Generative Engine Optimization (GEO): What Works in 2026

TL;DR

GEO means making your business easy for AI answer engines to find, understand and cite. The term comes from a 2023 research paper that reported visibility gains of up to 40% on a lab benchmark.
Google and Bing both say normal SEO is the foundation. No secret AI-only switch exists, but access matters: blocking OAI-SearchBot keeps you out of ChatGPT search answers.
Vendor studies suggest that 'best of' lists, comparison pages, review platforms and pages that closely match the buyer's question get cited most. Treat these as vendor data, not settled fact.
Current evidence shows little or no citation lift from llms.txt or schema markup. Keep them as cheap hygiene, not your strategy.
Engines cite different sources and answers change from one run to the next, so measure each engine separately with logged re-runs. Be wary of anyone who guarantees placement.

What is generative engine optimization, and do I need it?

Generative engine optimization (GEO) is the work of making your business easy for AI answer engines like ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Overviews to find, understand and cite. If your buyers ask AI tools who to hire or what to buy, you need it. GEO builds on SEO. It does not replace it.

The practice is older than the name. The term "generative engine optimization" was introduced in a 2023 paper by Pranjal Aggarwal and co-authors, later presented at the KDD 2024 conference. Other terms for the same concept include answer engine optimization (AEO), large language model optimization (LLMO), artificial intelligence optimization (AIO), and AI SEO. I use "GEO" here because Microsoft Bing refers to the practice as "generative engine optimization" in its documentation .

If you want the bigger picture first, read what AI search visibility means, then come back.

Why does GEO matter now?

A lot of people now get their answers from AI, and those answers name only a few businesses. Google says AI Overviews now has over 2.5 billion monthly active users, while AI Mode has surpassed one billion monthly users. (Google)

These tools don't run just one search. Google says both AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources, to develop a response. (Google Search Central) For AI Mode, Google says it divides your question into subtopics and searches for each one simultaneously. (Google Search Help) In plain terms: a page that clearly answers one narrow sub-question can get pulled into an answer even if it doesn't rank for the broad keyword.

Google rank also isn't the whole story outside Google. In a vendor study, Ahrefs reported that 28.3% of ChatGPT's most-cited pages had no Google visibility. That's one vendor's data, but it fits what the platform docs show: each engine has its own crawler and its own way of picking sources.

Where did GEO come from? The 2023 research paper

The short answer: Princeton and IIT Delhi researchers, among others, built a test bench of 10,000 queries and showed that simple content changes could raise a source's visibility in AI answers.

The paper is "GEO: Generative Engine Optimization" (arXiv 2311.09735). The authors describe GEO as a flexible black-box optimization framework for optimizing and defining visibility metrics. To test it, they built GEO-bench, a benchmark consisting of 10000 queries from diverse domains and sources, adapted for generative engines.

What they found:

GEO can boost visibility by up to 40% in generative engine responses.
Including citations, quotations from relevant sources, and statistics were among the methods that helped most (paper PDF).
The efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods.

How I read it: the paper is a lab study. It was done before today's versions of ChatGPT search, Claude web search and AI Mode existed in their current form. Its lasting lesson is about content habits, not tricks. Back up your claims with sources, use real numbers, and quote experts. Those habits are easy to adopt and also make pages better for people.

How do AI engines choose which sources to cite?

The short answer: when an engine searches the web, it pulls a set of pages from its index and writes an answer from them. If its crawler can't reach your page, or your page isn't in the index it uses, it can't cite you. So start by checking access for each engine.

