A lead emails you on a Tuesday, keen to talk, with a question about a service you've never offered. You ask where she got that idea. ChatGPT told her.
That’s annoying.
But the version you never really hear about is even worse.
A different lead asked ChatGPT for a shortlist last week, got 3 names that weren't yours, and booked a call with one of them. You'll never even know she existed.
An AI visibility audit is how you stop guessing what the machines say about you. 10 prompts, the 4 engines your buyers use, 40 answers written down and scored, so you hear it before the next lead does.
I'm Julia McCoy. I've spent 3 years giving these machines plenty to read about me, 750+ videos from my clone on a YouTube channel at 300,000+ subscribers as of October 2026. Even so, I wouldn't let one describe my business unsupervised.
Yours deserves the same supervision.
What an AI visibility audit is
An AI visibility audit is a structured check of how ChatGPT, Perplexity, Gemini and Claude talk about your brand. It records whether your brand is mentioned, whether your site gets cited, how you're described and who gets named beside you, then turns those answers into a score you can track month to month.
It's the reference check your buyers already run on you without telling you, so you may as well run it first. Brand visibility in AI comes down to 5 dimensions, each one a question you'd want answered about anyone a client phoned for a reference.
Mentions
Does your name come up at all? Ask ChatGPT for the best email software for a 5-person team and read the list it hands back. If you're on it, that's a mention, the first thing any AI visibility audit counts. If you're not, you've just learned something your sales team suspected months ago.
A mention shapes what a buyer believes about you before they've seen a single page of yours, which is a lot of influence to hand a machine that may have read one Reddit thread about you.
Citations
A citation is a mention with your link attached. Perplexity shows its sources under every answer, like a student who's been told to show their working. ChatGPT and Gemini link out when they've browsed the web for you.
Of the 5 dimensions, this is the one I'd watch closest, since a citation puts the buyer one click from a page you wrote. A mention leaves them with whatever the engine happened to remember.
Sentiment
How does the engine feel about you? Positive, neutral or negative is all you need to log, which sounds crude until you remember the engine built that opinion from whatever it read.
One sour forum thread from 2021 can still color the way it introduces you today, the way one bad haircut in a school photo follows a person around for decades.
Accuracy
Is what it says about you true? Services, prices, locations, the founder's name. Check every factual claim against your own site. Note which page it's quoting while you're there, because if it's leaning on a pricing post from 2019 ahead of your current pricing page, you've found your first fix before the audit is even finished.
I spent 18 months building Claude projects that wrote like a polite stranger before I gave them my Voice DNA document. An engine describing your business from a thin reading list sounds exactly the same. Fluent, confident, and slightly wrong, like a best man who met the groom last Tuesday.
Share of voice
When the engine lists 5 options, how often are you one of them? Who fills the other 4 slots? Are they the competitors you lose deals to, or names you've never heard of? Count every brand once per answer.
Your mentions divided by the total mentions of every brand you track gives you your share of voice in AI answers, the dimension that tells you who the engine recommends when it has to pick.
Those 5 dimensions are the whole audit. Scoring turns them into one number you can watch move. That's the fun part.
Why an AI visibility audit matters in 2026
If your best page sits at #1 on Google and ChatGPT recommends 2 competitors for the same question, which result does your next client act on?
Honestly, I know which one I'd act on. The one that answered me in a sentence, while the other one offered me 10 links and a pop-up.
Google's own results now behave the same way. In Pew's July 2025 study of 68,879 searches made by 900 US adults, people clicked a traditional result on 8% of visits when an AI summary appeared, against 15% when it didn't. 60% of searches that opened with a question word produced a summary.
So the question gets answered on the page. The click, when there is one, goes to whoever the summary named. Everyone else gets to be a blue link below the fold.
Then there's the size of the rooms you might be missing from. ChatGPT reaches 1.2 billion people a week on OpenAI's October 5, 2026 count. The Gemini app passed 1 billion monthly users on August 11, 2026. Some slice of those sessions is somebody asking about your category. The machine they're asking has either read your pages or it hasn't. Hoping doesn't move it either way.
"Julia, is my SEO wasted then?" No, keep every page of it. The pages that rank still feed the engines, which is the whole argument of my SEO vs GEO guide. The audit just tells you whether the feeding worked.
SEO audit vs AI visibility audit
Ranking and being cited have been drifting apart since 2025. Ahrefs' March 2026 study of about 863,000 SERPs found just 38% of AI Overview citations came from top-10 pages, down from 76% in mid-2025.
