LLM SEO: What Actually Changes What AI Says About You, and What Doesn't

LLM SEO is the work of changing what large language models such as ChatGPT, Gemini, Perplexity and Claude say about your brand. You will also see it called LLM optimization, GEO or AI SEO. Most advice under those names treats an AI answer as one thing you can tune. It isn't. An answer comes from the model's memory or from a live search, and you can reach only one of them.

This guide splits an answer into the stages that produce it, shows what you control at each one according to the platforms' own documentation, and sets out what LLM SEO tools can and cannot measure.

The four stages behind an AI answer

Stage	What happens	What you control
1. Memory	The model answers from training data frozen at a knowledge cutoff, with no search	Nothing, until the next model ships
2. Access	Crawlers fetch your pages and an index stores them	Almost everything: robots.txt, your firewall, rendering, indexing and snippet controls
3. Retrieval	For a given question, the platform runs its own searches and picks candidate pages	Whether your page answers the question. Not where it lands in the list
4. Generation	The model writes the answer from what it retrieved and chooses what to cite	Whether the facts on your pages are correct and agree with each other

The evidence for each stage is laid out in how LLMs prioritise information about brands. Here is what it means in practice.

Stage 1: memory answers don't read your website

When a model answers without searching, it uses what it learned in training. OpenAI lists a cutoff for each model. Its chat-latest model, which "points to the latest Instant model currently used in ChatGPT", shows an "Apr 20, 2026 knowledge cutoff" (OpenAI). GPT-6 Luna shows May 18, 2026. Two OpenAI models, two different snapshots of the world.

Nothing you publish this quarter changes a memory answer. It changes when the vendor ships a new model. The only route in is the training crawl: GPTBot for OpenAI, ClaudeBot for Anthropic. Anthropic says restricting ClaudeBot "signals that the site's future materials should be excluded from our AI model training datasets" (Anthropic). Neither vendor says allowing its training crawler improves how you appear, and we know of no study that measures it. Treat it as unmeasured.

Stage 2: access is where LLM SEO works

This is the one stage where you control every variable and the platforms document the consequence. Every lever here removes a block. None of them makes a page more likely to be chosen.

• Per-bot robots.txt. OpenAI: sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers, though can still appear as navigational links" (OpenAI). Anthropic says disabling Claude-SearchBot "may reduce your site's visibility and accuracy in user search results." Perplexity: "To ensure your site appears in search results, we recommend allowing PerplexityBot in your site's robots.txt file."
• Your firewall. robots.txt is half the rule. OpenAI's eligibility test is to "allow OAI-Searchbot to crawl the site and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses" (OpenAI). A 403 from your CDN removes you, and nothing in your analytics shows it.
• Rendering. ChatGPT's and Claude's crawlers fetch JavaScript but "don't execute them. They can't read client-side rendered content" (Vercel). A price that appears only after a script runs doesn't exist for them.
• Google eligibility. For AI Overviews and AI Mode, "a page must be indexed and eligible to be shown in Google Search with a snippet" (Google), and the site must be included in generative AI features in Search Console. noindex, nosnippet, max-snippet and data-nosnippet all limit what Google shows.

Our AI crawler list maps each bot to its platform, and GPTBot vs OAI-SearchBot explains the training-versus-search split.

Stages 3 and 4: retrieval and generation

At answer time the platform doesn't search for your customer's exact words. OpenAI: "ChatGPT search typically rewrites your query into one or more targeted queries". Google says AI Overviews and AI Mode may issue "multiple related searches across subtopics and data sources". You can't see those queries, with one exception: Bing Webmaster Tools shows "the key phrases the AI used when retrieving content that was referenced in AI-generated answers", from a sample (Bing).

The research here is narrow. A 2026 survey of 45 GEO studies found "topical relevance and context position are the most reproducible levers", and that "citation-oriented rewrites can impair retrieval." You control relevance. You don't control position. A benchmark of rewriting tactics found most "not only largely ineffective but also frequently have a negative impact on document ranking."