AI engine	What to allow or check	Key fact from the platform
ChatGPT search	OAI-SearchBot	Blocked sites are left out of search answers
Claude	Claude-SearchBot, Claude-User	All three Claude bots honor robots.txt
Google AI Overviews and AI Mode	Googlebot, normal Search indexing	No special optimizations needed
Microsoft Copilot	Bingbot, IndexNow	NOARCHIVE blocks Copilot use
Perplexity	PerplexityBot	Perplexity-User may ignore robots.txt

ChatGPT

OpenAI states it plainly: "Any public website can appear in ChatGPT search." Per OpenAI's crawler docs, OAI-SearchBot is used to surface websites in search results in ChatGPT's search features. Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links. Training is a separate setting: a webmaster can allow OAI-SearchBot in order to appear in search results while disallowing GPTBot to indicate that crawled content should not be used for training (OpenAI). More in how to get found in ChatGPT.

Claude

Anthropic's crawler docs, as reported by Search Engine Journal, list ClaudeBot (training data collection), Claude-User (fetching pages when Claude users ask questions), and Claude-SearchBot (indexing content for search results) as separate bots . Anthropic says all three of its bots honor robots.txt, including Claude-User. Blocking the training bot does not block the others: for Anthropic and OpenAI, blocking the training bot does not block the search bot or the user-requested fetcher. Our desk has also seen Claude's web results line up with Brave Search, so I check whether a site shows up in Brave too. See how to get found in Claude.

Google AI Overviews, AI Mode and Gemini

Google says AI features run on normal Search. Its documentation states that "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." (quoted here; see Google's guide). Google is also beginning to test a new control that lets website owners manage how their links and content appear in generative AI Search features. If you see that setting in Search Console, make sure it is set the way you want. Guides: AI Overviews, AI Mode for local businesses and Gemini.

Microsoft Copilot (and why Bing matters)

Bing's guidelines say SEO fundamentals like crawl efficiency, indexing accuracy, URL consolidation, content clarity and structure, and authority and trust signals also support eligibility for grounding results . Bing also warns about meta tags: NOARCHIVE prevents content from being used in Copilot responses and grounding results , and NOCACHE limits Copilot to using only the URL, title, and snippet, reducing citation depth and answer quality. (Bing Webmaster Guidelines)

Perplexity and Grok

For Perplexity, allow its crawler and keep your pages fast and readable. One difference worth knowing: OpenAI and Perplexity draw a sharper line for user-initiated fetchers, warning that robots.txt rules may not apply to ChatGPT-User and generally don't apply to Perplexity-User. Engine-specific steps are in Perplexity and Grok.

GEO vs SEO: what actually changes?

The short answer: the foundation stays the same, but the goal and the scoreboard change. In SEO you win a ranking. In GEO you win a mention or a citation inside an answer, across several engines that each pick sources differently.

Question	Classic SEO	GEO
What counts as a win?	A high ranking	Being named or cited
How many engines?	Mostly Google	Six or more AI tools
Who controls the inputs?	Mostly your site	Your site plus third-party pages
How stable are results?	Fairly stable	Changes run to run
Best content format	Keyword-targeted page	Page matching a full question

Microsoft's own definition is useful here. Its guidelines describe GEO as activity that improves a page's eligibility to be used as a grounding source or reference within AI answers, without guaranteeing citations. (summary) "Eligibility" is the honest word. For a local-business angle, see GEO vs SEO for local businesses and ChatGPT vs Google for local search.

What kinds of pages get cited most?

The short answer: vendor studies keep pointing to the same few page types. These are "best of" lists, comparison pages, review platforms, and your own pages when they closely match the question. Almost all of this data comes from companies that sell AI visibility tools, so treat it as direction, not law.

Here is what the studies report, in the authors' own framing:

Listicles. Ahrefs (vendor) says "Best X" listicles made up 43.8% of the page types ChatGPT cited. It also says ChatGPT cited about half of the URLs it retrieved.
Comparison and "alternative" pages. AirOps (vendor) looked at 380K answers across 81 B2B categories. It found that ChatGPT, Perplexity and Claude sent 52 to 72% of citations to off-site product and comparison pages and 16 to 21% to review platforms. Omnius (vendor, B2B SaaS) found "best-" in 17.7% of cited URLs, "-alternative" in 5.2% and "-vs-" in 5.1%.
Your own site, when it matches the prompt. In BrightLocal's local study (vendor, relayed on X) of 200,875 local AI searches, 93% of unique cited domains were individual business sites, and those sites got 42% of citations. In one B2B client case, 85% of citations went to third-party pages. But 87% of the answers that mentioned the client cited the client's own site.
YouTube. Ahrefs reported that YouTube mentions had the highest correlation (0.737) with AI brand visibility. That is a correlation, not proof that video causes citations.
Reddit: mixed. AirOps found Reddit in 39 to 52% of Google AI Overview and AI Mode answers, but said Claude rarely cites Reddit or YouTube. Omnius found Reddit in only 0.6% of B2B SaaS citations.

My takeaway: no single page type wins everywhere. Build pages that answer real buyer questions, and get onto the third-party pages each engine already trusts in your category. If a competitor keeps getting named instead of you, this breakdown explains the usual reasons.

What doesn't move citations (based on current evidence)?

The short answer: llms.txt files and schema markup show little or no measurable citation lift in the data available today. Keep them cheap and accurate, and put your real effort elsewhere.

llms.txt. Ahrefs log data shared on X found that 97% of valid llms.txt files across 137K domains got zero requests in May. The same post cites Google saying it doesn't need AI text files. Aleyda Solis shared an experiment calling the evidence "GEO astrology." On the other side, one developer reported seeing Claude fetch theirs. My view: it takes two minutes to add, so add it if you like, but don't pay for it. Details in llms.txt for small business.
Schema as a citation lever. Ahrefs (vendor) reported that schema had no meaningful impact on citations. Schema still helps search engines read your facts and can earn rich results. Keep it correct (see local business schema), but don't expect it to get you cited by itself.
Fake reviews and astroturf. Beyond the risk of getting caught, the FTC finalized a rule in 2024 banning fake reviews and testimonials. Earn real ones.

How do you do generative engine optimization? A 10-step plan

The short answer: make sure the engines can reach you, publish pages that answer real buyer questions, get listed where the engines already look, then measure each engine on its own.

Open the doors. In robots.txt, allow OAI-SearchBot, Claude-SearchBot, Claude-User, Googlebot, Bingbot and PerplexityBot on the pages you want cited. Training bots are a separate choice. Remove NOARCHIVE and NOCACHE from pages you want Copilot to use. Walkthrough: allow AI crawlers in robots.txt.
Get indexed in both Google and Bing. Verify in Google Search Console and Bing Webmaster Tools and submit sitemaps. Bing's guidelines say to use IndexNow to notify Bing when URLs are added, content is updated, or URLs are removed . Bing also advises against "hiding critical content behind client-side rendering" , so make sure your key text appears in the HTML (summary).
Write down your buyer prompts. List 20 to 50 questions in your customers' own words. Examples: "best payroll software for a 10-person company" or "is a heat pump worth it in a cold climate."
Build one page per real question. Answer in the first two sentences, then give the details. Use one section per question, specific numbers and plain HTML tables. That's practitioner advice (here and here), and it fits the GEO paper's finding that statistics, quotations and citations help.
Publish honest comparison and "best" pages. Write "X vs Y," "X alternatives" and "best X for Y" pages with real facts, and include competitors fairly. A one-sided page is easy for readers and engines to spot.
Get onto the pages engines already cite. Run your prompts and write down every cited domain. Then claim or improve your profiles on the directories, review platforms and roundups that show up. For "best agency" type prompts, our desk has seen Claude and Gemini cite directories like Clutch and The Manifest.
Keep your facts identical everywhere. Use the same business name, phone, description, hours and service list on your site, Google Business Profile, Bing Places, Apple Business Connect and every directory. Link your profiles to each other.
Earn real reviews on more than one platform. AirOps (vendor) reported that review platforms received 16 to 21% of assistant citations. Ask every happy customer, and never buy reviews.
Add video where it helps buyers. Short explainers and demos give engines another source to find, and the Ahrefs correlation suggests it's worth testing.
Refresh pages and keep dates honest. Update prices, numbers and screenshots. Change the visible "updated" date only when the content really changed, and ping IndexNow when you do.