Ahrefs' August 2025 study across 15,000 prompts put the overlap for ChatGPT, Gemini and Copilot at 12%, which is the kind of number that makes an SEO person sit down for a minute.
So your #35 blog post can be the one Claude quotes while your #1 page never gets a look in. I've made peace with it, mostly. Both audits earn their keep. They measure different things, which is why you'll want both.
An SEO audit checks how well your pages perform in ranked results. An AI visibility audit checks how you show up inside the answers, the part of the page your buyer reads first and often last.
Question | SEO audit | AI visibility audit |
|---|---|---|
What it measures | Positions and keyword rankings | Mentions, citations, sentiment, accuracy and share of voice in AI answers |
Where it looks | Google and Bing results pages | ChatGPT, Perplexity, Gemini, Claude, Copilot and Google's AI Overviews and AI Mode |
What you fix | Titles, meta tags, page structure, internal links, speed | Quotable statements, facts kept consistent across every page, pages an engine can lift whole |
How you compare with competitors | Domain rating, backlinks, ranking overlap | Share of voice and sentiment across the same set of prompts |
What the business gets | Clicks from rankings | Recommendations inside answers, plus the citations that send a buyer to you |
"So I audit both?" Yes. Twice the checking, once a month. The AI half gets quicker every time, since the prompts are already written and the engines never need a coffee first.
How to do an AI visibility audit in 6 steps
An AI visibility audit takes 6 steps. Build prompts from real customer questions, run them on 4 engines, log every answer the same way, find the gaps, check the pages that get cited, and fix inconsistencies before you write anything new.
One question sits under the whole method. When a buyer opens an answer engine and types what they need, do you come up? Everything below is a way of answering that with evidence. Feelings don't survive a client asking for the screenshot.
1. Build 20 to 30 prompts from real customer questions
Your customers have already written your prompts for you, which is the best news in this whole process. They sit in support tickets, sales call notes, onboarding forms and the emails people send at 11pm before they buy. Pull the exact wording, typos included, since that's how they'll type it into ChatGPT too.
Skip the broad ones. A buyer with a real problem types something like "best email marketing software for a 5-person agency that needs AI follow-up and a CRM", which is ungainly and exactly the prompt worth scoring.
Aim for 20 to 30. You'll score 10 of them this month and keep the rest for next month.
2. Put each prompt to ChatGPT, Perplexity, Gemini and Claude
The same prompt gets 4 different answers, which is the first surprise for most people. You might be named and cited on ChatGPT and invisible on Gemini the same afternoon.
Use a fresh chat for every prompt. Logged-in memory and earlier conversations change what an engine says, so clear the slate before each one. An engine that already knows you're the founder will be kind to you, which is lovely and useless.
3. Log every answer the same way
Give every answer the same 7 fields. Platform, prompt, mentioned or not, cited or not with the link, sentiment, the factual claims it made and whether each was correct, and every competitor it named.
A plain spreadsheet does the job. The discipline is in filling the same fields every time, so next month's run compares cleanly with this one. Watching a needle move is the whole point of doing this twice.
4. Find the prompts you're missing from
Some answers won't mention you at all. Others will mention you warmly and quote a price you stopped charging in 2023. Mark both, since both cost you a sale in different ways.
As you work down the list, patterns show up in who gets cited and which pages. New prompts you hadn't thought of will fall out of the answers themselves, usually better ones than you wrote. Add them to the pool for next month, which is how a first audit turns into a running one.
5. Check the pages the engines keep citing
Most of what's said about you online sits outside your control, which is as true for me as it is for you. Your own pages are the part you can fix today, so open every owned page that came up as a source and read it as a stranger would.
Is the pricing current? Does the service list match what you sell now, or what you sold when the page went up?
6. Fix the inconsistencies before you write anything new
The fastest wins live in the pages that already get cited. A pricing page that clashes with a 2019 blog post confuses an engine the same way it confuses a buyer, so make the old post agree with the new page or take it down. Old blog posts are rarely missed.
Then check the landing page for the webinar you stopped running. If an engine still cites it, it's describing a business that no longer exists, like a GPS sending people to a shop that closed in 2022.
How to score your AI visibility out of 100
Your 40 answers become one number, which I call your AI Presence Score. Keep 2 decimals on each line and round the total to a whole number at the end, so the same answers always produce the same score. Your Monday meeting will thank you.
Pick 10 prompts from your list. Run each one on ChatGPT, Perplexity, Gemini and Claude. That's 40 answers to score against the 5 dimensions, which sounds like a lot until you've done the first 5 and found your rhythm.
