Get Recommended by ChatGPT and AI Search: A GEO Guide

"How do we get ChatGPT to recommend us?" is now one of the questions founders ask me most. The honest answer: nobody can guarantee it, and anyone selling a guarantee is guessing. What you can do is remove every technical reason an assistant has to skip you, and make your pages the easiest ones to quote.

Here's the checklist I use, with a note on what's documented and what's still speculation.

What GEO actually means

Generative engine optimisation (GEO) is the work of making your content more likely to be used and cited in AI-written answers. The term comes from GEO: Generative Engine Optimization by Pranjal Aggarwal and colleagues, published in November 2023 and accepted to KDD 2024. They rewrote source pages in different ways and measured how visible each became in generated answers:

• Adding quotations, statistics and cited sources worked best, with visibility gains of up to about 40% on their metrics.
• Keyword stuffing didn't help; on their main metric it did slightly worse than changing nothing.
• Lower-ranked pages gained the most: citing sources lifted the fifth-ranked page's visibility by 115%.

The experiments ran on 2023-era models, so trust the direction more than the numbers.

How AI assistants find and cite pages

There are two routes, and they need different work.

Search-backed answers

ChatGPT search, Perplexity, Google AI Overviews, Google AI Mode and Microsoft Copilot run a search, read a few pages and write an answer that links to them. This is the route you can influence now, and each leans on an index:

• Google AI Overviews and AI Mode use Google's index. Google says there are no additional requirements beyond being indexed and eligible for a snippet. AI Mode has been available in more than 200 countries and territories since October 2025.
• Copilot draws on Bing.
• ChatGPT search uses OpenAI's crawler, OAI-SearchBot, and OpenAI has said it also uses third-party search providers.
• Perplexity crawls with PerplexityBot.

So GEO starts with plain technical SEO. If Google and Bing haven't indexed you, a large part of AI search can't find you.

Training data

The second route is what a model learned in training, from crawlers such as GPTBot and ClaudeBot. It only changes when a new model ships and rarely comes with a link. The best you can do is be described consistently across the web.

Crawl access: search crawlers versus training crawlers

AI companies now run separate bots for separate jobs:

• Search crawlers index pages so they can be cited. Block one and you drop out of that product's answers.
• Training crawlers collect content for future models. Blocking them opts you out of training, not out of search answers.
• User-triggered fetchers visit a page because someone asked the assistant to, and may ignore robots.txt.

At the time of writing:

Google-Extended doesn't keep you out of AI Overviews or AI Mode, which follow the normal Search controls (nosnippet, max-snippet, noindex). Because it also covers grounding in Gemini apps, I don't block it on sites that want to be recommended.

For a service business that wants to be recommended, this is the robots.txt I start from:

# Answer engines: let them index public pages
User-agent: OAI-SearchBot
User-agent: Claude-SearchBot
User-agent: PerplexityBot
Disallow: /account/

# Model training: a business decision. These lines opt out.
User-agent: GPTBot
User-agent: ClaudeBot
Disallow: /

# Everyone else, including Googlebot and Bingbot
User-agent: *
Disallow: /account/

Sitemap: https://example.com/sitemap.xml

A crawler obeys only the most specific group that names it, so repeat private paths in each group. The training block is optional: I leave it out on my own site because I want future models to know what I do.

Check your CDN and firewall too. Cloudflare has blocked known AI crawlers by default on new domains since July 2025, and rules aimed at scrapers can catch OAI-SearchBot. OpenAI publishes its bots' IP ranges as JSON for allowlisting.

Server-side rendering: put the facts in the HTML

A late-2024 analysis by Vercel and MERJ found that none of the major AI crawlers it observed, OpenAI's and Anthropic's included, executed JavaScript. Googlebot is the exception: it renders pages.

I build as if that still holds. If your Vue or React app renders content in the browser, an AI crawler may get an empty <div id="app">. Test it by fetching a page the way a bot does:

curl -s -A "OAI-SearchBot" https://example.com/pricing | grep -i "per month"

If your price, service or location isn't in that output, an assistant probably can't see it either. Nuxt and Next.js render on the server by default, which is a big part of why I build client sites with Nuxt. Still on Nuxt 2? I covered the move from Nuxt 2 and Vue 2 to Nuxt 4.

I've seen no evidence that ChatGPT or Perplexity look at Core Web Vitals. They still matter, because Google's ranking systems use them and people who click a citation leave slow pages. Aim for LCP under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1.

Entity information: say clearly who you are

An assistant that recommends a business has to be sure which business it means:

• Organization or Person schema. Add schema.org JSON-LD for your company, or for yourself if you're a consultant: name, URL, logo or photo, description, address, contact details.
• sameAs links. Point sameAs at your LinkedIn, GitHub, Crunchbase or Wikidata profiles to tie them to your site.
• Consistent NAP. Keep your name, address and phone identical on your site, Google Business Profile, LinkedIn and directories. Contradictions give assistants a reason to hedge.

