How to Get Named in AI Answers When 85% of Your Mentions Sit on Pages You Do Not Own

Someone who could be your next customer is asking ChatGPT which company to go with right now. You will not see that conversation and you will not know who got recommended instead of you. The buyer who gets steered elsewhere never lands on your site, so the lost deal leaves no impression, no click and no drop-off anywhere in your reporting.

This is the method Will Leatherman walked through in a Catalyst workshop on 30 October 2026, built on the work Catalyst has done with over 150 B2B clients. AI answer tracking means asking your buyers' real questions many times across ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews, then counting how often each company gets named. It is for B2B marketing and GTM teams who report AI visibility to a founder or a CEO and want a number that survives the follow-up question.

The buyer behavior behind it is already measured. G2 surveyed over 1,000 software buyers for its 2026 AI Search Insight Report. Half of them, 51%, now start research with an AI chatbot more often than with Google, and 71% use one somewhere in the process. 69% chose a different vendor than the one they expected, because of what an AI chatbot told them.

Why does your rank in ChatGPT change every time you check it?

Because the answer gets rebuilt from scratch every time, and the list of companies almost never repeats.

SparkToro had 600 volunteers run 12 prompts through ChatGPT, Claude and Google's AI Overview a combined 2,961 times. There is less than a 1 in 100 chance that two responses name the same list of brands. Claude repeats a list slightly more often and is even less likely to repeat the order.

No vendor is about to fix this. Thinking Machines Lab ran one prompt 1,000 times at temperature 0, every setting turned down to force the answer to repeat, and still got 80 different completions. The cause is how the servers batch requests when they are busy.

"The answer your buyer gets depends partly on who else is asking at the same moment." — Will Leatherman, Founder of Catalyst

Catalyst tested it live on the morning of the workshop. The same question went into 3 fresh ChatGPT sessions, "what are the best tools for tracking how your brand shows up in AI answers". The top spot came back Profound, then Peak AI, then Profound again, and the fifth slot was a different tool almost every time. Four tools appeared in all 3 answers. Three others appeared once each. Same question, minutes apart, 3 different lists handed to a buyer.

"A rank in ChatGPT is true for one specific ask, but then it's pretty much gone." — Will Leatherman, Founder of Catalyst

What should you count instead of a rank?

How often you get named across roughly 100 answers.

One hospital in the SparkToro data showed up in 69 of 71 answers and came first in only 25 of them. Report that as a rank and you look like you are slipping. Count how often you appear and you are in almost every answer. A rank is one phone call. A count is the whole poll.

Build the count like this.

1. Get 20 questions from your sales team, in the exact words buyers use when they say what they asked ChatGPT before the first call.
2. Ask each one 5 times in a fresh chat with no history, which gets you about 100 answers. Temporary chats in ChatGPT do this.
3. Write one line per answer with 4 fields, the question, whether you showed up, who showed up instead, and which pages the answer linked to.
4. Read the "who showed up instead" column against the list of competitors your sales team names. The two often do not match.

Read the change honestly once you have the number. At 100 answers your count can move about 10 points either way by chance alone, so 34 one month and 39 the next is noise. A move from 20 to 45 over a quarter is real. With fewer than 20 answers, report it as "3 of 8" and skip the percentage, because a percentage off 8 answers is a guess. We go deeper on separating signal from noise in 3 checks that tell you whether your AI visibility score really changed.

Which 20 questions should you track?

Spread them across 5 kinds so you catch buyers at every stage, roughly 4 questions each.

The questions matter more than the tool you track them in. Visibility trackers are largely interchangeable. A sales team that hears buyer language every day is a better source of queries than any keyword tool.

When your CEO asks where you rank in ChatGPT, the answer is that there is no fixed rank, you appeared in this many of the last 100 answers, and here is who appeared instead.

Which pages does AI read before it names anyone?

Mostly pages you did not write.

AirOps analyzed 21,311 brand mentions across ChatGPT, Claude and Perplexity and found 85% came from third-party sources rather than the brand's own domain. Almost 9 in 10 of those outside pages were roundups, comparison pages and reviews.

"Most of what AI says about you, you did not write." — Will Leatherman, Founder of Catalyst

The mechanism is the fan out. Google says its AI answers are rooted in core search ranking, so one buyer question triggers several background searches. A buyer asking which payroll tool suits a startup with contractors abroad might set off searches for startup payroll tools, international contractor payments, and a comparison of the two best known names. Each search returns its own page, a roundup here, a review site there, a comparison page somewhere else, and the answer gets stitched together from all of them. You are not trying to win one search. You are trying to sit on the pages that keep coming back across every search behind your buyer's question.

The split by question type is stark. Ask about a category in general and only about 2% of the links in the answer point at any company's own site. Ask about a company by name and almost 8 in 10 links go to that company's site.

