How ChatGPT picks sources, and why your brand isn't there yet

Generative AI doesn’t pick random sources. Behind every ChatGPT answer sits a surprisingly predictable system. Once you understand that system, you also understand why your brand isn’t being mentioned yet. And more importantly: what you can do about it.

I work as a freelancer under JSResults on SEO and on AI search optimization (GEO). In the audits I run for clients, the same picture keeps coming back: the content is usually fine, but it misses exactly the signals AI looks for. This article is my way of making those signals clear.

I break the system down into three main components: factual authority, structural clarity and repetition in the data. For each component I give you five concrete actions you can take today. No theory without execution, but a list you can start on tomorrow.

First this: ChatGPT finds sources in two ways

Before we get to the three components, you need to know how ChatGPT gets its sources in the first place. Because there are actually two ways, and they work very differently.

1. From the training data. This is the knowledge inside the model itself. It was collected during training and is therefore not live. You get in here by being mentioned often and consistently across the open web, so your information ends up as a pattern inside the model.
2. Through live searches. When ChatGPT looks something up in real time, it uses the Bing search index, supplemented by its own crawler called OAI-SearchBot. OpenAI has confirmed this connection with Bing itself. Your position in Bing therefore strongly predicts whether you get cited in a ChatGPT answer.

This distinction matters. For the training side, everything revolves around repetition and authority over time. For the live-search side, everything revolves around discoverability and structure right now. The three components below cover both sides.

Component 1: Factual authority

AI systems don’t want to cite the prettiest text. They want to cite the source they dare to trust. Authority is that trust. And authority for AI is something different from a high Domain Rating in an SEO tool.

For AI, brand authority weighs more heavily than classic backlinks. By that I mean: are you being mentioned? Are there reviews about you? Do you appear in places the model already trusts, such as comparison sites, industry platforms and well-known media? Those signals count for more than a link in a footer.

This is also where I see the most to gain in audits. Many brands have built their authority neatly within their own network, but not in the independent places an AI model checks. Closing that gap is often the difference between being mentioned and being skipped.

Five actions for factual authority

• Claim and complete your profiles on platforms AI trusts. Think G2, Capterra, Trustpilot and relevant industry registers. A complete, active profile increases the chance you get cited.
• Ensure consistent mentions of your brand name, plus what you do, across multiple independent sites. Not once, but the same story everywhere.
• Back up your claims with numbers, sources and data. A sentence with a concrete figure and a reference is picked up sooner than an opinion without proof.
• Tie your content to a real person. An author with a name, a bio and demonstrable experience is a stronger authority signal than an anonymous company text.
• Actively collect reviews and testimonials. For AI, a pile of positive, independent mentions is a sign that your brand really matters.

Component 2: Structural clarity

An AI doesn’t read your page the way a human does. It looks for chunks of text it can cleanly cut out and reuse as an answer. The easier you make that, the greater the chance your text becomes the answer.

What consistently helps: a clear heading structure, the answer right at the top, and short, self-contained paragraphs. Analyses of AI citations show that the beginning of a page is cited by far the most. So the first paragraphs under a heading do the most work.

Five actions for structural clarity

• Give the answer directly. Under each heading, put the short answer first in two or three sentences, and only then the explanation and nuance.
• Use a logical heading structure. One H1, then H2s below it, and only then H3s. Let the headings contain the reader’s actual question.
• Write headings as questions people really ask. ‘What is llms.txt?’ works better than ‘Introduction’. AI matches questions to questions.
• Add an FAQ section with short question-and-answer blocks. This format is almost exactly what an AI wants to return as an answer.
• Use tables and lists for comparisons and steps. Structured data is easier for a model to extract than a long running paragraph.

Component 3: Repetition in the data

The third component is the most patient one. An AI trusts a fact more the more places it has encountered that fact. Being mentioned once is noise. Encountering the same story everywhere becomes a pattern. And patterns become knowledge.

This is exactly why new or small brands aren’t there yet. Not because your content is bad, but because it hasn’t been repeated often enough and in enough places. Freshness plays a role here too: recently updated content is demonstrably cited more often than content that has sat still for years.

Five actions for repetition in the data

• Define your key messages and repeat them consistently. Use the same definitions and phrasing on your site, your socials and in guest articles.
• Be present on multiple channels. LinkedIn, YouTube, industry media, podcasts. Each extra channel is another place where the pattern repeats.
• Update your most important pages regularly. An update with a new date and fresh numbers keeps the content fresh in the eyes of AI.
• Seek guest publications and collaborations. Being mentioned on a site the model already trusts speeds up the whole process.
• Build a recognisable entity. Tie your brand name firmly to your field, so AI automatically connects ‘JSResults’ with SEO and GEO.

Why your brand isn’t there yet

When you hold these three components up against your own situation, the penny usually drops fast. Almost always it’s not quality that’s missing, but one of these three things.

• Your authority is there within your network, but not yet in the places AI checks.
• Your content is written well for people, but not structured for extraction by a model.
• Your story sits in one place, not ten. There’s no pattern to trust yet.

The good news: all three are buildable. It’s not a matter of luck or budget, but of consistently doing the right things.

What I would do if I started today

If I were to start working on AI visibility for a brand from scratch tomorrow, this is the order I would follow:

• Fix the structure first. Answer at the top, clear headings, FAQ blocks. This is fast and has an immediate effect on the live-search side.
• Then secure discoverability. Index the site in Bing Webmaster Tools, and give OAI-SearchBot access in robots.txt.
• Then build authority. Claim profiles, collect reviews, tie an author to the content.
• Finally, organise repetition. A rhythm of publishing, updating and telling the story across multiple channels.

This is not a sprint. But it is a path that works, because it lines up precisely with how these systems pick sources. Want this done for your brand? Take a look at my GEO audit or book a call.

Frequently asked questions

How does ChatGPT know which sources are trustworthy?

ChatGPT relies on two things: what it encountered often and consistently during training, and what it finds live through the Bing search index. Brand authority, mentions and reviews weigh more heavily than individual backlinks.

Why isn’t my website mentioned in ChatGPT?

Usually for one of three reasons: your content isn’t structured for extraction, you’re not (well) indexed in Bing, or your story hasn’t been repeated often enough in places the model trusts.

Does updating my content often help?

Yes. Recently updated content is demonstrably cited more often than old, static pages. An update with fresh numbers and a new date is one of the easiest wins.

Is GEO the same as SEO?

No, but they overlap strongly. Good SEO, especially a strong Bing position, helps directly on the live-search side of ChatGPT. GEO adds structure and authority for AI extraction on top of that.

Sources

This article combines my own hands-on experience with publicly verified information about how ChatGPT Search works. I checked the technical premises against these sources:

• OpenAI on ChatGPT Search and the connection with the Bing search index, and the role of the OAI-SearchBot crawler.
• Yoast, ChatGPT Search and how it uses Bing data: yoast.com/chatgpt-search
• Leapd, How ChatGPT, AI Overviews and Perplexity source information (2026): leapd.ai/blog

Where I write about patterns in citations (such as the importance of the start of a page and of freshness), I base myself on published analyses plus what I see in my own audits. I deliberately leave out specific percentages where the evidence differs per source.