How to appear in
AI search results.

Start with what the answers cite, not with advice

Most guidance on this subject is a list of sensible habits: use structured data, answer questions directly, build authority. None of it is wrong. All of it is written without looking at what the assistants actually pulled from.

In October 2026 we asked ChatGPT, Gemini and Perplexity five customer-phrased questions each about one industry in one city, in fresh private windows, signed out, on a single day. Fifteen answers. The full study and the verbatim run log are published.

The most useful thing in it was not the rankings. It was the citation list.

Every source the three assistants cited across fifteen recommendation answers in one industry.

Seven directories, several of which most people in that industry would struggle to name. Not the trade association. Not the big review platforms everyone obsesses over. Niche aggregators with thin traffic of their own, which the assistants nonetheless treat as a list of who exists.

Most of them are free to join. That is the practical point of this entire page.

Find the directories for your own industry

The seven above are specific to one trade in one city. Yours will be different, and you cannot guess them. You have to look.

Run five recommendation questions about your own trade and city across the three assistants, signed out, in fresh private windows, and record which sources each answer cites. Most assistants show them. The method takes about ten minutes and is written out step by step here.

What comes back is a list of the places that decide whether you are in the answer. Work through it. That list is worth more than any general advice on this page, including the rest of it.

What they are reading is not your Google ranking

Every source in that table is a directory. Not one of the fifteen answers was assembled from the local search results that businesses spend most of their effort on.

That does not mean local search is irrelevant, and we did not test whether it feeds these systems indirectly. It does mean that the thing you are optimising and the thing being read are not obviously the same object, and that a business can be doing well in one while being absent from the other.

It is worth sitting with how strange that is. A business can rank, be reviewed, be busy, and still not exist inside the answer a customer is given.

Being named is not the same as being described correctly

One business in the study was given a rating of 4.3 from 419 reviews by one assistant and 4.9 from 2,540 by another, in answers given on the same day.

Both figures are probably true. One cited Google and the other cited Yelp. We could not establish which platform each assistant was quoting in every case, so this is reported as platform inconsistency rather than error.

The practical consequence is the same either way. The number a customer sees attached to your name may come from a platform you stopped thinking about years ago. Finding out which one is part of the check, and it is why recording what it said about you matters as much as recording whether you appeared.

Six things to do, in the order worth doing them

1. Get listed where the answers already point. Start with the sources your own check surfaced, highest frequency first. Free listings first, paid ones only if the same source keeps appearing.
2. Make the facts about you identical everywhere. Name, address, phone, hours, the names of the people who work there, the services you offer. Assistants assemble a description of you from several sources at once, and disagreement between them is where errors enter.
3. Publish specifics that can be quoted. Prices, named methods, numbers, the names of your practitioners, what you do not do. Vague positioning statements cannot be lifted into an answer. Concrete facts can.
4. Fill in your Google Business Profile completely. Categories, services with descriptions, attributes, hours, photos. It is the most structured description of your business that exists, and it is free.
5. Earn reviews on the platforms that showed up in your check, not on every platform. Which brings us to a finding that surprised us.
6. Re-run the check in sixty days and compare. This is the only way to know whether anything you did mattered.

A very small number of reviews can be enough

Six businesses appeared only in Perplexity's answers and were never named by the other two assistants in any run. All were small, and all were highly rated on very few reviews.

One was recommended for "top rated" in its city on four reviews. Another appeared in a treatment recommendation on two.

That cuts both ways. If you are small, you are not automatically excluded, and the usual advice that you need hundreds of reviews before anything happens is not what we observed. If you are established, a competitor with four reviews can appear beside you in the same answer.

Consistency matters more than it sounds

In one answer, an assistant named the wrong surgeon at a real practice. We verified the correct name against three independent sources. The same assistant had given the right name three questions earlier in the same session.

Also logged: a thumbnail captioned for a different business entirely, and a sentence that broke off mid-name.

Nobody is monitoring this on your behalf. If the facts about you are inconsistent across the places an assistant reads, you are making this more likely, and the first you will hear of it is from a customer who arrives asking for someone who does not work there.

There is no single answer to optimise for

Across fifteen answers, twenty four businesses were named. Two were named by all three assistants. Sixty two percent were named by one assistant and were absent from the other two. Six different businesses came first depending on which one was asked.

One clinic appeared in four of five ChatGPT answers and in none of Gemini's. Another was the exact reverse.

So "appearing in AI search results" is not one outcome. It is at least three, they disagree with each other, and checking only ChatGPT will give you a confident answer about a third of the picture.

What this does not show

This is a snapshot of one industry in one city on one day, and it describes what was cited, not what caused the citation. We did not establish that being listed in those seven directories is why those businesses appeared. We established that those are the sources the answers drew on.

That distinction matters, because the honest version of this advice is: go where the answers are already looking, then measure whether it changed anything. Not: do these six things and you will be recommended.

Assistants also change. Anything measured in October 2026 may read differently in six months, which is the argument for re-running the check rather than trusting a result, including this one.