Trace the sources, then price the influence
ChatGPT does not have opinions about your category. It has sources. When it recommends products it leans on first-person, point-of-view content, reviews on Tom's Guide and TechRadar, Reddit threads where someone actually used the thing, because that is what reads as ground truth to a model trained to detect it. Anonymous roundups get consumed. Named experiences get cited.
So the working method is mechanical. Ask the buyer questions your customers ask, note exactly which sources the answers lean on, then assess the effort or cost of showing up in each one. Some are open doors, a review site that accepts products, a community you can legitimately contribute to. Some are priced, and some are closed. The point is that influencing AI answers is a source-acquisition problem with a cost column, not a mystery. It is the same discipline as LLM SEO with a purchase-intent filter. Run the exercise quarterly rather than once. The source mix shifts with model versions, and a source that carried the answer in one release can vanish from the next, which changes your acquisition priorities overnight.
What actually moves recommendations
Three things, in my experience running this at Superpower and for clients since. Genuine third-party reviews by named people, which are the citations ChatGPT reaches for first. Your own experts published where the model looks, because expert-attributed content earns a different class of trust than brand content. And live retrieval: when the model browses to ground an answer, an answer-shaped page on your site can be pulled in directly. I have watched a click-to-AI feature that RAGs a page lift how models summarise it, and the lesson generalises: pages structured as answers get retrieved as answers.
What does not move recommendations is what did not move rankings either: thin pages, keyword stuffing, and content with no author. The engines are pattern-matching on evidence of real experience, so the play is manufacturing genuine evidence, not simulating it. How the same logic runs inside Google is on the AI Overviews page, the measurement panel is on the GEO page, and the category-prompt playbook is in answer engine optimization. For what happens to those who ignore it, the traffic-impact numbers are the cautionary tale, and if you want the buyer-side system built, that is what I do for clients.
FAQ
How does ChatGPT choose which products to recommend?
From its sources: first-person reviews on sites like Tom's Guide and TechRadar, Reddit experiences, and expert-attributed content. Trace which sources appear for your buyer prompts, then work out the legitimate cost of showing up in each.
Can you pay to be recommended by ChatGPT?
Not directly. But the sources it cites each have their own economics: review programs, communities, contributor slots. Influencing AI answers is source acquisition with a cost column, not advertising.
Does my own website affect ChatGPT recommendations?
Yes, twice over. Expert-attributed pages feed training and retrieval, and when the model browses live to ground an answer, an answer-shaped page can be pulled in directly. Structure pages as answers and they get retrieved as answers.