ChatGPT SEO: How to Get Cited in AI Answers

Dominik Breitbach · Founder & Lead SEO Strategist

Dominik Breitbach founded taismo, an SEO and GEO agency from Munich, and works as its Lead SEO Strategist on ongoing SEO support and visibility inside AI answers. He makes companies findable where their customers search today.

⏱ Reading time: 14 min
🔄 Last updated: 26 July 2026

ChatGPT SEO covers every measure that gets your company named and cited inside ChatGPT’s answers. There is no list of ten results to climb. ChatGPT writes one answer and names a handful of sources, so visibility turns close to binary: You are in the answer or you are invisible. Two inputs decide it: What the model absorbed during training, and what OpenAI’s search crawler finds on your site at the moment somebody asks.

👉 Want to know whether ChatGPT already knows your company? An SEO audit with a GEO focus puts it in black and white.

ChatGPT SEO means two different things

Search “ChatGPT SEO” in English and you land on two different topics under one label.

Meaning one: Optimizing so ChatGPT cites you. The goal is a mention of your brand, your page or your expert inside the generated answer. This is the ChatGPT specific slice of Generative Engine Optimization (GEO) and what we build for clients as part of GEO and AI visibility.

Meaning two: Using ChatGPT as a tool to do SEO work. Keyword clustering, outlines, schema drafts, alt text at scale. That is a production question and gets its own honest section further down. The two pull in opposite directions, because mass producing pages with ChatGPT does nothing for your chance of being cited by ChatGPT.

Where ChatGPT gets its information

ChatGPT answers from two separate knowledge sources, and each responds to different work on your side.

The training corpus is fixed until the next model version ships. Whatever appeared often, consistently and in credible company across the open web before the cutoff is baked into the model’s weights. You cannot edit it. You can only make the picture of your company on the wider web consistent enough that the next training round gets it right.

The live retrieval path works like a search engine. When a question needs current information, ChatGPT runs a search, fetches pages and summarizes what it finds, with links. It reacts in days and rewards what classic SEO rewards, plus clarity.

Start with retrieval, not with a homepage rewrite: Get crawlable, get quotable, get found today. The training corpus follows on its own schedule.

Three OpenAI crawlers decide whether ChatGPT can see you

OpenAI operates three separate crawlers, each with a different job, and all three are controlled independently in your robots.txt. That is where a lot of accidental invisibility comes from.

The practical consequence catches many teams off guard: Blocking GPTBot does not remove you from ChatGPT’s search results, and blocking OAI-SearchBot does. Plenty of sites added a blanket AI block in 2024 to stay out of training, then wondered why ChatGPT never cites them. OpenAI documents the two as independent settings.

Google runs the same split: Googlebot handles Search, Google-Extended controls Gemini model training. Blocking Google-Extended has no effect on Search or on AI Overviews.

Writing the robots.txt rules without locking yourself out

The tokens in that table are exactly what you write into robots.txt. Matching runs on the product token, not on the full user agent string, and it ignores case. This block lets the search crawler in and keeps your pages out of model training:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

The trap sits in how groups are matched. A crawler obeys exactly one group, the most specific one that names it, and a named group is never combined with the global User-agent: * group. So the moment GPTBot gets a group of its own, every rule under * stops applying to it, including the disallows that keep staging paths, filter URLs and internal search results out of everyone’s reach. Whatever a named group should still respect has to be repeated inside it.

Conflicts inside a group go to the least restrictive rule, and the longer matching path wins. Disallow: /blog/ alongside Allow: /blog/ai/ keeps the AI folder crawlable and closes the rest of the blog.

The scope is narrower than most people assume. A robots.txt file governs only the host, protocol and port it is served from. Your bare domain, your www variant and every subdomain need a file of their own, which is how a forgotten staging host ends up handing an AI crawler content you never published.

And nothing here is instant. OpenAI puts the lag at roughly 24 hours from a robots.txt change to its systems adjusting, and Google caches robots.txt for up to 24 hours as well. Unblocking is a day, not a deploy, so make the change before a measurement run rather than during one.

