Original research

Why AI recommends your competitors: five patterns from recorded answers.

AI assistants don't rank a category. They match each buyer's situation to the firm with the clearest public evidence for it. Across our studies, five patterns decided who got named.

Based on
Recorded answers from Riseklix studies
Tools
ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews
Best for
Service firms

The short version

  • AI rates firms for jobs, not in general. One firm was recommended 27 of 32 times for one need and 4 of 32 for another.
  • Being known isn't being recommended. All four tools knew one agency by name. It was named in 0 of 24 buyer answers.
  • Specialists win situations. In three buying situations, one specialist was named in 15 of 16 answers.
  • Each tool reads a different internet. Perplexity used the firm's own website 88% of the time. Gemini: 11%.
  • Missing facts get said out loud. "I couldn't verify their current terms or pricing, so ask them directly."

How we collected the answers

For each firm we wrote the questions a buyer would actually type, with no company names in them. Each buying situation was asked several ways across the AI tools buyers use. We saved every answer and every source it showed.

The figures below come from three of our studies: a UK recruitment firm, a US training firm and a US commercial contractor. The first two were recorded on ChatGPT, Gemini, Claude and Perplexity. The third also covered Google AI Overviews.

1. AI rates firms for jobs, not in general

The same recruitment firm, with the same website, got very different results depending on the hiring situation. For an everyday hire it was recommended in 27 of 32 answers. For a senior US search it was recommended in 4 of 32.

The everyday work was well documented in public: job adverts, team biographies, a clear service page. The senior work wasn't. So "does AI recommend us?" is the wrong question. The useful one is "for which job?"

Most owners test AI by typing their own company name. That check passes easily. In one test, all four tools knew an agency when asked by name. When we asked the questions its buyers ask, which describe a problem and not a company, it was named in 0 of 24 answers.

Three of the four tools also described the business it used to be. They were reading old directory profiles and podcast biographies, not the current website.

3. Specialists win situations, famous names don't

In the training study, three different buying situations were each dominated by one specialist, named in 15 of 16 answers. McKinsey, BCG and Bain each appeared 15 to 17 times across all 160 answers, and none of them led a single situation.

When we read the pages cited for two of those winners, they worked like decision guides: who it's for, which route fits, what's required, and what support comes after.

4. Each tool reads a different internet

We traced the sources behind 365 recommendations of 16 leading firms. This is how often each tool showed the firm's own website as a source:

AI toolRecommendations tracedOwn website shown as a source
Perplexity6658 (88%)
ChatGPT6122 (36%)
Claude12628 (22%)
Gemini11212 (11%)

Across all four tools, about a third of recommendations drew on the firm's own site (120), a third leaned only on other sources (123), and a third showed no source at all (122). Your website is one voice among three.

5. Missing facts get said out loud

"I couldn't verify their current terms or pricing, so ask them directly."

That was Claude, about a firm it had just recommended. The other firms in the same answer got a course fee, a per-person price and a format. "Ask them directly" sounds harmless, but it turns the recommended firm into the only one that needs an extra step.

In the same study, a firm with dozens of well-written blog articles was cited in 0 of 160 answers. The pages AI used answered what a buyer asks before a call: cost, time, and what you leave with.

One check that comes before all five

In a study of a US commercial contractor, AI recommended the company in 15 of 50 answers, always under its parent company's name. It never used the name of the company's newer commercial brand, and never cited that brand's website. When we checked the website, all nine pages we sampled were set to stay out of Google's index.

The five patterns are about evidence. This check comes first, because evidence on a page that can't be indexed doesn't count. That study covered five tools, and Google AI Overviews named the company most often, in 7 of 10 answers. Read case file 003.

What to change on your side

  1. Check your pages can be indexed. Look for a noindex tag on your key pages, or use URL Inspection in Google Search Console.
  2. List your buying situations. Write the ten situations in which a buyer needs you, in their words.
  3. Build one page per situation. Say who it's for, how it works, how long it takes, a typical price range and what the buyer gets.
  4. Fix the old record. Update directory profiles, speaker biographies and partner listings that still describe who you were.
  5. Earn proof on other sites. Client pages, credible lists, podcasts with transcripts and reviews all feed the answers.
  6. Measure by situation and by tool. Ask each question several ways and more than once. Single answers vary.

Questions people ask

Is this the same as ranking in Google?

No. Google ranks pages for a query. AI assistants write an answer and name a few firms. The same firm can rank well in Google and still be left out of the answer, or the other way round.

How many answers do you need before you can trust a pattern?

One answer tells you very little, because answers vary. We ask each buying situation in four phrasings across every tool, so a pattern has to hold across many answers before we report it.

Do these findings apply to every industry?

They come from studies of service firms in recruitment, training and commercial construction. The mechanisms are general, but the numbers are specific to those studies. Your own questions and competitors will give different figures.

Want this checked for your firm?We ask five AI tools the questions your buyers ask, and trace every source.

Open your case

Next step

See why AI recommends your competitors.

A discovery study covers 10 buyer situations across ChatGPT, Gemini, Claude, Perplexity and Google AI Overviews. Fixed scope, agreed before we start.