What you control at generation is whether the page says the right thing. If your pricing page and your FAQ give two different prices, the model has two answers to choose from.

LLM SEO myths, briefly

Myth	What the source says
"Add an llms.txt file"	Google: "Google Search ignores them." See our llms.txt guide
"Add schema to get cited"	A matched test of 1,885 pages found "no major uplift in citations on any platform." See schema markup for AI
"Chunk and rewrite for AI"	Google: "You don't need to write in a specific way just for generative AI search."
"Raise your E-E-A-T score"	Google: E-E-A-T "isn't a specific ranking factor." See E-E-A-T for AI
"Allow GPTBot to appear in ChatGPT"	GPTBot is for training. OpenAI: "Each setting is independent of the others."
"Up to 40% more visibility"	Gains "conditional on a source already being present in a fixed context" (2026 survey)

Promises like these are also the red flags in our guide to hiring an AI SEO agency, and automation playbooks get the same treatment in our fact-check of AI-automated GEO tactics.

What LLM SEO tools can and cannot measure

LLM SEO tools, ours included, answer some questions well and are blind to others. Know which is which before you read a dashboard.

What they can measure

• Whether each AI crawler can reach your pages, as a yes or no.
• Whether your key facts are in the raw HTML.
• What a sample of AI answers says about you on a given day, and which competitors and sources they name.
• Whether those answers get your facts right.

What they cannot

• The real queries. Platforms rewrite prompts before they search, so a prompt set is a proxy. With memory on, ChatGPT "may use relevant saved memories when rewriting a search query."
• A true position. Google: "No third-party tool has access to our internal ranking or AI systems." Google's own generative AI report counts impressions, with no clicks and no positions. On ChatGPT search, OpenAI says "Placement is not guaranteed."
• One answer for "ChatGPT". A memory answer and a search answer are different measurements. A tool should say which it ran, with which model, on which date.
• A stable result. The same survey found "substantial run-to-run variability" in commercial audits. One run is a sample, not a trend.

LLM SEO checklist

1. Allow the search crawlers you want (OAI-SearchBot, PerplexityBot, Claude-SearchBot, Googlebot). Decide on training crawlers separately.
2. Request your key pages as each crawler and confirm a 200, not a 403 or a challenge page.
3. View the source of your pricing and product pages. Prices, product names and claims belong in the raw HTML.
4. Remove accidental noindex, nosnippet, max-snippet:0 and data-nosnippet, and check the generative AI setting in Search Console.
5. Make every fact agree across your site: pricing page, FAQ, About page, footer and structured data.
6. Ask the AI platforms about your brand. Record the model, the date and whether it searched, then check each claim.
7. Re-check after model updates. Memory answers change only when the model does.

99Visibility handles the measuring side. It audits five AI platforms (ChatGPT, Gemini, Perplexity, Google AI Overviews and Claude), checks whether AI crawlers can reach your site, checks the answers against the brand facts you enter, and tracks your competitors. Your first audit is free, with no credit card, and covers ChatGPT and Google AI Overviews. See the plans for all five platforms. Inside the dashboard, Agent V reads your audits and your site and drafts the fix, such as page copy for you to paste. It never edits your site itself.

FAQ

What is SEO for LLMs called?

LLM SEO, LLM optimization, GEO (generative engine optimization), AEO (answer engine optimization) and AI SEO all describe the same work: changing what AI platforms say about you. The names differ more than the work does.

Is SEO dead now with AI?

No. Google: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." Indexing and crawlability still decide eligibility there. What's new is that other platforms run their own crawlers, so access has to be checked per platform.

Can I change what ChatGPT says from memory?

Not this quarter. A memory answer reflects training data up to the model's cutoff. Make the facts correct and reachable on your own site, so search answers get them right now and future training crawls find the right version.

See what AI says about your brand today. Run a free AI visibility audit: no credit card.