How ChatGPT Works

ChatGPT started as a writing tool. Today it is one of the places where your future customers look for answers, compare options and ask for recommendations. When someone types "which accounting software suits a small online shop", they do not get ten links. They get an answer in which a handful of brands are named, and a handful of sources are cited.

That makes the question a practical one for anyone responsible for marketing: what decides whether ChatGPT mentions you? The answer sits in how the tool finds what it knows, so that is where this starts. And the work itself is SEO, even though the surface is a chat window rather than a result page.

Two layers of knowledge

ChatGPT draws on two separate layers, and each needs different work from you.

The model's built-in knowledge. The language model behind ChatGPT was trained on very large amounts of text from the web. Everything written about your brand up to the model's cut-off date shapes what it "knows" about you: your own site, brand mentions, reviews, industry coverage, directories. If you are well documented and described consistently across sources, you exist in the model's picture of your market. If you are not, it either guesses or names a competitor.

Live web search. When a question needs current information, ChatGPT searches the web, reads a selection of pages and cites them with links in the answer. Here it behaves much more like a search engine: your page has to be findable, fetchable and understandable at the moment the question is asked.

The two layers reinforce each other. A brand the model already knows gets picked and cited more often when it searches live, and live citations in credible sources eventually become training data.

What decides whether ChatGPT cites you

There is no published ranking factor list, and anyone who claims otherwise is selling something. The mechanism, though, points consistently at four things.

• Crawl access. OpenAI operates its own crawlers, including OAI-SearchBot for search and GPTBot for training data. If your robots.txt blocks them, you have opted out of the visibility. Check this first. It takes five minutes and it determines whether anything else you do matters.
• Content that can be quoted. ChatGPT lifts passages that answer a question precisely and hold up on their own. Clear headings, definitions, numbers and structured comparisons are far easier to quote than long unbroken prose. This is the same discipline as writing for featured snippets, taken further.
• Authority across sources. The model weighs what other people say about you. Mentions in media, trade publications and review sites build the external picture that recommendations rest on. It is the same dynamic that makes digital PR one of the strongest disciplines in AI search.
• A credible sender. The E-E-A-T signals, named experts, documented experience, clear company information, decide which sources get treated as fact rather than as marketing.

Your experience is the raw material

One point has held since the first arguments about AI and content: a language model has no value without data, and generic content that already exists in a thousand versions adds nothing the model needs. What the models lack is what only you have. Your own numbers, your cases, what you have learned from doing the work.

That inverts the fear. Plenty of people expected AI to make content work redundant. In practice original content became more valuable, because it is what both Google and ChatGPT need in order to have something to cite. Publish a study with your own data and the answer exists only with you, which means the citation is yours.

Working on visibility in ChatGPT

1. Open the doors technically. Verify that OpenAI's crawlers have access, that the site renders cleanly without JavaScript execution, and that important content is not hidden behind scripts or a login. Most AI crawlers do not run JavaScript, which is covered in more detail under JavaScript SEO and rendering.
2. Answer the questions customers ask in conversation. Prompts are longer and more specific than classic keywords. Someone types "cheapest way to ship pallets from Denmark to Germany", not "pallet shipping". Build content that matches whole questions, not single terms.
3. Make your brand facts consistent. The same description of who you are, what you do and who you help, on your site, in your profiles and in third-party mentions. Inconsistency produces uncertain models, and uncertain models leave you out.
4. Invest in brand mentions. Every credible third-party source that names you is an argument for the model when it picks who to recommend.
5. Publish original knowledge on a rhythm. Your own data and clear professional positions are the shortest route from background noise to being the source.

Measure it rather than guessing

Visibility in ChatGPT can be measured. Which questions in your category trigger recommendations, who gets named, on which sources, and how that moves over time. We call that discipline LLM monitoring, and it is the foundation under any serious effort, because without it you cannot tell whether the work is landing.

There is one honest caveat worth carrying with you: the models are not deterministic. The same prompt can produce different answers on different days. Measure patterns across repeated runs rather than single answers, and neither celebrate nor panic over one result.

If you want a concrete picture of where you stand today, our generative AI optimization work is built for exactly that: making your brand one of the ones the language models know, trust and recommend.