Why ChatGPT Recommends Your Competitor (And How to Change It)
A prospect asks ChatGPT which tool they should use for your exact category. It names three companies. You are not one of them. The same thing happens in Gemini, in Perplexity, and in Google's AI Mode, and by the time that buyer reaches your website they have already built a shortlist that you are not on.
This is not a ranking problem you can fix on your own site, and it is not something you can buy your way out of. Here is what is actually happening and what changes it.
Why the model picks someone else
A model answering "best X for Y" is not consulting a ranked index of vendors. It is producing the most probable answer given everything it has read about your category, and for live answers it is also retrieving a handful of pages at the moment you ask. Both paths run through content other people wrote about you.
That retrieval pool is far narrower than the open web. Semrush analysed roughly 230,000 prompts and over 100 million citations across ChatGPT search, Google AI Mode and Perplexity in weekly snapshots between July and October 2025. A small set of domains supplied a large share of the answers: Reddit and Wikipedia were the two most-cited domains on ChatGPT, while Google AI Mode leaned on LinkedIn, YouTube and Reddit, and Perplexity drew heavily on Reddit and LinkedIn. A separate analysis of 30 million sources by Peec AI, reported by Search Engine Land in March 2026, put Reddit first overall, followed by YouTube and LinkedIn, with Perplexity favouring Reddit, LinkedIn and G2 for B2B queries.
Read that back as a marketing problem. If your category conversation on Reddit, Quora and LinkedIn does not mention you, the model has very little to work with when it builds a shortlist. Your own website is rarely the deciding input, because a vendor describing itself is exactly the source these systems weight least.
Being talked about is necessary, but it is not sufficient
Seer Interactive ran a useful study during the 2026 Winter Olympics, measuring 231,347 responses across seven AI platforms over 52 days. It found three signals working in sequence. Entity authority came first, and Wikipedia was the single strongest predictor of visibility in the data. Third-party validation came second: independent publications and analysts explaining why a brand matters. Community discussion came third, and Reddit was cited across platforms at three to fifteen times the rate of Instagram, TikTok or X. Brands with all three signals present averaged around 12,174 mentions against roughly 1,565 for brands with an incomplete set.
The important finding is the order. The study concluded that community discussion amplifies authority that already exists but cannot manufacture visibility on its own. Flooding Reddit while nobody credible has ever described what you do is not a strategy, it is noise. That is why we treat community work, published articles and search content as one program rather than three services.
How we change the answer
The method is deliberately narrow, because trying to influence every possible prompt at once produces nothing measurable.
- You pick one buyer prompt. Something a real prospect would type, in the form "best X for Y". One prompt, not a keyword list.
- We measure where you stand today on that prompt across ChatGPT, Claude, Gemini and Perplexity, including which competitors appear and how you are described when you do.
- We map the sources those answers actually pull from. Not the sources we assume. The specific threads, answers, articles, comparison pages and review listings behind the response.
- We work those sources. Reddit and Quora discussions where the question genuinely comes up, LinkedIn and Medium articles published under a named person, and the comparison and review pages that carry weight in your category.
- We re-measure monthly and keep defending the position, because the answer moves whether or not you touch it.
Most published advice on this points at YouTube and Wikipedia. Both are heavily cited, and both are extremely hard to influence deliberately and slow to move when you can. Reddit, Quora, LinkedIn and Medium are cited constantly, index within hours, and are genuinely open to a credible contributor, which is why we concentrate the effort there.
What this cannot do
Nobody can force a model to recommend a brand, and any agency promising that is guessing. Citation patterns are also unstable: in the Semrush data, Reddit and Wikipedia's share of ChatGPT citations fell sharply after a change in September 2025, from roughly 60 and 55 percent to somewhere around 10 to 20 percent, while Forbes and PR Newswire gained. Models update on their own schedule.
So this is maintenance, not a fix you buy once. Expect movement over months, expect some months to go backwards before they go forwards, and judge the work on your share of the answers you chose to compete for rather than on any single screenshot.