Generative engine optimization (GEO): the complete guide

Generative engine optimization (GEO) is the practice of getting your business named and your pages cited in answers written by AI: ChatGPT, Claude, Perplexity, Microsoft Copilot and Google’s AI Overviews and AI Mode. It comes down to four things: let the AI search crawlers in, write content an AI can quote accurately, make sure the wider web describes you consistently, and measure with a fixed set of real prompts.

What is generative engine optimization?

The term comes from “GEO: Generative Engine Optimization” by Pranjal Aggarwal (IIT Delhi), Vishvak Murahari, Karthik Narasimhan and Ameet Deshpande (Princeton) and Tanmay Rajpurohit and Ashwin Kalyan, first posted on arXiv in November 2023 and accepted at KDD 2024. They defined a generative engine as a system that retrieves sources and writes one answer from them, and GEO as the work of making your content more visible inside that answer.

That distinction matters for what you can influence. An AI answer can draw on:

• Model memory. What the model absorbed in training. You can only shape this slowly, through what the public web says about you, and you cannot check it on demand.
• Live retrieval. Pages fetched at answer time through a search index or a crawler (ChatGPT search, Claude’s web search, Perplexity, Copilot, Google’s AI features). This is where most of the practical GEO work pays off, because it depends on things you control: access, indexing and the page itself.

GEO overlaps heavily with answer engine optimization, which focuses on the answer format on your own page. See the complete AEO guide for that side, and AEO vs GEO vs SEO for how the three fit together.

What did the GEO research actually find?

The paper is worth knowing in detail because it is cited constantly, and often loosely. The authors built GEO-bench, a set of 10,000 queries across many domains, and tested nine ways of rewriting a source page, measuring how much of the generated answer drew on that source.

The authors single out quotations, statistics and citations: the best methods improved the main visibility metric by about 30 to 40% over the baseline, while keyword stuffing scored below the page it started from, and a test on Perplexity showed gains of up to 37%. The most interesting finding for smaller sites: the gains were largest for sources that ranked low in the retrieved results. Adding citations raised visibility for the fifth-ranked source by 115% while slightly lowering it for the top source.

Keep the limits in mind before treating this as a recipe:

• It measured how much of an answer drew on a source, not clicks, leads or sales.
• Most tests ran on a simulated engine built on 2023-era models; today’s assistants retrieve and rank differently.
• The effect varied by domain, which the authors themselves stress.

The practical reading holds up anyway: pages with specific, sourced, quotable facts give a writing model something to use. Pages full of generic claims do not.

How do AI assistants find the pages they cite?

Each company publishes the crawlers it uses, and they split by purpose. Getting this table right is the single most important technical step in GEO.

Sources: OpenAI crawlers, Anthropic crawlers, Perplexity crawlers, Google’s common crawlers. For Google’s AI Overviews and AI Mode, the crawler is plain Googlebot: a page needs to be indexed and eligible for a snippet, as covered in how to show up in AI Overviews and AI Mode.

How do you do GEO, step by step?

Step 1: Let the AI search crawlers in

Check your robots.txt and, just as important, your CDN or firewall. Cloudflare, for example, offers settings that block AI crawlers, and a site can block them without anyone on the team realizing it; see is Cloudflare blocking ChatGPT and Claude? The free AI crawler checker shows which bots your robots.txt allows. A common setup that stays visible in AI search while opting out of training:

# AI search and user fetches: allowed
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: Claude-SearchBot
User-agent: Claude-User
User-agent: PerplexityBot
Allow: /

# Training crawlers: blocked
User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
Disallow: /

User-agent: *
Allow: /

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

Whether to block training is a business decision; nothing published shows that allowing training crawlers improves citations in live answers. What is documented is the other direction: OpenAI states that “sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers.”

Step 2: Make your pages easy to quote

1. Answer the question in the first sentences under a question-style heading (the full method is in the AEO guide).
2. Add concrete numbers, each with its source and date. This is the “statistics addition” the GEO paper tested.
3. Cite primary sources with links, especially for claims a model might otherwise doubt.
4. Quote named people where it adds authority: your licensed engineer, a published study, a regulator.
5. Use tables for comparisons and specs. Facts in rows are easier to lift correctly than facts spread across paragraphs.
6. Keep key facts in HTML text, not only in images, PDFs or scripts that need a browser to run.