Google says AI features need no special schema, and that markup should match the visible text. I treat structured data as a way to remove ambiguity, not as a secret AI signal. On liviu.dev every page carries a Person entity with sameAs links to my profiles, posts carry BlogPosting and listing pages carry ItemList, all generated with nuxt-schema-org.

Content that can be quoted

Assistants quote passages that answer the question on their own, so:

• Answer first. Put the direct answer in the first sentence under each heading.
• Use numbers. "Audits start from €900 and take a few working days" can be quoted. "Competitive pricing" can't.
• Write real FAQs from the questions clients ask on calls.
• Publish comparisons. "X vs Y" pages and tables are exactly what people ask assistants for.
• Cite sources. In the GEO paper this was one of the most effective changes.
• Give each service its own page, rather than one homepage that lists twelve.

llms.txt: cheap to add, unproven

Jeremy Howard of Answer.AI proposed llms.txt on 3 September 2024. It's a markdown file at /llms.txt with your site's name, a summary and curated links, meant to help a language model use your site at inference time, not to guide crawlers.

The evidence that AI search reads it is thin:

• No major AI platform has committed to it. Google says you don't need AI text files to appear in AI Overviews or AI Mode.
• An Ahrefs study published in June 2026 covered 137,210 domains. About 38,000 had a valid llms.txt, and 97% of those got no requests for it in May 2026. Among AI bots, GPTBot and Claude Code requested it most.
• Common Crawl found 584,107 llms.txt files in its July 2026 crawl. About half followed the proposed structure, and a few contained prompt injections.

I add it anyway: it takes an hour, coding agents fetch it, and Chrome's Lighthouse now checks for it in an experimental audit. On liviu.dev, Nuxt SEO's AI Ready module generates it, along with markdown versions of each page. Do it last.

For a SaaS product, the more direct option is an MCP server that lets ChatGPT and Claude use your product, not just read about it. More on that in building an MCP server for your SaaS.

Measuring AI search traffic

No single source is reliable, so combine them:

• GA4. Create a channel group matching chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com and claude.ai. ChatGPT often adds utm_source=chatgpt.com to cited links, but some clicks arrive without a referrer and land in Direct, so treat the numbers as a floor.
• Search Console. AI Overviews and AI Mode clicks count in the Performance report under Web search, and at the time of writing you can't separate them out.
• Bing Webmaster Tools. The AI Performance report, in public preview since February 2026, shows citations in Copilot and Bing's AI summaries and the grounding queries behind them, but not clicks.
• Server logs. OAI-SearchBot, Claude-SearchBot and PerplexityBot hits confirm you're crawled. ChatGPT-User or Claude-User hits mean an assistant fetched the page mid-conversation.
• Your enquiry form. Add "ChatGPT or another AI assistant" to "How did you hear about us?".

A prioritised checklist

1. Get indexed in Google and Bing. Verify the site in both webmaster tools and submit a sitemap.
2. Let search crawlers in. Allow OAI-SearchBot, Claude-SearchBot and PerplexityBot, and check your CDN and firewall.
3. Decide on GPTBot and ClaudeBot deliberately, not by accident.
4. Server-render key pages, and confirm with curl that services, prices and locations are in the HTML.
5. State the key facts plainly: what you do, for whom, where, from what price.
6. Add entity markup with sameAs, and keep NAP consistent.
7. Rewrite top pages for answers: FAQs, comparisons, numbers, sources.
8. Fix Core Web Vitals on pages that bring in leads.
9. Set up measurement across GA4, Bing, logs and your form.
10. Add llms.txt.

Items 1 to 4 are where I most often find problems, and they're the core of my technical SEO audit. Audits start from €900 and include an AI search readiness check. If you'd like me to look at your site, send me the details and I'll reply within one working day.

FAQ

Can I submit my website to ChatGPT so it recommends me?

There's no submission form for ChatGPT's answers. You control access and findability: allow OAI-SearchBot, don't block OpenAI's published IP ranges, and be well indexed in the search engines ChatGPT draws on.

Does blocking GPTBot remove my site from ChatGPT search?

No. GPTBot collects training data and OAI-SearchBot controls ChatGPT search, and OpenAI says the settings are independent. Block GPTBot, allow OAI-SearchBot, and your pages stay eligible for ChatGPT search answers.

Do I need llms.txt to appear in Google AI Overviews?

No. Google says AI Overviews and AI Mode use pages that are indexed and eligible for a snippet, with no special files or markup. llms.txt does no harm, but it won't get you into Google's AI features.

How long does it take for ChatGPT or Perplexity to cite my site?

For search-backed answers, it depends on when your pages are crawled and indexed. OpenAI says robots.txt changes reach its search systems in about 24 hours. Whether you're then chosen depends on the question and the competition, so nobody can promise a date.

Is GEO different from SEO?

Mostly it's SEO with a different finish line. Crawling, indexing, speed and clear content matter for both. GEO adds more crawlers to manage, writing for quotes rather than clicks, and measuring citations, which is why I handle it as part of technical SEO.