What 25-Question Test Runs Keep Showing Us About Who Gets Recommended (2026)

Every check we run works the same way. Twenty-five questions a real buyer would type, asked across several assistants, with every business name that comes back written down and dated.

Do that enough times across different niches and the same handful of patterns keeps turning up. Not opinions about how AI works. Just things that happen over and over when you ask the questions and read what comes back.

This post is those patterns. Everything below is described in illustrative terms drawn from our own runs. None of it is measured client data, and we will not publish that until clients have given permission for their numbers.

What do the test runs keep showing?

Five things, and the first one governs the rest. Answers to hiring questions converge on a short list of about three businesses. The same names refill that list across repeat runs. The winners are usually described on pages they do not own. Rewording a question moves the names more than changing assistant does. And the best website in a niche is regularly not on the list at all.

None of this is surprising once you have watched an answer being built. An assistant is not ranking a market. It is writing a short, confident paragraph that has room for a few names, drawn from whatever the web currently says about the category, in the shape Google describes for AI features in Search.

The practical value of running it yourself is not the insight. It is the list. You get a written record of who occupies the three slots in your market, which is a more useful competitor analysis than anything you would have assembled by hand.

Answers name about three businesses, and the list barely moves

Ask the same question five times over a few weeks and most of the names come back. Some churn happens at the edges, and there is usually one slot that rotates. The core does not.

FIVE RUNS OF THE SAME QUESTION SET, SAME NICHEILLUSTRATIVE

Run 1

Run 2

Run 3

Run 4

Run 5

The lit block is the group of businesses named in every run. It is the same width each time. The muted block is the churn: names that appeared once and did not return.

AnswerHalo answerhalo.com

CLICK TO ENLARGE

Written out, because nothing here may live only inside a figure. Across repeat runs of the same questions in one niche, a stable group of businesses is named every time, and a shifting group is named once and not again. The stable group is roughly the same size in every run. The shifting group varies, which is why a single run tells you very little and five runs tell you almost everything. Those proportions are illustrative.

What that means for a coach outside the stable group is worth saying plainly. You are not competing for a position on a long list. You are competing for one of about three slots, most of which are occupied by the same names every time somebody asks.

The named business is usually described somewhere else

When we trace back where an answer came from, the sources are mostly not the websites of the businesses being named. They are roundups, association listings, local press, review pages, and podcast episode pages that describe those businesses.

That inversion catches people out. A coach with an excellent website assumes the website is the asset. In these runs it is closer to a reference: something an assistant confirms details against once it has already decided to mention you. The thing that gets you mentioned is somebody else writing about you, which is the same finding at the center of what kinds of pages assistants quote most.

Perplexity, which shows its sources more openly than most, makes this easy to verify on your own market. Its help center describes searching the web and answering with the sources attached, so you can ask a category question and read the source list directly. Do it for your niche and count how many of the sources belong to a business that got named. In our runs it is usually a minority.

Five patterns, what causes each, and the move

Here is the whole set in one place, with what appears to cause each and the thing to do about it.

Row four is the one that saves the most time. People spend weeks deciding whether to optimize for ChatGPT or Perplexity when the bigger swing comes from asking the question the way a buyer actually types it, which is the craft covered in the questions to ask about your own business.

What to do if you are not on the list

Run the test properly once before doing anything else. Twenty-five questions, fixed wording, several assistants, every name written down. That takes an afternoon and it replaces guessing with a list of the businesses currently occupying your market's three slots. The method is laid out step by step in how to check if AI recommends you.

Then work backwards from the winners rather than forwards from your own site. Find the pages that keep supplying those names, and get yourself described on them: the association listing, the local roundup, the podcast with a written episode page. That is slower than publishing a blog post and it is the thing that moves the list.

Give the specific questions your attention first. Broad category questions in most niches are held by national brands and directories, and an independent is not going to displace them this year. Narrow questions about a particular situation, in a particular city, are frequently won by whoever has one clear page and one third-party mention.

And write in a way that can be quoted. Princeton researchers testing generative engine optimization found that citing sources and quoting authority raised how often a page was used in generated answers. That is a writing instruction, and it matches what we see: the businesses that get named tend to be described in sentences somebody could lift without rewriting.

One honest caveat about all of this. These are patterns from our own runs, not laws. Niches differ, the assistants change month to month, and a pattern that holds in ten markets can fail in yours. The only run that describes your business is the one done on your questions.

Our free check runs twenty-five buyer questions from your niche at $0 and emails a plain-English readout within the hour, including who currently holds the slots. It runs as soon as you ask for it.