A fourth agent is worth knowing about, even though it changes nothing for your organic visibility. OAI-AdsBot checks the safety of pages submitted as ads inside ChatGPT, its data does not train foundation models, and it has no say in whether an answer cites you.

Proving the setting does what you think it does

Robots.txt records what you asked for, never what happened. The evidence sits in your server logs. Filter them for GPTBot, OAI-SearchBot and ChatGPT-User and you see which agent arrived, on which URLs and with which status code. A 403 or a challenge page in that list is the edge rule from the pro tip above, caught in the act.

Then confirm the hits were genuine, because a user agent string takes one line to fake. OpenAI publishes the current IP ranges of each agent as JSON at openai.com/searchbot.json, openai.com/gptbot.json and openai.com/chatgpt-user.json. A request whose address is missing from the matching file did not come from OpenAI, whatever it called itself.

ChatGPT, Google AI Mode and Perplexity pull from different places

“ChatGPT SEO” is the common label, but four systems shape AI answers and they source information differently enough to matter.

ChatGPT combines training knowledge with its own search layer, fed by OAI-SearchBot. Google AI Overviews and AI Mode read the regular Google index, and Google states plainly that there are no extra requirements and no special optimizations for appearing in them. Both use a query fan-out technique, issuing several related searches across subtopics before writing the response, which is why they surface a wider set of sites than a classic result page. Perplexity leans hardest on live search and shows its sources in the open. Microsoft Copilot builds on Bing, so your Bing indexing status quietly matters again.

All four reward the same qualities: A page that is crawlable, current, unambiguous and backed by a recognizable organization gets picked up everywhere. They differ in speed. Our guide to GEO and AI visibility holds the systematic comparison, and the glossary entry on agentic SEO covers what changes once AI agents browse on the user’s behalf.

How to rank in ChatGPT: Four levers

Whether an AI system pulls you into its answer comes down to four levers that compound.

1. Citable content: Statements that survive being lifted out of their paragraph.
2. Entities and structured data: Machine readable facts about who you are and what you do.
3. Authority and mentions: Consistent references to you across sites you do not own.
4. Technical access: Crawlers can reach, render and quote your pages.

Start at the bottom of that list. Technical access gates everything above it, and authority takes longest to build.

Lever 1: Write answers that survive being lifted out

An AI system does not quote your page. It quotes a sentence from your page. Five principles cover most of what that changes:

• Answer in the first sentence. Put the conclusion at the top of the section, then explain. A model extracts the opening claim far more reliably than one buried three paragraphs down.
• Make every statement self contained. Write “OAI-SearchBot controls ChatGPT search visibility” instead of “this crawler controls it”. Out of context, the second sentence is worthless.
• Use numbers and named entities. “Three OpenAI crawlers” beats “several crawlers”.
• Structure with lists and tables. A comparison table hands a model clean, separable rows.
• Keep it current and say so. Systems with live search favor maintained pages, which is why content freshness has moved from nice to have into a ranking relevant habit.

None of this is a trick. It is the clarity that helps a reader in a hurry. Glossary and guide content is the strongest raw material for AI answers, and our own SEO glossary is the part of taismo.de that AI systems quote most.

Lever 2: Entities and structured data remove the ambiguity

AI systems reason in entities: Bounded things like companies, people, products and places, plus the relationships between them. State those relationships explicitly instead of hoping a model infers them from prose.

Structured data in JSON-LD does exactly that. Mark up your organization, authors, services and defined terms as one connected graph and you cut the biggest source of AI error: Confusing you with a similarly named company. The same markup is the groundwork for entering Google’s Knowledge Graph and for earning rich snippets.

Two constraints keep this honest. Google requires that structured data matches the visible text on the page, so markup claiming something the page does not show is a policy problem rather than a shortcut. And there is no AI specific schema type: Google says outright that you do not need new machine readable files or special markup to appear in AI features.

Lever 3: Authority is built off your own domain

The third lever is the one you cannot pull on your own website. AI systems weigh how often and how credibly other people talk about you. Mentions in trade publications, directories, review platforms and partner sites tell a model you are an established option rather than a name on a page.