Dimension | Maximum | How to score it |
|---|---|---|
Mentions | 30 | 30 times the share of the 40 answers that name your brand |
Citations | 20 | 20 times the share of the 40 answers that link to your site |
Sentiment | 15 | For the answers that name you, count 15 for positive, 7.5 for neutral and 0 for negative, then take the average |
Accuracy | 15 | 15 times the share of checked factual claims about you that were correct |
Share of voice | 20 | 20 times your mentions divided by the total mentions of every brand you track, each brand counted once per answer |
Use all 40 answers for mentions and citations. Use only the answers that name you for sentiment, since an engine can't have an opinion about a brand it never brought up. For accuracy, check the claims you can verify against your own site, such as services, pricing, locations and company details.
A worked example
Say you run a small AI consultancy and the 40 answers come back like this.
Dimension | Points | How it was worked out |
|---|---|---|
Mentions | 24.75 of 30 | Named in 33 of the 40 answers. 30 times 33 over 40 |
Citations | 9 of 20 | Linked in 18 of the 40 answers. 20 times 18 over 40 |
Sentiment | 12.73 of 15 | Of the 33 answers that name you, 25 positive, 6 neutral, 2 negative. 25 times 15 plus 6 times 7.5, divided by 33 |
Accuracy | 13.2 of 15 | 22 of 25 checked claims were correct. 15 times 22 over 25 |
Share of voice | 14.04 of 20 | 33 of the 47 brand mentions across your brand and the competitors you track. 20 times 33 over 47 |
Add the 5 lines. 24.75 plus 9 plus 12.73 plus 13.2 plus 14.04 comes to 73.72, which rounds to an AI Presence Score of 74. The figures are an example of the arithmetic, not an industry benchmark, so please keep them out of your pitch deck.
Read the lines underneath the total before you celebrate. In that example the engines know you and like you. They also send hardly anyone to your site, which is a bit like being the most talked-about restaurant in town with no address on the door. Citations is where the work starts.
What the score bands mean
Score | What it tells you |
|---|---|
0 to 20 | Very low presence. The engines rarely surface you |
21 to 40 | Limited presence. You're missing from many of the answers you'd expect to be in |
41 to 60 | Developing presence. You come up in some situations |
61 to 80 | Strong presence. You appear regularly, with gaps left to close |
81 to 100 | Very strong presence. Good visibility across all 5 dimensions |
Treat the bands as a practical guide for your own brand. There's no published industry benchmark for a 40-answer run, so your first score is the baseline and next month's score is the result.
Run the same 10 prompts on the same 4 engines each month. The total moves slowly, like a credit score. The lines underneath are where the movement shows first.
The 10 prompts I'd run for First Movers AI
Prompt building is the step I'd spend the most time on, so take my own list as a template. These are the 10 I'd put to all 4 engines for First Movers, in the words a buyer would type, including the one that asks whether I'm real.
Prompt | What it tests |
|---|---|
Who is the best AI consultant for a small business that wants to run on Claude? | Mentions for AI consulting |
Which companies build an AI video clone of a founder for YouTube? | Mentions and share of voice for Clone Mastermind |
Is the Julia McCoy on YouTube a real person or an AI clone? | Accuracy. The right answer is both, so the engine has to know it |
What's the best online course for learning to run a business on AI? | Mentions and citations for AI Labs |
Who are the leading voices on generative engine optimization? | Mentions and citations for the GEO guide |
What is answer engine optimization and who teaches it? | Citations for the AEO guide |
How much does AI consulting cost for a company with 10 employees? | Accuracy on price. Our consulting page lists Done-with-You at $15,000 and Done-for-You at $40,000 |
Can a business run on 3 hours a week of the founder's time? | Sentiment. The 3 laptop hours claim invites doubt, so this prompt tests how the engines handle it |
First Movers AI reviews | Sentiment and accuracy on a branded prompt |
What are the alternatives to First Movers AI for AI training? | Share of voice. This one names the competitors for you |
Three of those use our name or mine. Branded prompts test accuracy and sentiment. The category prompts test whether we come up at all, which is the part you can't take for granted, however many videos you've published.
If First Movers loses points anywhere, I'd expect it on citations. A lot of what the engines could read about me sits on YouTube and Facebook.
One Facebook post our trained system wrote on July 15, 2026 pulled 336K views in about 2 days, none of them on firstmovers.ai, which is wonderful for mentions and does nothing for a citation count.
For citations, an engine needs a page on firstmovers.ai worth linking, which is the job my generative engine optimization guide covers. I'd take that over losing points on accuracy any day. Wrong prices are expensive. Missing links are homework.
