The short answer
- LLM SEO is the work of getting your pages retrieved, quoted and cited when ChatGPT, Gemini, Perplexity or Google's AI Overviews answer a question in your category.
- LLM-powered search runs a normal search first, then picks passages to quote. If a crawler cannot fetch your page, or a search index does not rank it, the model never sees it.
- The changes that move it: let the AI crawlers in, put the answer in the first two sentences of each section, write FAQ answers that stand alone, publish honest comparison pages, and describe your company the same way everywhere.
- Measure it with a fixed list of prompts run monthly in ChatGPT, Gemini and Perplexity, plus referral traffic and Search Console.
What LLM SEO means
LLM SEO, also called LLM optimization, is search optimization for answers instead of blue links. The goal is a mention or a citation inside the answer an assistant writes.
You will see the same work called generative engine optimization (GEO) or answer engine optimization (AEO). The labels overlap almost completely. This page is the practical version: how the systems pick sources, what to change on a site, and how to tell if it worked. For the longer background on each label, read the GEO guide and the AEO guide.
One thing LLM SEO is not: a replacement for SEO. Google says its generative AI features rely on its core Search ranking systems, and that the best practices for SEO continue to be relevant. Most of LLM SEO is good SEO with a sharper focus on answer-shaped writing and on what other sites say about you.
How LLM-powered search picks its sources
An assistant answering a current question does not read the whole web. It runs a pipeline, and you can lose at any step.
1. Retrieval
The assistant sends one or more search queries to an index and pulls back a short list of pages. Each engine has its own route:
| Engine | Crawler to allow | What its docs say about retrieval |
|---|---|---|
| ChatGPT search | OAI-SearchBot | OpenAI says ChatGPT search may send de-identified queries to Bing and other search providers, and that any public site can appear if it does not block OAI-SearchBot. |
| Perplexity | PerplexityBot | Perplexity says PerplexityBot surfaces and links websites in its search results and is not used to train foundation models. |
| Google AI Overviews and AI Mode | Googlebot | Google says there are no extra requirements to appear; pages must be indexed and eligible for a snippet in normal Search. |
| Gemini app | Googlebot | Google's Gemini help says it retrieves information from external sources such as Google Search and links to sources under the answer. |
Practical consequence: if your page ranks nowhere in Bing, it has a weak shot at ChatGPT search. If it is not indexed in Google, it will not show in AI Overviews.
2. Passage selection
The model reads the retrieved pages and lifts the passages that answer the question most directly. A paragraph that opens with the answer and contains the specific number, price or step gets quoted. A paragraph that warms up for three sentences gets skipped.
3. Citation
The answer links to the pages it used. You can be mentioned without being cited (the model names your brand from a review site) or cited without being recommended (your pricing page is the source for a competitor comparison). Track both separately.
4. Entity clarity
The model has to know what you are. If your homepage says "the operating system for growth" and a review site says "a Shopify reporting app", the model will repeat the clearest description it found, which may not be yours.
5. Freshness
For pricing, "best X" and comparison prompts, assistants favor recent pages. A comparison table last updated two years ago loses to a newer one with current prices. Update the facts and the visible content, not just the date.
What to change on your site
These are the changes worth doing, in the order they usually pay back.
Answer-first sections
Every H2 should be a question or a claim, and the first sentence under it should be the answer. "Plan X costs $49 a month billed annually" beats "Pricing is an important consideration for many teams." Keep the answer paragraph to one to three sentences so it can be quoted whole.
FAQs with self-contained answers
Write each FAQ answer so it makes sense pasted alone into a chat window. Repeat the subject ("Acme's Pro plan includes...") instead of "It includes...". Four to six real questions per page beat twenty padded ones. Mark them up with FAQPage structured data where it fits the page.
Comparison pages
"X vs Y" and "best X for Y" prompts are where buying decisions happen in assistants. Publish fair comparisons with a table of plans, prices, limits and the use case each product fits. Say where a competitor is the better choice. Assistants quote balanced tables; they skip pages that read as ads.
llms.txt
An llms.txt file is a plain Markdown index of your most important pages at the root of your domain. Be clear about what it does: Google's guidance says Google Search does not use llms.txt and that it neither helps nor hurts there. It is cheap to add and some AI tools and agents read it, so treat it as hygiene, not a ranking lever.
