How businesses get recommended by ChatGPT: the AEO playbook

Your buyers have quietly changed how they shortlist. Instead of typing a search and scrolling ten blue links, plenty of them now ask an assistant: "who's good at X in Y city?" They get three or four names and a sentence about each. That's the whole shortlist. If you're not in it, you're not in the conversation, and you'll never see a bounce or a missed click to tell you.

The discipline of getting named there is called AEO, answer engine optimization. It overlaps with SEO but it isn't the same job, and the difference matters.

How does an assistant decide who to name?

Simplifying a lot, but usefully: the model has a broad memory of the public web from training, and for anything current it goes and reads live sources when you ask. Then it composes an answer from what it read, favouring things it can state confidently.

Confidence is the key word. A model is more comfortable naming you when several independent sources agree that you exist, do the thing, and operate where you say you do. That's a very different bar from ranking first for a keyword.

Search rewards the best page about you. AI answers reward the clearest consensus about you.

So the work isn't optimising one page. It's tidying the story the whole web tells about your business until it's boringly consistent, and then getting other people to repeat it.

Get mentioned by people who aren't you

This is the largest lever and the one most businesses skip because it isn't in their control panel. Your own website is treated as marketing, which is fair, because it is. Third party pages are treated as evidence.

What counts as a third party mention:

• Directories and marketplaces in your category, filled in properly rather than half abandoned.
• Local and regional press, including small publications. Being named in a city roundup is worth more than a beautiful page you wrote about yourself.
• "Best X in Y" listicles from anyone who publishes them. Ask. Most people who make these lists accept submissions and nobody sends them.
• Partner and supplier pages. If you work with venues, studios, suppliers or associations, get your name and link on their site.
• Forums and communities where real people answer real questions, if you show up as a useful participant rather than an advert.

You don't need hundreds. You need enough independent pages saying the same true thing that a model can repeat it without hedging.

Say the same thing everywhere

Now the unglamorous part. Go and look at how your business appears across your website, Google Business Profile, Instagram bio, directory listings, old profiles you forgot about. Most businesses find three versions of their own name, two phone numbers, an address from a previous office, and service descriptions that don't match.

Every inconsistency is a small reason for a model to hedge or to name someone tidier instead. Fix these until they're identical:

• Business name, exactly, including whether you use "and" or "&".
• Category, in the plain words a customer would use, not a clever internal label.
• City and service area, stated in words on the page and not only implied by a map pin.
• Phone, email, and the URL you want treated as home.
• What you do and don't do, in one consistent sentence.

This is a boring afternoon that pays for years. It's also the step that makes everything else work, because the mentions you earn later need something consistent to point at.

Publish things that are easy to quote

Models quote text that answers a question directly. That's it. Not text that's beautiful, not text that's clever, text that's liftable.

The practical rules:

• Use the question as the heading. "How far ahead should you book?" gets used. "Timelines and process" doesn't.
• Answer in the first two sentences. Put the payoff up top, then explain underneath. Models take the top, and so do impatient humans.
• Be specific and be willing to say numbers. Ranges, durations, capacities, what's included. Vagueness reads as nothing to quote.
• Cover the questions buyers actually ask you. Your inbox is the keyword research. The awkward questions about cost and process are the ones that get asked to assistants too.
• Write once, keep it current. An out of date page is worse than none, because contradiction is what makes a model hedge.

If you want the fuller structural version of this with page level detail, we've written it up in what AEO means and how to do it.

Keep reviews fresh, specific and answered

Reviews do double duty. They're third party text about you, and they're recent, which matters because assistants prefer sources that look current.

Three habits do most of the work. Ask every happy customer, right after the job, when the feeling is real. Nudge people gently towards specifics, because "great service" is unquotable while "they redesigned the layout two days before, in the rain" is a sentence a model can use. And reply to every review, including the bad ones, in a way that reads like a person who cares rather than a template.

A steady trickle of detailed reviews beats a pile from two years ago. Recency is part of the signal.

How do you measure any of this?

You can't see AI answer traffic in your analytics the way you see search. So you test directly:

1. Write down ten questions a real buyer would ask, in their words, with your city in them.
2. Ask each one in ChatGPT, Perplexity and Google's AI answers, in a fresh session with no personalisation.
3. Record whether you're named, in what position, and who else appears.
4. Repeat monthly. Same questions, same day of the month.
5. Track the trend across months, never the reading from one day.

The trend is the only number worth caring about. Also watch your competitors in the results, because the fastest lesson available is noticing which of their pages keeps getting quoted.

This is one of the loops we built into Kavi, because it's slow, repetitive and easy to abandon, which is the exact profile of work software should carry. Panigrahana, the company behind Kavi, runs it on its own marketing across 16 channels, and the measurement pass is the part we'd never go back to doing by hand.

The honest caveats

Anyone who guarantees you a spot in an AI answer is bluffing. Nobody outside these companies controls what the models say, the answers vary between users and sessions, and the share of results any business holds moves week to week without warning. Models get updated. Sources get reweighted. A good month can quietly become an average one.

What you control is the evidence. Consistent facts, real third party mentions, quotable answers, fresh reviews. That work also happens to improve your search results, your conversion rate and your credibility with humans, which is the reassuring part: even if the assistants change their minds next quarter, none of it is wasted.