"Your site talks to people who already know who you are, which is great for bottom-funnel, but other people's pages decide whether new buyers hear about you at all." — Will Leatherman, Founder of Catalyst

It works like a reference check. Nobody hires off the resume alone, they call the people who have worked with you, and AI does the same thing. That co-citation effect is the whole game, and what co-citation means for AEO explains why one mention next to an authority outperforms another post on your own blog.

Build your own list straight from the 100 answers. Write down every page they link to, count which sites keep coming back, and take the top 10. It usually lands on a couple of review sites, a few roundups and a comparison page or two. A review site showing in 31 of your 100 answers, one roundup in 22 and a comparison page in 14 means those 3 pages sit behind most of your answers. Start there. For how to tell the page types apart, see the 4 types of pages that drive AI citations.

Does schema markup get you recommended?

No, and the test is public.

Ahrefs tracked 1,885 pages that added JSON-LD schema between August 2025 and March 2026 and matched them against about 4,000 control pages. Citations moved by minus 4.6% in Google AI Overviews, plus 2.4% in AI Mode and plus 2.2% in ChatGPT, all statistically indistinguishable from zero. In a separate test, a price written only in hidden code was missed by all 5 assistants.

"It's like writing your price on the back of a price tag. AI only reads the front the same way your buyer does." — Will Leatherman, Founder of Catalyst

Your own pages still have a job. Make every heading a buyer question and put the answer in the first 2 lines underneath it. Rewriting just the heading and the structure that way lifted how often AI cited a page by about 17% in the testing Will walked through on the call. That is the whole on-site task. Hidden code is not going to get you recommended.

How do you get onto the pages that decide the answer?

Three plays, and this is where most of the budget should go.

Reviews. G2 found buyers go to review sites and Reddit to confirm what AI told them. Ask 5 happy customers for a review this month, starting with the ones who just renewed or said something kind on a call. Send them to the single review site your 100 answers link to most, give them one link, and tell them it takes 5 minutes.

Roundups. Most have a named author and a date. Send that author what has changed since they published, with one number a reader would care about. Keep it short, something like "you wrote the best payroll tool article in March, since then we've added this feature and this many companies use it, happy to send details if you update it."

Comparisons. Buyers ask AI how you compare to a named competitor more than they ask anything else, and somebody's page is going to answer that. Open every comparison page about you and check 4 things. Is the pricing current. Are the features they credit you with still right. Do they describe who you are for the way you would. Is the date recent. Anything wrong gets a short correction with a link to the source, which makes it easy for the author to say yes. Where no comparison exists for a question buyers keep asking, publish your own and keep it fair to the competitor, because AI quotes it when buyers ask about you by name.

"The default is more blog posts. That's what everyone tries to do. But what changes the answer is actually getting reviewed, getting into the roundups, and getting the comparisons right." — Will Leatherman, Founder of Catalyst

Plain mentions carry more weight than links now. Ahrefs studied 75,000 brands and found brand web mentions correlate with AI Overview visibility at 0.664 against 0.218 for backlinks, roughly 3 times stronger. YouTube came out stronger still, so a product walkthrough on someone else's channel or a customer talking about you on video counts. So does a real answer posted as yourself on Reddit when someone in your category asks for advice. Three ways to build entity authority covers how these compound.

Who owns each move?

Give every move one name, because this work falls between teams and quietly stops.

Put those 4 people in a 30 minute meeting once a month with 3 questions. Who showed up instead of us. Which pages changed. Which 3 do we go after next.

What do you send your CEO next month?

Three parts, in this order.

1. Who AI recommends when buyers ask about your category, and how often you appeared across the last 100 answers.
2. The pages AI reads to decide, which of those you are already on, and your plan for the rest.
3. What you are going to stop paying for. Usually a rank dashboard with no count behind it, or another round of hidden code. Killing it frees the budget for reviews, roundups and comparisons.

Take the live test as a worked example. Peak AI would report showing up in 3 of 3 answers, Profound showing up every time as well, and Athena HQ, Scrunch and Semrush once each. Then the 7 sites those answers linked to, and which of them mention Peak AI today. That holds up when the CEO opens ChatGPT and checks it.

The next 2 weeks are small. This week, get the 20 questions from sales. Next week, ask each one 5 times and write down who shows up and which pages get linked. Then pick the 3 pages you are missing from and go.

The takeaway

A buyer steered to a competitor never appears in your analytics, so go read the answers yourself, many times. AI mostly repeats what other people's sites say, which means the pages that matter most are not on your domain. Getting reviewed, getting into the roundups and fixing the comparison pages is what changes who gets named.

Start with the count. Run the free AEO audit with your work email. It writes the questions your buyers are actually asking and runs each one 10 times across ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews and Grok. You get who gets named instead of you, which best-of lists you are missing from, and a 90-day plan you can hand to the 4 owners above.