often, and what the model believes about each of them.
The findings are frequently uncomfortable. Businesses discover the assistant has their old hours, believes they do not offer a service they have offered for years, or names three competitors and not them.
Those are fixable, which is the point of doing it. Not by influencing the model, which nobody can do, but by correcting what it is reading.
Why this is worth watching now
BrightLocal's 2026 consumer survey, base 1,002 United States adults. A vendor study with a stated method, reported as such.
SourceLocal Consumer Review Survey, BrightLocal, 2026, base 1,002 US consumers
The third figure is the one most businesses have not considered. A summary of your reviews is being written and read back to prospective customers, and its substance depends on what your reviews actually say rather than on your average.
How the monitoring actually works
This is deliberately boring. Consistency in method is what makes the results comparable over time.
Write the prompts a customer would actually use
Not keyword phrases. Whole questions in the words a person would say out loud. Who is the best dentist in Oak Brook for implants. Which urgent care near Elmhurst is open now. Is there a family law firm in Downers Grove that handles collaborative divorce.
- Category plus place, as a real question
- Specific service questions, not only the general one
- The comparison question, which is where competitors get named
Run them across the assistants that matter
Which assistants matter depends on your customers rather than on which is most discussed. The set typically includes the major general assistants and Google's own AI features, and it changes as products change.
Repeat on a schedule, in clean sessions
Outputs vary between sessions, so a single run tells you very little. Repetition in sessions with no prior history is what turns an anecdote into a pattern, and it is the step most vendors skip.
See the remaining steps: How the monitoring actually worksHide the remaining steps: How the monitoring actually works
Record three things each time
Were you named. Was what it said accurate. Which sources were cited instead. The third is the most actionable, because it shows what the model treats as authoritative in your category and therefore where you need to be present.
Fix the inputs it is reading
Contradictions between your website, your Google Business Profile and your listings. Services missing from your profile. Out of date hours. Thin pages that do not answer the question. This is where the actual work is and it is mostly ordinary local search work.
Re-measure and report as an observation log
Reported as what was asked, what came back, and what changed. Not as a score, because a score would imply a precision that does not exist.
None of this requires a proprietary tool. It requires somebody doing it consistently and writing down what happened.
What you can and cannot influence
You cannot influence the output directly. There is no submission process, no placement mechanism and no advertising product that puts you into an organic AI answer. Anybody offering one is describing a thing that does not exist.
What you can influence is what the systems read about you, and that turns out to be quite a lot.
Read the full breakdown: What you can and cannot influenceHide the full breakdown: What you can and cannot influence
Consistency is the first lever. When your website says one thing and your Business Profile says another, a model either hedges or picks one, and it may not pick yours. Making every source agree is unglamorous and it is the single most effective action available.
Completeness is the second. Services that are not listed anywhere cannot be attributed to you. A profile with one generic service entry is invisible for every specific question.
Third party mention is the third and the strongest. What other sources say about you carries more weight than what you say about yourself, which makes professional directories, associations, local press and genuine industry citations more valuable here than anything on your own site.
And reviews, because assistants summarise them. Their substance, not just the average, is what gets read back.
Claims to be sceptical of
AI visibility is currently the least regulated corner of marketing and the easiest place to be sold something meaningless.
- A guaranteed citation in ChatGPT, Perplexity or anywhere else. No placement mechanism exists.
- An AI visibility score with no stated method. Ask which assistants, which prompts, how many runs, and over what period. If those cannot be answered, the number is decorative.
- A proprietary algorithm for AI ranking. Google states plainly there are no additional requirements or special optimisations for its AI features.
- A special file to add to your site for AI. Google states explicitly that no AI specific files or markup are needed.
- A single screenshot of an assistant naming you, presented as evidence. Outputs vary between sessions, and one favourable run proves nothing.
See the full checklist: Claims to be sceptical ofHide the full checklist: Claims to be sceptical of
- Content published at volume to feed the models, which Google's spam policies address as scaled content abuse regardless of how it was produced.
- Blocking AI crawlers described as an optimisation, which removes you from the answers rather than improving your position in them.
- Any statistic about AI search with no named publisher, year and link. In this field the invented numbers outnumber the real ones.