Structured data
Google states that structured data is not required for generative AI search and that there is no special schema to add. Keep using Organization, Product, FAQPage and Article markup where it is accurate, because it earns rich results and states your company name, logo and product facts in a machine-readable way.
Text the crawler can read
Prices hidden in images, specs inside tabs that load on click, and comparison tables rendered only by client-side JavaScript are often invisible to AI crawlers. Put the facts in server-rendered HTML text.
The crawler and index checklist
- robots.txt. Allow OAI-SearchBot, PerplexityBot and Googlebot. Blocking GPTBot only opts you out of OpenAI model training; it does not remove you from ChatGPT search, which uses OAI-SearchBot. OpenAI notes robots.txt changes can take about 24 hours to apply.
- CDN and firewall. Bot protection rules sometimes block AI crawlers even when robots.txt allows them. Check your firewall logs for the user agents above.
- Bing. Verify the site in Bing Webmaster Tools and submit your sitemap. Bing supports IndexNow for pinging new and updated URLs.
- Google. Confirm key pages are indexed in Search Console and are not blocked from snippets with nosnippet.
- Internal links. Link every comparison and FAQ page from somewhere crawlable, so the crawler finds it without a sitemap.
Entity clarity and third-party mentions
Assistants trust what several independent sources agree on. Your job is to make the agreement easy.
- Write one plain sentence that says what you are, who it is for and what it costs to start. Use it on your homepage, About page, directory listings and social bios.
- Keep the company name, category and pricing consistent across review sites, marketplaces and partner pages. A stale price on a directory will be quoted back to buyers.
- Earn real mentions: customer reviews, inclusion in independent roundups, answers in community threads where your product is relevant. Google warns that seeking inauthentic mentions across the web is less helpful than it might seem, and assistants are getting better at ignoring them.
How to measure LLM SEO
There is no rank position in an AI answer, so measure presence across a fixed prompt set.
- Build 20 to 50 prompts a buyer would actually type: "best accounting software for a 10-person agency", "Acme vs Rival pricing", "is Acme worth it". Mix category, comparison and brand prompts.
- Run them monthly in ChatGPT, Gemini and Perplexity, in a logged-out or clean session.
- Record four things per prompt: mentioned (yes or no), cited (which URL), position in the answer, and which competitors appeared.
- Watch referral traffic. OpenAI adds utm_source=chatgpt.com to links from ChatGPT, so it shows up as its own source in GA4.
- Check Search Console. Google added a Generative AI performance report that shows how people find your content through its AI features.
A worked example of what a month of tracking looks like for 30 prompts:
| Engine | Prompts with a mention | Prompts with a citation to your site | Top competitor mentions |
|---|---|---|---|
| ChatGPT | 9 of 30 | 4 of 30 | 17 of 30 |
| Gemini | 12 of 30 | 6 of 30 | 15 of 30 |
| Perplexity | 11 of 30 | 8 of 30 | 14 of 30 |
Read it this way: the gap between mentions and citations tells you whether your own pages are being used or only third-party pages about you. The competitor column tells you which prompts to write a comparison page for first. Re-run the same 30 prompts next month; changing the prompt list breaks the trend line.
For a quick baseline before you build a prompt list, run your domain through the AI visibility checker.
Common LLM SEO mistakes
- Blocking OAI-SearchBot by accident while trying to block training crawlers. Block GPTBot if you want; leave OAI-SearchBot open.
- Writing for the model instead of the reader. Keyword-stuffed "AI-optimized" pages do not get quoted. Clear answers do.
- Measuring one prompt once. Answers vary between runs. Trends across a fixed set are the signal.
- Ignoring Bing. Teams that only check Google miss the index ChatGPT search draws on.
- Treating llms.txt as the strategy. It is a five-minute file, not a substitute for pages worth citing.
Doing it with Sprites
Sprites' AI visibility agent checks how your brand shows up in ChatGPT, Gemini and Perplexity answers for the prompts you choose, then writes content aimed at being cited: FAQ pages, comparison pages and llms.txt. Pages publish to WordPress, Webflow, Shopify or Framer only after you approve them. AI visibility is included from the Launch plan at $495 a month; see pricing, which starts with a 7-day trial for $1.
Start with the 30-prompt baseline. Fix crawler access the same day, then write the comparison page for the prompt where a competitor shows up most.