The 5 Step AEO System That Gets AI to Name Your Company
Run the same 5 steps Catalyst runs for every client, starting with a 20 minute fact check, and you will know which pages are writing the answer your buyers see and which one to fix first.
Will Leatherman
Founder, Catalyst

TLDR
Answer engine optimization is the work of getting AI tools to name your company when a buyer asks, and for a B2B company that AI does not already know from memory, the pages it reads that day decide the answer. The system has 5 steps. Fact check what AI already believes about you, audit 20 real buyer questions to see who gets named instead, rewrite your own pages so a model can quote them, earn mentions on the pages AI already reads, then track 3 numbers every week. The first 3 moves take about 2 hours total and tell you more than most teams learn in a year.
A buyer types "which platforms should we look at" into ChatGPT, gets 3 names back with a reason for each, and contacts one of them. That shortlist gets written before anyone visits your website, and nothing in your analytics tells you the conversation happened.
The scale of the shift is already measurable. Bain found that almost half of buyers now start in an AI tool or use it right alongside Google, and that small and mid-size business buyers build their vendor shortlist the same way. TrustRadius found most buyers seriously consider 3 products or fewer. Your buyer is looking at roughly 3 companies, and more and more an AI picks those 3 for them.
Catalyst has taken more than 100 B2B companies from invisible in AI search to the top recommendations across ChatGPT, Gemini and Perplexity, worth more than 8,000 AI-driven sessions and roughly $8 million of pipeline influence in 12 months. What follows is the 5 step system behind that, taken from a live workshop, with a made-up payments company called XYZ Pay carried through every step so you can see what good looks like before you run it on your own.
What is the difference between AEO, GEO and AI SEO?
Nothing meaningful. They are the same work under different labels. AEO stands for answer engine optimization, which means getting AI to name you when buyers ask. Google still calls the broader discipline SEO. Each label is unpacked here if a stakeholder is arguing about which one to put in a budget line.
What you actually measure is AI visibility, which is how often AI names you across the questions your buyers ask. There are 3 levels of being named, and only one of them puts you in a deal.
| Level | What it means | What it gets you |
|---|---|---|
| Mentioned | Your name appears in the answer | Awareness, no click |
| Cited | A page of yours is linked as a source | Some traffic |
| Recommended | AI tells the buyer to use you | The meeting |
"Being cited sends you some traffic, but being recommended is actually what gets you the call." — Will Leatherman, Founder of Catalyst
The gap between the second row and the third is where most companies get stuck, and it is covered in why AI search names you but never recommends you.
How does AI decide which companies to name?
From 2 places. Training data, which is what it learned before release, and live search, which is the pages it reads the moment someone asks. The biggest brands get named from memory, and that part moves slowly on the model provider's schedule. Live search is the part you can move this quarter.
For a company your size, live search is most of the game, because the model usually does not know you from memory. Researchers at the University of Toronto shuffled the pages a model was reading and found the ranking of well-known brands barely moved, while smaller companies moved almost twice as much. What is on the pages it reads is what decides your answer.
Watch how it builds one. A single buyer question turns into several searches, so platforms that compare well, what each one charges by country, and the tax rules for contractors. The model reads what comes back, picks the passages that answer each piece, and checks whether more than one source agrees. Names that appear across several sources make the cut. A name on only one page usually does not.
The sources behind one answer are a map of the whole problem, split between your own pages, other people's pages, and whether the two agree about who you are. Ahrefs compared the sites each engine cites most and found only a handful appear in all 3 of Google, ChatGPT and Perplexity, so getting named on one does not carry over to the others.
"AI names you when the pages it reads, yours and everyone else's, say the same clear thing about you." — Will Leatherman, Founder of Catalyst
All 5 steps hang off that sentence.
Step 1. What does AI already believe about you?
Find out before you try to get named more, because it is wrong more often than teams expect. Ask the model what it knows about your company and have it list every fact, where that fact came from, and how sure it is. The check takes about 20 minutes.
On XYZ Pay it came back with 3 kinds of answer. The Salesforce integration was true. The number of countries was wrong. A security certification it listed had expired. Every one of those facts lands in a buyer's head before they ever speak to you.
Mark each fact as true, false or out of date. Then build a fact sheet, which is one document holding every fact about your company and where it came from, so your pricing, your customers, what you integrate with, and the claims you never want made. Every page you write afterwards draws from it, so AI hears you described the same way everywhere.
Then fix the sources, because a wrong fact came from somewhere. An old pricing page, a review site still listing last year's features, an article about a different company with your name on it. The prompt tells you where each one came from, so you know which page to update and which site to email.
