ChatGPT chooses which brands to recommend by searching the web, reading the pages that come back, and naming the brands those pages name. For most commercial questions (“best accounting software for a small law firm”, “reliable HDPE supplier in India”, “which running shoes last longest”) ChatGPT runs a live search, pulls in a few dozen pages, and writes an answer grounded in them. The brands that appear across those pages, especially in independent comparisons, reviews and community threads, are the brands that get recommended. Your own website contributes, but on ChatGPT it is a minority of what gets cited.
That is the whole mechanism. There is no list of approved brands, no advertising slot, and no way to pay OpenAI for a recommendation. What ChatGPT says about your category is a reflection of what the rest of the web says about it. This post explains how the retrieval works, which sources ChatGPT leans on, what the correlation studies show, and what to do about it.
ChatGPT recommends the brands that third-party pages already recommend, so the fastest way to change its answer is to change what the independent pages it retrieves say about you.
Does ChatGPT actually search the web before recommending a brand?
Yes, for most questions where the answer depends on current information. OpenAI’s own documentation for ChatGPT search describes the process: when a prompt needs web information, ChatGPT searches, reads the results, and writes a response with inline citations the user can open. OpenAI says it ranks results on multiple factors aimed at relevance and reliability, and that placement is not guaranteed.
Two details in that documentation matter for brands.
First, ChatGPT search uses OpenAI’s own index, crawled by a bot called OAI-SearchBot, alongside third-party search providers. To be eligible for search answers, a site has to allow OAI-SearchBot in its robots.txt and let traffic through from OpenAI’s published searchbot IP ranges. Many sites blocked every OpenAI bot in 2023 and 2024 to keep their content out of training data. If yours did, you may have opted out of being recommended. GPTBot is the training crawler; OAI-SearchBot is the one you need to allow.
Second, ChatGPT consults more pages than it cites. OpenAI’s developer documentation for its web search tool notes that the full list of URLs the model consulted is usually longer than the list of inline citations. A page can shape the answer without appearing as a visible source. This is one reason brand presence across many pages matters more than a single prominent link.
When ChatGPT does not search (a general-knowledge question, or a user with search turned off), it answers from training data. Those answers reflect what the web said about your brand when the model was trained, which is again mostly third-party text. Either way, the input is what other people have written about you.
Which websites does ChatGPT cite most often?
Independent studies agree on the shape, if not the exact numbers.
Semrush tracked around 230,000 prompts across ChatGPT, Google AI Mode and Perplexity from July 14 to October 12, 2025, taking weekly snapshots of the 25 most-cited domains. On ChatGPT, Reddit and Wikipedia were the two most-cited domains across the period. The share swung sharply: Reddit appeared in close to 60 percent of ChatGPT responses in early August and around 10 percent by mid-September, a drop widely attributed to Google removing the parameter that let tools fetch 100 results at a time, though Semrush’s own analysts doubted that was the whole cause. The rest of the top 25 was dominated by publishers, communities and reference sites rather than brand homepages.
Profound’s 2026 analysis of 11.8 billion AI citation data points, summarized by WARC, found that brand websites account for 57 percent of AI citations globally but only about 47 percent on ChatGPT (and about 69 percent on Gemini). So on ChatGPT, slightly over half of all citations go to pages the brand does not control: media, communities, institutions and review sites.
Here is what that mix looks like in practice, and what a brand can do about each part of it.
| Source type | Examples | Why ChatGPT retrieves it | What you can do |
|---|---|---|---|
| Community threads | Reddit, Quora, niche forums | First-hand opinions and direct comparisons, often recent | Earn genuine mentions in relevant threads; answer questions honestly |
| Reference sites | Wikipedia, industry wikis, standards bodies | Neutral, structured, heavily linked | Make sure facts about your company are accurate and sourced |
| Publishers and trade press | National business media, sector magazines | “Best of” lists, reviews, expert quotes | Pitch data, commentary and case studies; get into listicles |
| Review and comparison sites | G2, Capterra, Trustpilot, consumer test sites | Structured ratings, pros and cons | Claim profiles, gather reviews, keep listings current |
| Brand websites | Your own pages | Specs, pricing, documentation | Clear, factual pages that answer questions directly |
Do backlinks or brand mentions matter more to ChatGPT?
Mentions, on the best evidence available, though the two travel together.
