What LLM SEO covers
You will see the same work called LLM optimization (LLMO), ChatGPT optimization, generative engine optimization (GEO) and answer engine optimization (AEO). The labels differ mainly in emphasis; our AEO vs GEO article compares them. Whatever you call it, LLM SEO has three parts:
- Access: AI crawlers can fetch your pages and read the main text without running JavaScript.
- Answers: you have pages that answer the questions buyers actually ask, in plain language that is easy to quote.
- Reputation: other sites (lists, reviews, forums, docs) mention you, because AI answers lean heavily on them.
How LLMs find the sources they cite
An assistant can answer in two ways. It can answer from model knowledge, what it learned in training, which changes only when a new model ships. Or it can search the web while answering and cite the pages it read. For LLM SEO the second route matters most, because you can influence it this month.
OpenAI documents three crawlers, and they do different jobs (see its crawler documentation):
| User agent | What it does | If you block it |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT search | Your pages are left out of ChatGPT search answers |
| GPTBot | Collects content that may be used to train models | Signals your content should not be used for training |
| ChatGPT-User | Visits a page when a user’s request needs it | Does not control search visibility; robots.txt may not apply |
OpenAI’s ChatGPT search help page says that ChatGPT sometimes sends rewritten queries to third-party search providers, and may follow up with more specific queries after reading the first results. It also says ranking depends on a number of factors and top placement can’t be guaranteed. OpenAI does not publish more than that, so treat anyone who claims to know “the algorithm” with care.
Outside studies fill in some of the picture. AirOps and Kevin Indig ran 16,851 queries through ChatGPT three times each and published the results in April 2026. A page in the first search position was cited 58.4% of the time, against 14.2% at position ten. Pages whose headings closely matched the query were cited more often, and focused pages beat ones that tried to cover everything. Only 2.3% of page-and-query pairs were cited in all three runs, which is a good reminder that single answers prove little.
What Vercel changed for LLM SEO
The most cited example is Vercel. In a June 2025 post, How we’re adapting SEO for LLMs and AI search, the company wrote: “ChatGPT now refers around 10% of new Vercel signups.” It said that was up from 4.8% the month before and about 1% six months earlier. What Vercel says it does:
- Watches X, Reddit, GitHub, Discord and forums for new questions, and writes in depth where competitors are thin.
- Publishes original data, code and quotes, with consistent terms and short, self-contained points.
- Uses a clean heading hierarchy, tables and JSON-LD, and serves static or server-rendered HTML because most AI crawlers don’t run JavaScript.
- Takes part in community threads, changelogs and open source rather than buying links.
- Reviews content at 30, 90 and 180 days, fixes 404s and keeps the sitemap current.
- Tracks AI referrers, citations in answers that show sources, and mentions.
These are Vercel’s own figures and Vercel’s own account of what helped. For more companies, see ChatGPT traffic case studies.
LLM SEO vs traditional SEO
| Traditional SEO | LLM SEO | |
|---|---|---|
| Goal | A high-ranking link and a click | A mention in the answer, ideally with your page as a cited source |
| Unit of work | Keyword | Question (often long and specific) |
| Off-site | Backlinks | Mentions on lists, reviews and forums the answers cite |
| Measurement | Rank and clicks | Mention rate across repeated answers, citations, AI referrals |
The foundation is shared. If your pages are not indexed or rank nowhere, the search step behind an AI answer is unlikely to find them.
An LLM SEO checklist
Access
- robots.txt allows OAI-SearchBot (and the other AI crawlers you want). Decide on GPTBot separately; it is about training.
- The main text of key pages is in the server HTML, not injected later by JavaScript.
- sitemap.xml lists your important pages with honest last-modified dates.
- Your site is indexed in Google and Bing; check both webmaster tools.
Answers
- One page per real buyer question: comparisons, “best X for Y”, pricing, alternatives, how-tos.
- Headings that match the question, and the answer in the first sentence or two under them.
- Specifics a model can quote: numbers, limits, prices, supported platforms, dates.
- Original data or examples that nobody else has.
- Organization and Product or SoftwareApplication JSON-LD with a plain description.
Reputation
- Find the third-party pages cited for your category’s questions and work on being included, accurately.
- Answer questions in communities where your buyers ask them, with disclosure and without link-dropping.
- Use the same one-line description of your product everywhere.
Maintenance
- Review top pages every few months and update facts that changed.
- Track AI referrals and re-run the same questions on a schedule.
Tactics with weak evidence
Some popular LLM SEO advice has little support so far. SE Ranking’s study of 129,000 domains found llms.txt had negligible impact on ChatGPT citations, and pages with FAQ sections were not cited more than pages without. Webflow’s growth team reportedly tried llms.txt and saw no significant lift. Long “ultimate guides” did worse than focused pages in the AirOps data when relevance was held equal. None of this means these things hurt; it means they should come after the basics. All of these studies are correlational.
How to measure LLM SEO
Start with analytics: ChatGPT adds utm_source=chatgpt.com to the links people click in its answers, so that traffic is easy to segment. Add a “How did you hear about us?” option for AI assistants too, because many people read an answer and then search your name, which shows up as search or direct traffic.
Then measure the answers themselves. Pick 10–20 questions a buyer would ask, run each more than once, and record whether you are mentioned, where, and which pages are cited. You can start with the free AI visibility checker: it asks five buyer-style questions on OpenAI with web search and checks your robots.txt rules for the main AI crawlers, llms.txt, JSON-LD, server-rendered text and sitemap, with no sign-up. The one-time $9 report asks 15 questions on OpenAI and Claude with web search plus Gemini from model knowledge, each twice, and lists every cited page with a suggested action. It uses official APIs, so answers can differ from the consumer apps, and it doesn’t test Perplexity or Google AI Overviews.
For the citation side in detail, read how to get cited by ChatGPT.
LLM SEO FAQ
What is LLM SEO?
LLM SEO is the work of getting your brand mentioned and your pages cited when large language models such as ChatGPT, Claude and Gemini answer questions. It covers crawler access, pages that answer specific questions clearly, and presence on the third-party pages those answers draw from.
Is LLM SEO different from regular SEO?
It builds on it. When ChatGPT searches the web it sends queries to search providers, so pages that rank well are more likely to be picked up. The differences are the goal (a mention or citation inside an answer rather than a click from a results page) and how you measure it.
Is LLM optimization the same as ChatGPT optimization?
Mostly. LLM optimization (LLMO), ChatGPT optimization, GEO and AEO are overlapping names for the same work. ChatGPT optimization usually just means focusing on the assistant that sends the most referral traffic today.
Does llms.txt help with LLM SEO?
There is little evidence that it does. SE Ranking's study of ChatGPT citations found it had negligible impact, and Webflow's growth team reportedly saw no significant lift. It is cheap to add, but it should not be a priority.
How long does LLM SEO take to work?
Crawler changes can take effect in about a day (OpenAI says roughly 24 hours after a robots.txt change). Content and third-party mentions take longer, and answers vary between runs, so measure rates over several weeks rather than single answers.