Study says vendors shape three-quarters of AI product recommendations

A study by AI visibility consultant Maria Dykstra found that 30 of 40 pages behind AI product recommendations were published by companies selling in the same category. More than half of the pages promoted their publisher’s own products, raising a pointed question about how neutral an AI-generated shortlist can really be.

Watch Desk analysis

What happened

Dykstra examined 1,855 citations across 160 AI answers to buyer questions, then looked more closely at 40 of the source pages. Digital Journal reports that 30 were written by vendors in the market being discussed, and 22 ranked or included the publisher’s own products. The work is part of Dykstra’s ongoing Algorithmic Authority Index, which she says examines recommendations from ChatGPT, Perplexity and Google AI Overviews.

Those numbers describe the sample she studied, not every AI recommendation or search result. The article gives no detail here on how the 160 answers or 40 pages were selected, so the findings are a useful warning, not a market-wide census.

Why it matters

A buyer may see an AI-generated shortlist as an impartial comparison, while the sources informing it include competitors describing their own market. That creates a practical challenge for businesses trying to understand how they appear in AI answers, and for buyers deciding whether a recommendation deserves their trust.

Our read

The finding is worth attention precisely because it is specific, but it should not be mistaken for proof that AI recommendations are generally pay-to-play or systematically biased. The more useful next step is to ask which sources support a recommendation, who wrote them, and whether the answer distinguishes a vendor’s own claims from independent evidence. Even the machine’s shortlist has a bibliography, of sorts.

What to watch

• Whether Dykstra publishes the selection methods and results for a larger sample.
• Whether other researchers reproduce the vendor-page finding across different questions and AI systems.
• Whether AI products make the sources behind recommendations easier to inspect.

Discussion spark: If an AI recommends a product using comparisons written by vendors selling in that market, is clear citation enough to make the shortlist trustworthy, or should the system do more to separate self-promotion from independent evidence?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.