For the training corpus, distributed and consistent mentions are close to the whole ballgame, which makes consistency the cheapest win available. Your company name, address, positioning and core service terms should read identically everywhere. Contradictory data dilutes the entity and forces a model to guess which version is real. This is the discipline E-E-A-T asks for in classic search, applied to a wider set of sources. AI visibility rests on the same authority that produces good rankings, which is one more reason to treat GEO as part of ongoing SEO support.

Lever 4: Technical access is the gate everything else passes through

If a crawler cannot reach, render or quote your page, the other three levers never get evaluated. Four checks cover the failure modes we see most often.

Crawlability. Robots rules, CDN bot filters and hosting firewalls all block AI crawlers, and only the first is visible in your repository.

Snippet eligibility. To appear as a supporting link in Google’s AI features, a page must be indexed and eligible to be shown with a snippet. A nosnippet directive, a max-snippet:0 setting or a data-nosnippet wrapper around your key paragraph removes that eligibility. One line, total effect.

Text in the HTML. Important content has to exist in textual form in the delivered markup. Content that appears only after client side rendering is a gamble with every crawler that does not execute JavaScript.

Findable structure. Internal links are how crawlers discover your depth. A pillar page that bundles its subtopics also teaches AI systems that you cover a field rather than one lucky article, which is why cornerstone content and dedicated grounding pages earn their keep here. Sites that machines parse without effort are the core of our SEO web design.

How to measure AI visibility in three steps

AI answers have no position tracker, but visibility is measurable. Three steps turn a gut feeling into a number.

Step 1: Build a prompt set. Write down 20 to 30 questions your customers actually ask, in their words: “Who does X in Munich”, “best provider for Y”, “how much does Z cost”.

Step 2: Run them across systems and record the answer. Ask ChatGPT, Google AI Mode, Perplexity and Copilot the same set. Note three things per question: Whether you are named, who else is named, which URLs get cited. The competitor list is usually the more useful half.

Step 3: Repeat monthly and read the trend. One run is a snapshot, twelve runs are a trend line. Track the share of prompts that name you, not just yes or no on your favorite question.

One quantitative source sits alongside this. Clicks and impressions from Google’s AI Overviews and AI Mode are counted inside the normal Search performance data under the Web search type, so they already sit in your Google Search Console reports. Since June 2026 there is also a separate generative AI performance report. It shows impressions only, split by page, country, date and device, it leaves out Search Labs experiments, and Google is rolling it out to a subset of site owners rather than everyone at once.

Using ChatGPT as an SEO tool: Where it helps and where it costs you

The other half of the “ChatGPT SEO” search demand is about production. Here is the honest split from daily agency work.

Where it saves hours: Clustering a keyword export into topics, drafting outlines from a brief, generating JSON-LD skeletons you then verify, writing alt text for a hundred images, building regular expressions for log analysis. A human checks every output against real data before it ships.

Where it costs you: Three jobs go wrong reliably. Facts and figures, because a fluent wrong number is harder to catch than an obvious gap. Competitor and search volume data, because the model produces plausible numbers it never measured. And copy that ships unedited, because it gives a reader no reason to stay.

Google’s position is clear: The method of creation does not decide rankings, quality does. Mass produced pages fail on quality, and quality is what AI systems screen for when they choose a source. We do not use AI detection scores as a quality gate. They measure the wrong thing.

What separates a useful prompt from a useless one is not the phrasing, it is what you hand over. Three rules cover most of it. Supply the data instead of asking the model to recall it: Paste the keyword export, the page source, the log extract. A model asked for search volumes invents them, the same model asked to cluster a list you provided is dependable. Ask for structure, not for facts: Grouping, reformatting, turning a table into JSON-LD, drafting an outline. The truth comes from you, the shaping comes from the model. Validate every machine readable output before it ships: A schema block that reads well and carries an invented property still fails, and a validator catches in seconds what a read-through misses.

Four myths about ChatGPT SEO

Myth 1: You can buy your way into AI answers. As a rule you cannot. AI systems select sources by relevance, clarity and authority. Advertising formats inside AI products carry an ad label and sit on a different surface from the cited sources.