Ahrefs studied 75,000 brands to see which signals predicted a brand appearing in Google’s AI Overviews. Branded web mentions (the number of pages that name the brand, linked or not) had the strongest correlation at 0.664. Backlink count correlated at 0.218. The study covered Google’s AI, not ChatGPT, but the logic transfers: a language model learns what a brand is and what it is good for from the text that mentions it, and a bare hyperlink carries no words.
This does not make links irrelevant. A link helps the engine connect a mention to the right website (there are many companies called Apex), it helps your pages rank in the conventional search that ChatGPT’s retrieval leans on, and a linked mention in a reputable article is almost always a mention in context, which is the thing that matters. The practical rule: pursue articles that describe your brand, and take the dofollow link (one that passes authority to your site) when you can get it. Do not pursue links in articles that say nothing about you.
In engagements we track, brands we have placed in third-party articles, guest posts and listicles have seen AI citations rise by roughly 70 to 80 percent on the prompts where those articles are cited. The rise comes from the article being retrieved and read, not from the link inside it.
Why does ChatGPT recommend a competitor instead of you?
Almost always because the competitor is in the pages ChatGPT retrieves and you are not. Here is how to see it for yourself in about twenty minutes.
- Write down five to ten questions a buyer would ask before choosing you. Keep them specific: “best payroll provider for a 40-person charity” rather than “payroll software”.
- Ask each one in ChatGPT with search on. Note which brands are named and in what order.
- Open every cited source. Count how many name your competitor and how many name you.
- Run the same prompts again on two more days. Answers vary; you are looking for the pattern, not one result.
The typical finding: the competitor is in two or three listicles, one trade publication piece and a Reddit thread where someone asked the exact question. You are on your own website and perhaps one directory. No amount of homepage copy will fix that, because the homepage is not what ChatGPT is reading. We cover the listicle side in detail in why “best of” listicles drive AI citations, and the measurement side in how to measure your brand’s AI visibility.
What can you do this quarter to get recommended?
Four things, in order of leverage.
Get into the comparison pages that already exist. Search for your category plus “best”, “top”, “vs” and “alternatives”. The articles that rank are the ones ChatGPT retrieves. Contact the authors with a reason to include you: a data point, a customer result, a feature the list is missing. This is the highest-return activity for most brands, and it works for a polymer manufacturer or an accounting firm as readily as for a software company.
Be present where buyers already talk. Find the Reddit threads and forum discussions that come up for your prompts. Participate honestly, disclose who you are where the rules require it, and give useful answers. ChatGPT’s retrieval treats a helpful thread as evidence. Our Reddit marketing guide covers how to do this without being removed.
Earn coverage with something worth writing about. Original data, a survey, a clear opinion from your founder. Press and trade coverage puts your brand in the publications ChatGPT cites, in a context (an expert quote, a ranked list) that the model treats as a recommendation. Our digital PR guide explains what journalists actually pick up.
Make your own pages quotable. Allow OAI-SearchBot in robots.txt. Put a direct answer in the first paragraph of each product or service page. Publish pricing, specifications and comparison pages that state facts plainly. This is the on-site half of AEO (answer engine optimization) and GEO (generative engine optimization), covered in AEO and GEO explained. It will not get you recommended on its own, but it determines what ChatGPT says about you once a third-party page has brought your name in.
For the full argument, with the citation data across every major engine, read why third-party articles decide which brands AI recommends.
How Model Marketing helps
We do the off-site work described above. Our link building service places your brand in guest posts and listicles on vetted sites: DR 50+, 10,000+ monthly organic visits, real editors, no “write for us” page, and a check on how often the site is already cited by ChatGPT, Perplexity, Gemini, Claude and Google AI Mode. Digital PR earns press and trade coverage through data-led campaigns. Reddit and community marketing puts honest, useful answers in the threads your buyers read, billed only after a post has stayed live for seven days.
On the on-site side, our AEO/GEO service starts with a two-week audit of your current citations and the prompts that matter, then makes your pages the ones ChatGPT quotes when your name comes up. Clients include Uniqode, SM Polymers Private Limited and Sim Films, across software, manufacturing and media.
Everything is billed per live unit, twice a month, with no retainer and 30 days’ notice. Guest posts start at $250, Reddit posts at $110, and a data-led PR campaign is $8,000. Full details are on the pricing page.