Why does ChatGPT recommend some brands and not others?

What ChatGPT is actually doing when it recommends#

ChatGPT is not ranking brands from a private scoreboard; it is summarizing the pages it just read. Prefer shows you which pages those are for your own buyer questions. When you ask for the best option in a category, it searches the web, pulls a small set of pages, and names the brands those pages name. OpenAI’s own documentation describes this: the model searches, reads, and cites its sources.

That reframes the whole question. The reason a competitor gets recommended and you do not is usually not that the model judged you worse. It is that the competitor showed up in the sources it read for that query, and you did not. Recommendation follows presence in the cited set.

The sources that decide it#

A few recurring sources carry most of the weight, so the brands they mention have an outsized advantage. Two patterns from our measurements make this concrete:

• A few third-party sources do heavy lifting. In our September 2026 ChatGPT run, the techradar.com domain was cited on 12 of 47 answers, across its tested-tools roundup and its per-tool reviews. Being named on a source like that touches a large share of related questions at once.
• Community threads can be the top source. In our August 2026 study, Reddit was the most-cited source on ChatGPT for every one of 25 B2B buyer queries, averaging about 30% of the sources per answer. What people say about you in those threads becomes what ChatGPT repeats.

Because grounded answers cited only about 3.8 sources each, the set is small and the competition to be in it is real. A brand missing from the listicles, review sites, and threads for a query is effectively invisible to that answer.

How to become a brand it recommends#

Work on being present and well-regarded in the sources ChatGPT reads, not on the model itself. The levers are concrete, and most map to sources we saw cited in our run:

• Get into the tested roundups and category reviews. TechRadar-class hands-on articles that name you get repeated across many related answers; the techradar.com domain alone was cited in 12 of 47 answers, through five of its pages. Editorial coverage that names you is the highest-leverage surface.
• Complete your review-site and community presence. G2 category pages were cited for alternatives and pricing questions, and Reddit was the top source type in our study, so being well-represented on the relevant subreddits and review grids is what gets quoted. It cannot be faked at scale.
• Give the engine an answer-first passage it can lift. A crawlable page that answers the exact question in its first sentence hands the model a clean, self-contained fact to cite when your brand comes up.
• Fix the pages you already appear on. A directory or review entry that names you but describes you wrong or out of date will feed that version into answers. Correcting existing mentions is often faster than earning new ones.

One caution: because a small number of sources carry so much weight, do not confuse the tactic with gaming it. Seeded self-ranking listicles and fake reviews get discounted once they are spotted, and community platforms remove inauthentic posts. The durable version is real editorial coverage and genuine participation, which is exactly why it compounds. In our data, the editorial and community sources that recur most (the techradar.com domain on 12 of 47 answers, Reddit as the top source on all 25 software buyer queries in our August study) are the ones that earned their place, not the ones that manufactured it.

The pattern is consistent: win the sources, and the recommendation follows. See which sources ChatGPT cites for software buyers in our study, and run a free AI visibility audit to find which sources are naming your competitors instead of you.

Sources

1. In Prefer's 2026-09-02 ChatGPT run, grounded answers cited about 3.8 sources each, and the techradar.com domain was cited on 12 of 47 answers, showing recommendations track a few recurring sources. (Prefer AEO loop run, 2026-09-02, dated research note)
2. In Prefer's August 2026 source study, Reddit was the single most-cited source on ChatGPT across all 25 B2B software buyer queries, averaging 30.4% of the source mentions per query. (Prefer Reddit source study, 2026-08-19, dated research note)
3. OpenAI's Help Center explains that ChatGPT searches the web and cites the sources it reads, which is why the brands named in those sources are the brands the answer repeats. OpenAI Help Center: Searching the web with ChatGPT
4. Prefer, our product, runs your buyer prompts on ChatGPT, Gemini, Perplexity, Google AI Overviews and Google AI Mode, shows which domains each answer cites and which competitors it names, and includes AI articles on every plan; Prefer Managed adds off-page work such as Reddit participation and PR pitches. Prefer pricing (our product)