Myth 2: Classic SEO is finished. Google states that the same SEO fundamentals apply to its AI features with no additional requirements, and ChatGPT’s search layer behaves like a search engine.

Myth 3: You need an llms.txt file to be visible. Google says explicitly that you do not need new machine readable files, AI text files or markup to appear in its AI features. A curated file can still serve as a briefing document, and we maintain one. Treat it as a bonus, never as the reason your visibility is stuck.

Myth 4: More content means more citations. Twenty pages that each answer one question completely outperform two hundred that circle a topic, because a model quotes a specific claim, never a general vibe.

All four share one point. AI visibility is maintenance: Models get retrained, the web moves, competitors catch up. That is why it belongs inside continuous SEO support.

Your first 30 days of ChatGPT SEO

Starting is less complicated than the topic sounds. This sequence has held up across client projects:

1. Days 1 to 3, baseline: Run your customer questions through ChatGPT, Google AI Mode and Perplexity and record who gets named. That gives you your starting point and your real AI competitors.
2. Days 4 to 7, unblock: Check robots.txt for OAI-SearchBot, verify no nosnippet directives sit on your money pages, confirm your CDN is not filtering AI agents.
3. Days 8 to 14, mark up your entity: Publish a connected JSON-LD graph for your organization, authors and services. One graph, consistent IDs, matching the visible page.
4. Days 15 to 24, make three pages quotable: Rewrite the openings of your three most important pages so each section answers its question in the first sentence.
5. Days 25 to 28, fix your mentions: Make company name, address and positioning identical across your directory and review profiles.
6. Day 30, re-measure: Run the same prompt set again. Live retrieval systems often move inside this window, so this is where you see whether the fixes landed.

Nothing on that list is exotic. It is SEO fundamentals aimed at clarity and machine readability. Starting early buys a head start, because most competitors still file this under next year.

Frequently asked questions about ChatGPT SEO

What is ChatGPT SEO?

ChatGPT SEO covers every measure that gets your company named and cited inside ChatGPT’s answers. The goal is a mention as a source in the generated answer rather than a position on a result page.

How do I rank in ChatGPT?

You get into ChatGPT answers through four levers: Content that answers a question in one liftable sentence, structured data that defines your entity, credible mentions on sites you do not own, and technical access for OpenAI’s search crawler. Start by confirming OAI-SearchBot is not blocked in your robots.txt.

Where does ChatGPT get its information from?

ChatGPT draws on two sources: A training corpus holding what was widely published before the model’s cutoff, and a live search layer that fetches current pages while the answer is written. Only the live path moves within weeks.

How is Perplexity different from ChatGPT for SEO?

Perplexity relies more heavily on live web search and shows its sources openly, so clean, current pages pay off there faster. ChatGPT blends training knowledge with its own search layer, which gives brand mentions across the wider web more weight.

How do I check whether ChatGPT knows my company?

Ask it the questions your customers would ask, such as “who offers X in my region”, and record whether you are named and which sources get cited. Repeat monthly, because answers change with every model update.

How do I block ChatGPT from using my content?

Decide first which of the two settings you mean. Disallow: / inside a User-agent: GPTBot group keeps your pages out of OpenAI’s model training and leaves your search visibility untouched. The same rule under User-agent: OAI-SearchBot takes you out of ChatGPT’s search results. Give each agent a group of its own, repeat any global disallow rules inside it, and allow around 24 hours for the change to register.

Which user agent does ChatGPT use?

Three, with separate jobs. OAI-SearchBot feeds ChatGPT’s search features, GPTBot collects content that may train OpenAI’s foundation models, and ChatGPT-User fetches a single page when a user action asks for it. Each one sends a token you can filter for in your server logs.

Sources

• Google Search Console Help: Generative AI performance report, Google, 2026.
• OpenAI: “Overview of OpenAI Crawlers“, OpenAI Developer Documentation, 2026.
• Google Search Central: “AI features and your website“, Google, 2026.
• Google Search Central: “Overview of Google crawlers and fetchers“, Google, 2026.
• Google Search Central: “How Google Interprets the robots.txt Specification“, Google, 2026.
• Google Search Central: “Creating helpful, reliable, people-first content“, Google, 2026.