Brand Mentions in ChatGPT: How to Earn Yours

ChatGPT mentions brands through two routes: what its models learned during training, and what its web search finds when a buyer asks a question in your category. You earn mentions by keeping your entity consistent everywhere, showing up in the third-party sources ChatGPT actually cites (comparison lists, communities, press), and publishing pages that answer questions in their first sentences. And you track them with a fixed set of prompts run monthly, because ChatGPT’s citation patterns shift faster than most people assume.

A brand mention in ChatGPT is not one thing. It can be a recommendation (“for invoicing software, consider X”), a neutral reference inside an explanation, or a cited source with a link. Each has different value, and each is earned differently. This post covers where mentions come from, what the data says about the sources behind them, and how to build and measure yours.

Mention, citation, link: three different assets

A mention is your brand named in the answer. A citation is a source ChatGPT links to support the answer. A link is a click opportunity. You want all three, but the mention is the one that shapes buying decisions, because most ChatGPT answers are consumed without a single click.

Most answers never touch the web at all. In Semrush’s analysis of 17 months of U.S. clickstream data, more than a billion rows of it, only about a third of ChatGPT queries (34.5% as of February 2026) used web search. The rest were answered straight from the model. If your brand only exists where a live search can find it, you are absent from the majority of answers in your category.

Where ChatGPT’s citations actually come from

When ChatGPT does search the web, it leans on a recognizable set of domains. Semrush’s study of the most-cited domains in AI, which tracked more than 230,000 prompts and over 100 million citations across ChatGPT, Google AI Mode and Perplexity between July and October 2025, found Reddit, Wikipedia, LinkedIn and YouTube dominating the citation tables.

The more useful finding is how unstable that table is. In September 2025, Reddit’s share of ChatGPT responses dropped from close to 60% to around 10% in the study’s data, and Wikipedia fell from roughly 55% to under 20%. Meanwhile Forbes doubled its cited responses, and press-release and editorial domains like PRNewswire and Medium gained ground. The sources reshuffled within weeks.

The lesson for a brand is direct: do not build your AI visibility on a single channel. A footprint spread across communities, editorial mentions, comparison pages and your own quotable content survives a reshuffle. A footprint that is only Reddit karma, or only your blog, may not.

What a mention is worth in traffic

Referral traffic from ChatGPT is real and growing, but concentrated. Semrush’s ChatGPT traffic analysis measured outbound referral traffic growing 206% year over year into January 2026, with the number of domains receiving any referrals rising from about 71,000 to about 170,000. It also found that the top 10 domains capture around 30% of all that traffic.

Read those numbers honestly: the pie is growing fast, a big slice goes to a handful of large sites, and your realistic prize as a brand is not raw traffic. It is being named in the consideration set when a buyer asks “who should I look at for X”. That is a positioning asset, and it compounds, because the same footprint moves Perplexity, Gemini and Google AI Overviews too.

What earns a mention

Three things, all covered in depth in our guide on how to rank in ChatGPT:

A consistent entity. Same name, same one-line description, same key facts on your site, your schema, your llms.txt, your LinkedIn and your directory profiles. Models mention brands they can identify without ambiguity.

Presence in cited sources. Run your own buying prompts, note which comparison posts, review platforms and threads ChatGPT cites, and earn your way into them with something concrete: data, an honest quote, a differentiated angle. Given the volatility above, spread this across source types on purpose.

Quotable pages. Pages that answer a buyer question completely in their first two or three sentences get lifted into answers. Brochures get skipped. Our page-level checklist for ChatGPT covers the mechanics.

How to track your mentions

One screenshot proves nothing. The measurement that supports decisions is a fixed set of 20 to 40 buying prompts in your category, run monthly against the same engines, scored on whether your brand is recommended, mentioned or absent, with the cited sources logged each run. The locked set is what makes months comparable, and the source log tells you where to earn presence next.

Every tool counts mentions, citations and share of voice its own way; our GEO glossary sets out the definitions and their sources.

You can run a light version of this in a spreadsheet. If you want a first read without the setup, our free AI Visibility Diagnosis runs a starter set of buying prompts on all five engines. And if you want the whole discipline handled end to end, from entity to citations to monthly reporting, that is our generative engine optimization service.