SEO for ChatGPT is the set of measures that increase the odds that ChatGPT mentions your brand, uses your pages as a source and recommends you when someone asks for advice, a comparison or a product. It differs from classic SEO on three points: there is no results page with ten links, the answer is assembled fresh in every conversation, and the source that gets picked is the source that makes the answer easiest. This article explains how ChatGPT arrives at its answers, what that means for your content and your brand, and how you measure whether it works.
How ChatGPT arrives at its answers
A ChatGPT answer comes about in two ways you need to keep apart. The first is the model itself: a large language model trained on an enormous amount of text up to a certain date. Everything it 'knows' about your brand without searching comes from that training data. You cannot edit that directly; you can only make sure that, over time, more consistent information about you exists so that the next model versions know you better.
The second way is search. For questions that need current or specific information, ChatGPT searches the web, reads a number of pages and writes an answer with source references. That mechanism is called retrieval-augmented generation: retrieve first, then generate. This is the part SEO for ChatGPT really acts on, because here the system decides per question which pages it opens and which it cites.
At the time of writing, OpenAI uses several bots with their own names for this. GPTBot collects pages for training models. OAI-SearchBot fetches pages so they can be shown in search answers. And ChatGPT-User fetches a page when a user explicitly asks for it in a conversation. Each of them reads your robots.txt; what you allow or block there determines which part of ChatGPT you can appear in. Read more about the bots in our explanation of the AI crawler.
How SEO for ChatGPT differs from SEO for Google
The biggest difference is that there is no position. Google shows a list; ChatGPT writes one answer and picks a handful of sources for it. You are part of the answer or you are not. Who ends up in the answer also varies from run to run: asking the same question ten times does not return the same sources ten times. That is why, for ChatGPT, we talk about how often you are mentioned, not where you rank.
The second difference is the shape of the question. People type 'shopify seo app' into Google; in ChatGPT they ask 'which seo app would you recommend for a Shopify store with 2,000 products that also wants to sell in Germany'. Questions are longer, contain context and ask for a recommendation. Content written only for short keywords misses that context.
The third difference is the role of your brand. A brand mention without a link counts fully in ChatGPT: if the model keeps encountering your name together with a category ('Traze, an autonomous AI SEO platform'), it links you to that category. Links remain important for the search part, but the reputation of your name across the whole web weighs more heavily than it does with Google.
Step 1: make sure ChatGPT can read your pages
Everything starts with reachability. Check your robots.txt and allow at least OAI-SearchBot if you want to appear in search answers. Whether you allow GPTBot is a separate trade-off between visibility in future models and control over your content; in any case, do not block it by accident with a blanket rule that turns away every unknown bot.
Then make sure the content is in the HTML the bot fetches. AI crawlers generally do not execute JavaScript the way Googlebot does. Content that only appears after a script runs — a product description loaded through an API, an FAQ that expands on click and only then enters the DOM — does not exist for ChatGPT. Server-side rendering of everything that matters is no longer a luxury.
Keep your pages light and fast. Bots have a time limit per page and give up on a slow page sooner. A clean HTML structure with real headings (H2, H3), lists and tables also helps to cut your text into usable chunks. An llms.txt file is optional, but a cheap way to tell AI systems in one overview what your site is and which pages matter.
Step 2: write answers, not texts
ChatGPT cites the passage that answers the question most directly. That means every page and every section has to open with the answer, not with a run-up. Put the definition, the advice or the number in the first two sentences and explain afterwards. Exactly the way this article opens.
Cover one question per section and use the question, or a recognisable variant of it, as the heading. An H2 like 'Should I allow GPTBot in robots.txt?' followed by a four-sentence answer is a perfectly citable block. An H2 like 'Our vision' followed by five paragraphs is not.
Be concrete and verifiable. Name versions, prices, quantities, dates and names, and cite the source when it is not your own data. Models weight passages with specific, checkable information more heavily than general claims — and generic text is something the model can write itself, so it does not need your page for that.
Use structure where it fits: a table for comparisons, a numbered list for steps, an FAQ for the questions your customers actually ask. Add structured data for what the page is (Article, Product, FAQPage) so the meaning of the page is unambiguous to machines too. And keep content current: a visible last-updated date is a signal to a model that the information still holds.
Step 3: turn your brand into an entity
A model attaches information to entities: things with a name and attributes. The more unambiguously your brand exists as an entity, the easier it is for the model to name you for the right question. Start with one sentence that is the same everywhere: what you are, for whom and how you differ. Use that sentence on your homepage, your about page, your LinkedIn profile and in every guest article.
Then make sure that sentence also appears outside your own site. Mentions in directories, review platforms, partner pages, podcasts and trade media together build the picture the model has of you. Consistency outweighs volume: ten mentions that say the same thing are worth more than a hundred that contradict each other.
Publish facts the model can point to: prices, integrations, founding year, location, team. An about page with real details and Organization schema is a more reliable anchor for ChatGPT than a mission page without facts.
Step 4: measure whether it works
Because ChatGPT answers vary from run to run, you measure visibility with prompt sampling: you ask a fixed set of questions repeatedly and count the outcomes. Build a list of twenty to fifty questions your customers really ask, in their own words and language: informational questions ('what is a good seo approach for an online store'), comparison questions ('which is better, X or Y') and recommendation questions ('which tool would you recommend for').
Ask each question several times per measurement period and record, per answer, whether your brand is mentioned, whether your pages are cited, in what wording, and which competitors appear next to you. From that follows your share of voice in AI answers and, more importantly, where you are missing while a competitor is named. If you want a first impression without counting yourself, run the free AI visibility check: it asks the questions for you and shows whether ChatGPT, Gemini and Perplexity mention you.
Measure monthly, not daily. The variation between days is larger than the trend, and a monthly cadence is enough to see the effect of new content and mentions.
What not to do
A few tactics keep coming up and do not work, or turn against you. Writing 'the best' everywhere without evidence makes your text generic and therefore unnecessary for the model. Hiding text from visitors but exposing it to bots — sometimes with instructions to the model inside — is treated as manipulation by OpenAI and Google and damages your brand the moment someone discovers it. Rolling out hundreds of thin pages to 'increase the odds' mostly produces thin content nobody cites. And blocking all bots as a precaution is the surest way to disappear from AI answers.
Conclusion: start with reachability and directness
SEO for ChatGPT is not a separate discipline next to SEO, but a shift in what you optimise: from positions to mentions, from keywords to questions, from pages to entities. The order is clear. First make your pages reachable for OpenAI's bots, then write answers instead of texts, build your brand consistently across the web and measure with a fixed set of questions whether it has an effect. If you would rather not keep track of this yourself, Traze does the research, the writing, the publishing and the measurement per engine for you — see how our AI SEO service works.