"The thing with AI is it's very agreeable. If you don't provide the accurate information, it will go and find what it thinks is the most accurate and provide that to the user." — Will Leatherman, Founder of Catalyst
Step 2. Which buyer questions name you and which name a rival?
Pick one use case first, which is one job buyers hire you for. For XYZ Pay that is paying contractors abroad. Give it a goal you can check, so being the company AI recommends first on contractor payouts, and a number you can hit, so getting named on a couple more questions within 3 months. Small and real beats ambitious and unmeasurable.
Then write 20 buyer questions in the buyer's own words, pulled from where buyers actually talk. Your sales calls, your support tickets, Search Console, and the forums where your category gets discussed. Tag each one as research or buying.
| Question | Type |
|---|---|
| "How do I pay contractors in Brazil without setting up a company there?" | Research |
| "Which payout platform has the lowest fees?" | Buying |
Four checks keep the list honest. Every question has to be one a real buyer would say out loud. It has to match the use case you picked. Search volume sets the order and never rules a question out, because the buying questions that close barely register in volume. And you add the questions your competitors already win, so you can see what you are up against.
Run each question at least 10 times. Language models are not deterministic, so they do not return the same answer twice. SparkToro found you almost never get the same list of brands back. A marketing leader asked about this recently after a Catalyst scan said AI never named them, while someone on their team found themselves in an incognito tab on a VPN. Both results were real.
"Every time you ask, the AI runs a fresh set of searches and writes a fresh answer. So one person checking once is really just a roll of the dice." — Will Leatherman, Founder of Catalyst
The number that matters is how often you get named across many runs and whether it is going up. On XYZ Pay's buying questions run 10 times each, Rival A got named almost every time, Rival B most of the time, Rival C about half, and XYZ Pay never. That zero is common. About a third of the companies in Catalyst's AI Visibility Gap report were not named once on their own buyer questions. How to audit your AI search ranking in 20 minutes walks the manual version of this step.
Then pull it onto a map. For every use case you sell, count how many of its questions name you, how many a competitor already holds, and how many are still open, then weight that by the share of new revenue the use case brings in. On the sample, contractor payouts comes first because most of its questions are open and it carries the most revenue. Global payroll is already owned by one competitor, so it waits. Teams that skip the map start where they feel like writing, which is usually the use case a rival already owns.
Can AI even read your website?
Two things have to be true, and both fail quietly.
The bot has to be let through. That is set in your robots.txt file, and also in your CDN or firewall, which can turn a bot away before your site ever sees the request. Catalyst tested almost 800 sites as if they were AI bots, and most of the sites turning bots away had nothing in robots.txt saying so. The block was coming from the cloud host, with Cloudflare the most common culprit.
The words have to be on the page. Most AI bots outside Google's do not run JavaScript, so a page that builds its text with JavaScript arrives effectively empty. A study of small business and B2B homepages found a quarter of them were hiding key content from ChatGPT's bot for exactly this reason.
Here is the 1 minute version you can run today. Turn JavaScript off in your browser and reload your homepage. Whatever is still visible is roughly what most AI bots get. On the XYZ Pay example, the headline and the pricing table both disappear.
While you are in robots.txt, check which bots you are blocking. OpenAI documents that blocking its search bot removes your site from ChatGPT search answers, so let the search bots in. Training bots are a separate decision and genuinely your call. If you use the new no-training line that Google, OpenAI and others agreed on, write a Google-Extended rule too, or Google may drop you from Gemini's answers as well.
Step 3. What makes a page AI can actually quote?
AI lifts individual passages out of a page, so each passage has to carry the answer on its own. Four things make one liftable.
- Answer the question in the first 2 sentences.
- Be specific about who it is for and what job it does.
- Use something only you have, so your own product data or what your experts know.
- Show a real last-updated date.
"Picture a reporter on a deadline skimming for a quote. If your answer is buried in paragraph 4, they're going to grab someone else's." — Will Leatherman, Founder of Catalyst
The XYZ Pay page opened with "we're reimagining global payments for the modern workforce" and now opens with "XYZ Pay lets finance teams pay contractors in 30 countries from one account, with no local entity in any of them." Same facts, except now the answer sits in the first 2 lines and a buyer can tell in one read that the page is for them.
Specificity matters more than most teams assume. HubSpot found a buyer's question to AI is often a full paragraph loaded with context about their company, where a Google search is a few words. A page written for everyone matches nobody.
The proprietary data point is the edge most teams never use. A number from your product, a pattern your support team sees every week, a mistake your best customer stopped making. The cheapest way to get it is a 30 minute recorded call with one of your experts where you ask what buyers get wrong and what they would tell a friend, then build the page from what they say.