Step 3: Make the web agree about who you are

AI systems describe a business by stitching together what many sources say. If your site calls you a “digital solutions partner,” your LinkedIn says “marketing agency” and your directory listings have an old address, the description an assistant writes will be vague or wrong. Fix the basics:

• One plain sentence that says what you do, for whom and where, used on your homepage, about page and every profile.
• Organization or LocalBusiness structured data with sameAs links to your official profiles. Check it with the schema checker.
• Complete, matching Google Business Profile and Bing Places listings if you serve customers locally.
• An about page with the people, credentials and history behind the business.

Step 4: Earn mentions where AI assistants look

Ask ChatGPT “what is the best [your category] in [your city]” and look at the sources it cites: usually review sites, industry directories, “best of” roundups, forums and news. That is where you need to be named. An Ahrefs analysis of 75,000 brands found that branded web mentions had the strongest correlation (0.664) with how often a brand appeared in AI Overviews, ahead of branded anchors and branded search volume. That is a correlation in one vendor’s data, not proof of cause, but it matches what practitioners see.

Legitimate ways to earn those mentions: get listed in the directories that matter for your industry, ask real customers for reviews on the platforms your category uses, pitch genuinely useful data to journalists, and answer questions in communities where you have real expertise. SEO Rank Fixer’s “Get listed” checklist gives each site a list of trusted places to list itself, matched to its type, and checks which listings are live. Never buy reviews or post fake ones; it breaks platform rules and consumer protection law, and AI systems repeat what they find.

Step 5: Decide on llms.txt with clear eyes

llms.txt is a proposed file that lists your key pages for language models. It is cheap to add, but there is little evidence major assistants rely on it, and Google says its AI features need no special AI text files. Read the evidence in does llms.txt actually work? before spending time on it; if you want one, the free llms.txt generator builds a starting file.

How do you measure GEO?

There is no single “AI rankings” report. Combine these:

1. A fixed prompt set. Write 20 to 50 questions real customers ask, in their words, including “best [category] for [need]” and “[your brand] vs [competitor]”. Run them with web search on and record: were you mentioned, were you cited with a link, which competitors were named, which URLs were cited.
2. Repeat on a schedule. AI answers vary between runs and users, so one check proves little. Look at the share of runs that mention you over weeks.
3. Referral traffic. In analytics, watch referrals from chatgpt.com, claude.ai, perplexity.ai and copilot.microsoft.com. OpenAI says ChatGPT search adds utm_source=chatgpt.com to its links (Publishers and Developers FAQ); in-app and copied links can still arrive as direct traffic.
4. Search engine AI reports. Bing Webmaster Tools’ AI Performance report shows Copilot citations and grounding queries; Search Console’s Search Generative AI report shows impressions in Google’s AI features. Both are covered in the AEO guide.

SEO Rank Fixer automates the first two steps for ChatGPT and Claude: save your customers’ questions, and the AI visibility check asks both assistants with live web search, then shows whether they name or cite you and who they point to instead (Pro plan; see pricing). It does not track Perplexity, Gemini or Google’s AI Overviews yet. If you need those engines today, our comparisons of Otterly and Profound lay out the trade-offs honestly.

What is still unproven in GEO?

A lot is sold as GEO that nobody has shown to work. Treat these with care:

• “AI rankings” as a stable number. Answers change between runs, phrasing and users. A tool that reports one fixed position is simplifying.
• Special files and markup. No major assistant has said llms.txt, special schema or “AI-only” pages improve citations. Google says none are needed for its AI features.
• Writing for bots. Pages of keyword variants or hidden text did not help in the GEO paper and risk spam penalties in the search indexes assistants depend on.
• Shortcuts into training data. You cannot pay or submit your way into a model’s memory. Consistent, widely repeated facts about you are the only lever, and it is slow.

What does a 30-day GEO plan look like?

For the specific question of getting recommended in ChatGPT, the ChatGPT recommendation guide goes deeper, and getting cited in Perplexity, Claude and Copilot covers the differences between those engines.