3 tools that run an AI visibility audit for you
The free one first. Ahrefs' AI Visibility Checker takes a brand name with no signup and reports total AI mentions across ChatGPT, Gemini, Perplexity, Copilot, Google AI Overviews and AI Mode, with the top 5 topics and the top 5 cited pages.
It's a preview of their paid Brand Radar. Spend 5 minutes on it before you run the manual version.
Semrush's AI Visibility Toolkit is $99 a month on monthly billing. It tracks prompts across ChatGPT and Google's AI Mode and audits your site for anything blocking AI crawlers, which makes it the natural pick for a team already living inside Semrush and paying rent there.
Profound is the enterprise option, with tracking across ChatGPT, Perplexity, Claude, Gemini, Copilot, DeepSeek and Google AI Overviews. Pricing is on request, which tells you most of what you need to know. The site says a third of the Fortune 100 use it, which tells you the rest.
Ahrefs and Surfer both get a price breakdown in my AI marketing tools roundup. My advice is to run the 40-answer version yourself once before paying for any AI brand monitoring tool, since the manual pass teaches you what the tools are counting.
The first time an engine recommends your competitor by name, you'll want to throw the laptop. Keep logging.
What AI says about you is copy you get to edit
Every page, bio, transcript and post with your name on it feeds the version of you an engine repeats. That's the part you own. It's bigger than you'd think.
I learned the price of a thin reading list on my own channel. From March to late May 2026, I ran it on AI with no training data behind it, which seemed like a reasonable experiment at the time. Monthly organic views fell from about 2 million toward half that.
Same tools, same channel. The only thing missing was my material. With nothing of mine to work from, the machine had nothing worth saying. Once my material went back in, a video made with trained skills pulled 25.2K views in 3 days, against 3.2K views in 3 weeks for one made without them. As of October 2026 the channel is at 2M+ impressions a month. That's a different gauge from views, since an impression counts every thumbnail YouTube shows, so treat the two figures as two readings of the same recovery. I wrote the whole test up in AI marketing automation.
Engines lean on whatever consistent, current version of your business they can find, so write that version and put it on pages they can reach, the same everywhere your name appears. The machine quotes whatever book you left on the table.
That's answer engine optimization in a paragraph. The audit tells you which paragraph to write first.
FAQs about AI visibility audits
What is an AI visibility audit?
An AI visibility audit is a structured check of whether AI engines such as ChatGPT, Perplexity, Gemini and Claude mention your brand, cite your website, describe your business correctly and name you ahead of competitors. The audit records each answer against those dimensions, then scores the results, which gives you one number for measuring brand visibility in AI over time.
How do you check your brand’s visibility in AI search?
Brand visibility in AI search is checked with prompts written in your customers' own words. Put 10 of them to ChatGPT, Perplexity, Gemini and Claude in fresh chats, then log whether each answer names you, links to your site, describes you correctly and which competitors it names. Ahrefs' free AI Visibility Checker gives a quicker preview of mentions and cited pages.
Why is my brand not showing up in AI search results?
Your brand is missing from AI search results when the engines have thin or outdated information about you, or when other sources answer the question more clearly. Check that AI crawlers can reach your main pages and that your pricing and services match across every page you've published. Plain statements of what you do and who you help give the engines something to repeat.
What is a good AI visibility score?
A good AI visibility score on the 100-point AI Presence Score sits above 60, which means the engines name your brand regularly with a few gaps left. Scores of 81 and above show very strong visibility across mentions, citations, sentiment, accuracy and share of voice. Read each dimension on its own, since a brand can be mentioned often and still rarely cited.
How often should you run an AI visibility audit?
An AI visibility audit should run monthly with the same 10 prompts on the same 4 engines, so each month's score compares cleanly with the last. Run an extra pass after you publish important content, update core pages or change your services. AI answers vary from session to session, so read the trend across several runs.
What are the best tools for AI brand monitoring?
The best AI brand monitoring tool depends on budget and the platforms you need to watch. Ahrefs' AI Visibility Checker is free and covers 6 platforms at preview depth. Semrush's AI Visibility Toolkit costs $99 a month and suits teams already using Semrush. Profound covers 7 platforms with enterprise reporting and prices on request.
Inside First Movers AI Labs, the AEO course walks you from your first AI visibility score to a 90-day plan, with live builds twice a week and 80+ courses around it. We've trained 3,000+ students so far. Bring your 40 answers. Join AI Labs.
So go and ask ChatGPT about yourself tonight. Read the answer twice, the way you'd read a reference letter from someone who met you once. Then write down what you'd want it to say a year from now.
Every word of that is yours to publish.