The score question is the fastest filter. A real measurement has a method you can repeat; a marketing score does not.
How this differs from rank tracking
Treating this like rank tracking produces false confidence in both directions.
| Item | Rank tracking | AI visibility monitoring |
|---|---|---|
| Output | A position | A written answer |
| Stability | Comparable day to day | Varies between sessions |
| Official data | Search Console | None |
| What you measure | Position and clicks | Named, accurate, cited instead |
| What you change | The page | The consistency of every source about you |
| Guaranteed placement | No | No |
The last row is the same in both columns, and it is worth saying because in one of them it is still widely promised.
What a useful report looks like
The prompts used, verbatim, so you can run them yourself and get a comparable result.
For each prompt and assistant: whether you were named, what was said about you, and which businesses and sources were named instead.
Read the full breakdown: What a useful report looks likeHide the full breakdown: What a useful report looks like
A list of inaccuracies found, with where each one probably originated. Wrong hours usually trace to a stale listing. A missing service usually traces to an incomplete profile. A confused description usually traces to contradictory pages.
What changed since the last run, stated plainly, including when nothing changed.
And alongside it, the proxy measures that are actually trackable: referral traffic from assistant domains where analytics can see it, and branded search volume over time, which remains the most reliable available indicator that more people have heard of you.
What will not be in it is a percentage score, because we have no defensible method for producing one.
We do not publish a price for this piece of work on its own, because the right scope depends on what already exists. What is published is the bundle pricing: 2,400, 3,600 or 4,800 dollars a month depending on which channels are running. You can read the full breakdown on the pricing page, and you will get an exact number in writing before anything starts.
How this connects to the rest
This is the measurement half of generative engine optimisation. That page covers the work; this one covers finding out whether it did anything.
The fixes it produces are almost always ordinary local search work: Google Business Profile management, local listings consistency and better answers on your own pages.
Review management matters here more than most people expect, because assistants summarise reviews and the substance of yours is being read back to prospective customers.
And link building and digital PR is the lever with the most direct influence, since being cited by sources a model reads is the closest thing anyone has to affecting what it says.
Ask an assistant about your own category tonight.
Type the question a customer would ask, in a fresh session, and see who gets named. It takes two minutes and it is usually the moment this becomes a priority.
Straight answers.
Can you guarantee we will be cited by ChatGPT?
No, and nobody can. There is no placement mechanism, and outputs vary between sessions for the same question.
What can be done is improve the inputs and measure what comes back. That is designed to improve the odds, not to control the result.
Why does it say something different every time I ask?
Because these systems generate rather than retrieve a fixed list, and the output depends on phrasing, session history and the model version.
That is exactly why one run proves nothing and why the monitoring has to be repeated in clean sessions to produce a pattern.
An assistant said something wrong about us. Can it be corrected?
Not directly, and usually yes indirectly. Most inaccuracies trace to a source the model read: a stale listing, an incomplete profile, or contradictory pages on your own site.
Correcting those changes what it reads, and these systems retrieve current information rather than working entirely from a fixed snapshot, so corrections can show within weeks.
Should we block AI crawlers?
Only if you do not want to appear in AI answers. Blocking the crawler that reads your site for an assistant removes you from its answers.
For a publisher selling access to content that can be rational. For a local business trying to be recommended it is self defeating.
Is there a tool that does this automatically?
Several vendors sell one. They are doing what is described on this page, with a method you should ask about before trusting the number.
Which assistants, which prompts, how many runs, over what period. If those questions cannot be answered, the score is decorative.
Where this comes from.
Primary documentation and published research behind the guidance on this page.
- Google Search and AI features (opens in a new tab)Google's own statement that no special AI optimisation exists.
- GEO: Generative Engine Optimization, Aggarwal et al., arXiv (opens in a new tab)The peer-reviewed paper that introduced the term.
- Google users are less likely to click when an AI summary appears, Pew Research Center (opens in a new tab)
- Google spam policies for web search (opens in a new tab)Link spam, scaled content abuse and site reputation abuse.
- Guidelines for representing your business on Google (opens in a new tab)
Talk to the team
A short call, a look at how the business currently shows up, and a straight answer on what we would do first.