Generative Engine Optimization (GEO): What Actually Works

How to get cited by AI search — Google AI Overviews, ChatGPT search, Perplexity — based on what the research and the platforms' own documentation say, not guesses. Crawlers, content, structure, and what doesn't work.

AAAayush AdhikariSeptember 24, 2026 7 min read

Generative engine optimization (GEO) means making your pages more likely to be used and cited when AI search engines — Google's AI Overviews and AI Mode, ChatGPT search, Perplexity — write an answer. The evidence points to a short list: let the AI search crawlers reach your pages, be eligible for normal search, answer questions directly, and make content verifiable with statistics, quotations and cited sources. Google says no special markup or files are needed; a peer-reviewed study found keyword stuffing does little, while citations and numbers help most.

What GEO is, and what it isn't

Traditional SEO is about ranking in a list of links. Generative engines retrieve several pages, then write a synthesized answer that cites some of them. "Visibility" becomes: is your page among the sources, and how much of the answer comes from it?

The term comes from a 2023 paper, GEO: Generative Engine Optimization (Aggarwal et al., accepted at KDD 2024), which built a benchmark of queries and measured how content changes affected a page's visibility in generated answers. Its headline: GEO methods can boost visibility by up to 40% in generative engine responses.

GEO isn't a replacement for SEO. The same pages still need to be crawlable, indexable and genuinely useful. It's a set of content and access decisions on top.

What the research found

The GEO study tested content rewrites on its benchmark. Reported results for the best methods, as relative improvement in how much of the answer drew on the page:

  • Adding quotations from relevant sources — around 40%.
  • Adding statistics — around 33%.
  • Citing sources — around 28%.
  • Fluency optimization (clearer writing) — around 29%.
  • Keyword stuffing — little to no improvement.

Two findings matter for small sites:

  1. Lower-ranked pages gained most. The paper reports that citing sources lifted visibility of pages ranked fifth in the underlying search results by about 115%, while pages ranked first lost visibility with the same method. GEO partly levels the field.
  2. Effects vary by domain. What helps for history questions differs from what helps for technical ones. Treat the numbers as direction, not guarantees.

The practical reading: make claims checkable. A sentence with a number and a source is more usable to an answer engine — and to a human — than the same claim without them.

What the platforms say

Google AI Overviews and AI Mode

Google's documentation is explicit: there are no additional requirements to appear in AI Overviews or AI Mode, and no special optimizations are necessary. A page must be indexed and eligible to be shown with a snippet in normal Search. Google also says you don't need new machine-readable files, "AI text files", or special schema.org structured data for these features. The usual preview controls — nosnippet, data-nosnippet, max-snippet, noindex — apply to AI features too.

So for Google, GEO is SEO: helpful content, crawlable pages, snippets allowed. Google's own guidance on helpful, people-first content still applies, and its spam policies name scaled content abuse — mass-producing pages to manipulate rankings — as a violation, whatever tool produced them.

OpenAI documents three crawlers:

  • OAI-SearchBot — surfaces websites in ChatGPT's search results. OpenAI recommends allowing it in robots.txt (and allowing its published IP ranges) to appear in search results.
  • GPTBot — used for training OpenAI's foundation models.
  • ChatGPT-User — visits pages when a user's question prompts it.

Blocking GPTBot (training) doesn't block OAI-SearchBot (search). You can choose separately.

Perplexity

Perplexity documents PerplexityBot, which surfaces and links websites in Perplexity's results and isn't used for training, and recommends allowing it in robots.txt to appear in results. Perplexity-User fetches pages for a user's question.

A practical GEO checklist

1. Access

  • Allow OAI-SearchBot, PerplexityBot and ordinary search crawlers in robots.txt. If your rule for * already allows them, naming them explicitly protects against someone tightening * later. That's what LetRelay's blog does.
  • Don't hide content behind JavaScript that crawlers may not execute. Server-rendered HTML is safest.
  • Make sure pages allow snippets — a nosnippet directive also keeps you out of Google's AI features.

2. Discovery

  • A sitemap with accurate lastModified dates — the date of a real edit, not the date of the last deploy.
  • For Bing and other participating engines, IndexNow lets you announce new or changed URLs immediately; up to 10,000 URLs per submission, verified by a key file on your site.

3. Content that answers

  • Answer first. Put a direct, complete answer to the page's main question in the first paragraph, then the detail. Answer engines extract passages; make the best passage easy to find.
  • One question per heading. Clear H2/H3 structure lets both people and retrieval systems find the right section.
  • Numbers with sources. State the figure, name the source, link to it.
  • Quotations from primary sources where they add authority — sparingly, and within copyright norms.
  • A FAQ section in plain question-and-answer form for the questions people actually ask.
  • Original information. Your own measurements, examples and lessons are what no other page has. On LetRelay's blog, posts include numbers we measured ourselves — latencies, token budgets, test results — alongside documented limits; our free AI gateway write-up is an example.

4. Trust signals

  • A named author with a real bio.
  • Accurate dates: published and last updated.
  • Structured data that matches what's visible (BlogPosting, BreadcrumbList). Google says it's not required for AI features, but it helps search engines understand the page and costs little.

What doesn't work

  • Keyword stuffing. The GEO study found little to no benefit.
  • Mass-produced pages. Publishing hundreds of thin pages is exactly what Google's scaled-content-abuse policy targets. Fewer, substantive pages published at a steady pace is safer and more useful.
  • Special "AI files" as a shortcut. Google says they're unnecessary for its AI features. Other engines may read extra files, but no documentation we've found says they replace crawlable, useful pages.
  • Invented statistics. A made-up number is worse than none: it's the kind of claim answer engines and readers can check.

Measuring GEO

There's no GEO equivalent of rank tracking yet, but you can watch:

  • Referral traffic from AI search products in your analytics (they often send a recognizable referrer).
  • Crawler hits from OAI-SearchBot and PerplexityBot in your server logs — no hits means no chance of being cited.
  • Manual checks: ask the questions your posts answer in each engine every month and note whether you're cited.
  • Search Console: Google reports AI Overviews traffic as part of normal Search performance.

How we apply this on LetRelay's blog

Every post is held to a linter before it's published: a direct answer in the opening paragraph, a FAQ, a Sources section with working links, internal links to related posts, and a meta description of the right length. New posts are scheduled a few per week rather than published at once. Robots names the AI search crawlers, the sitemap uses real edit dates, and IndexNow announces changes. None of it is a trick; it's making good pages easy to find and easy to trust. The technical side of building the blog is in a free-tier SaaS architecture.

FAQ

What is generative engine optimization?

Optimizing content so AI search engines — which write answers from several sources — are more likely to use and cite it. The term comes from a 2023 research paper that measured which content changes improved visibility in generated answers.

Do I need special markup to appear in Google AI Overviews?

No. Google says there are no additional requirements or special optimizations: the page must be indexed and eligible for a snippet in Search, and normal SEO best practices apply.

How do I get my site into ChatGPT search results?

OpenAI recommends allowing OAI-SearchBot in your robots.txt and allowing requests from its published IP ranges. Blocking GPTBot, the training crawler, doesn't affect search.

The GEO study found keyword stuffing gives little to no improvement. Adding statistics, quotations and cited sources performed much better.

Sources

AA
Aayush Adhikari

Building Relay — the internal request desk with AI triage and SLA tracking.

Run your internal requests on LetRelay

AI triage, SLA-tracked queues, and bottleneck analytics — the help desk your team actually likes. Free to start.

Try LetRelay free No credit card required
Ad spaceYour Google AdSense unit shows here once approved.

Keep reading