"A competitor can copy your keyword list, but they can't copy what your experts actually know." — Will Leatherman, Founder of Catalyst
Step 4. How do you get onto the pages AI already reads?
Treat it as a reference check, because AI is asking around about you before it recommends you. Muck Rack found more than 8 in 10 AI citations come from earned media, which is coverage you did not pay for, while paid placements barely register. HubSpot looked at millions of AI citations and found most of them were blog posts and lists, so the best tools for this and the top platforms for that. Getting onto those lists in your category is worth more than another press release.
Go back to the sources behind that first answer and sort them into 3 piles.
| Pile | What it is | The move |
|---|---|---|
| Sites to pitch | Review sites and roundups you should be listed on | Get added before their next update |
| Pages to beat | Competitor articles getting cited | Publish a better one on the same question |
| Communities to join | Forums and threads where buyers ask | Answer as yourself |
That is your outreach list, and it came straight from what AI already reads in your category. Three quick checks before you pitch any site. Does AI cite it in your category, does it get real readers, and does it sell links to anyone who pays. Skip the last kind, because AI and Google both learn to discount them.
Pair every placement you earn with a page on your own site saying the same thing, so AI finds 2 sources that agree. The longer version of this play is in how to get named in AI answers when 85% of your mentions sit on pages you do not own.
Step 5. Which 3 numbers tell you it is working?
Check these every week, and give each one an owner.
- Your AI visibility score across your 20 questions, each run more than once.
- AI referral traffic in your analytics.
- A "how did you hear about us" field on your forms and discovery calls, logged in your CRM.
The third one carries more weight than it looks. UserMaven studied B2B software sites and found most conversions from buyers who first arrived through an AI answer got credited to direct or organic traffic, because the buyer read about you in ChatGPT and then typed your name into Google. Ask them, or you will never know.
Weekly beats monthly because the answers move. HubSpot found about half the pages AI cites change every month. The working rhythm is one decision per page per month. Refresh it if it is slipping, with new facts and a new date. Reformat it if it ranks but never gets cited, so the answer sits at the top. Leave it alone if it is working. How to monitor your AI search ranking every week covers the tracking setup.
Ownership usually lands across 4 people. Whoever runs the website owns the technical fixes, content owns the pages, PR or partnerships owns the sites to pitch, and one person owns the weekly number and brings it to the marketing meeting.
"If no one owns it, it just doesn't happen." — Will Leatherman, Founder of Catalyst
What should you skip?
Plenty of popular AEO advice measured near zero, and a lot of the tools selling it want you to buy them.
- Schema markup, which is code labelling what is on a page. Google's own guidance says you do not need special schema, chunked pages, or everything rewritten for AI.
- llms.txt, a file some sites add listing their pages for AI. Ahrefs found almost none of those files get read.
Spend that hour writing one clear page instead.
How long before any of this shows up?
Weeks, not days, and far better than SEO's months. One study found most AI search bots do not even re-read your robots.txt in the first week. Make a change, then give it a month of weekly checks before you judge it.
The takeaway
Three moves get you further than a year of reading about this.
| When | Move | Time |
|---|---|---|
| Today | Run the fact check and mark what AI has wrong about you | 20 min |
| Tomorrow | Sit with sales and write down what buyers actually asked last month | 30 min |
| Tomorrow | Turn JavaScript off and reload your homepage | 1 min |
That is about 2 hours, and at the end of it you will know which use case to attack, which facts to correct, and whether AI can read your site at all. Start with the fact check today, then run the free AEO audit on your own company to see how often AI recommends you, which competitors it picks instead, and what to fix first.
The Content Engineer
Enjoyed this article?
Frameworks like this, weekly. No fluff, just original research and actionable insight.
More in AEO

Animalz vs Foundation vs an AEO Specialist, Three B2B Content Models Compared
Compare three B2B content marketing models on five criteria, retainer articles, campaign assets and citation work, and see which one your buyers need.

Who Offers Answer Engine Optimization Services in 2026, 9 Providers Compared on Scope and Published Proof
Compare answer engine optimization services on nine criteria, from tracked prompt counts to guarantee remedies, with Catalyst's answers on each.

Who Provides an AI Visibility Audit, 2026 Providers Ranked by Funding Stage
Learn how to choose an AI visibility audit by funding stage, from query counts and AI assistant coverage to published method and tracker ownership.
See where AI ranks you
The free AEO audit runs your buyers' questions through ChatGPT, Claude, Perplexity and Google AI Overviews, and shows where you are named and who is named instead.