How do you measure GEO results?

The short answer: run a fixed set of prompts in each engine on a schedule, log who gets named and what gets cited, and track AI referral traffic in your analytics. Never judge from a single screenshot.

Engines don't agree with each other. A study shared on X reported only 11% source overlap between ChatGPT and Perplexity, and 13.7% between AI Overviews and AI Mode. Results also shift on the same engine. BrightLocal data relayed on X says only 20 to 33% of results overlapped on a repeated search.

Pick your 20 to 50 prompts and freeze the wording.
Run each prompt in ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Mode, logged out or in a clean session.
Log three things: whether you're named, where you appear in the answer, and which URLs are cited.
Repeat on the same schedule. Track how often you appear across runs, not one result.
Watch referrals. OpenAI notes that publishers who allow OAI-SearchBot to access their content can track referral traffic from ChatGPT in standard analytics tools.

Step-by-step guides: check if AI recommends your business and the AI visibility audit checklist.

What are the most common GEO mistakes?

Blocking the search bot when you meant to block training. GPTBot and OAI-SearchBot are separate settings, and so are ClaudeBot and Claude-SearchBot.
Ignoring Bing. Copilot relies on Bing's index, and Bing's guidelines cover Copilot directly.
Writing for keywords instead of questions. AI tools break questions into subtopics. Pages that answer one clear question are easier to use.
Only working on your own site. The vendor data shows a large share of citations go to third-party lists, comparisons and review sites.
Buying "guaranteed" AI placement. Bing's own guidelines say GEO, like SEO, does not provide any guarantees of rankings or citations.
Treating llms.txt or schema as the whole plan. Both are fine as hygiene. Neither shows clear citation lift in current data.

Should you do GEO yourself or hire help?

The short answer: steps 1, 2, 7 and 8 are doable for most owners in a few afternoons. The ongoing parts are prompt research, building comparison content, getting onto third-party lists and logged measurement. Those take steady time, and that's where outside help tends to pay off.

If you're weighing help, read how to choose an AI search optimization agency and what AI search optimization costs. Industry examples: accountants and veterinarians. You can see how we work, our services, the overview of Get Found with AI, or browse the blog.

How I put this together

On October 22, 2026, I checked the top three US results for "generative engine optimization":

HubSpot: a broad, recently updated guide ( Updated: 09/14/26 ) covering the definition, GEO vs SEO, tactics, tools and mistakes. It lists structured data markup as something engines favor, but it doesn't mention the vendor data showing no clear citation lift from schema.
Built In: an expert opinion piece, updated by Brennan Whitfield on Aug 13, 2025 . It's useful for framing but thin on platform docs and over a year old.
Mangools: a definition plus per-engine tips. It still names ChatGPT, Perplexity, and Bard , an outdated product name, and its per-engine advice doesn't cite platform sources.

All three explain what GEO is. None of them checks crawler access against OpenAI's, Anthropic's and Bing's own documentation. None covers the llms.txt log evidence, the low source overlap between engines, or the run-to-run variation that makes single screenshots misleading. None labels its data as vendor data, and none gives a repeatable way to measure. I added those pieces. I also verified the paper's claims against the arXiv original and flagged every vendor study as a vendor study. I can't promise any placement, and nothing here should be read as a promise of results.

Frequently asked questions

What is generative engine optimization in simple terms?

It is the work of making your business easy for AI tools like ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Overviews to find, understand and cite when people ask questions in your category. It builds on SEO rather than replacing it.

Is GEO different from SEO?

It sits on top of SEO. Google says there are no special optimizations needed for AI Overviews or AI Mode, and Bing says SEO fundamentals support eligibility for AI answers. What changes is the goal (being named and cited, not ranked), how many engines you track, and how much third-party pages like reviews and 'best of' lists matter.

Does llms.txt help with generative engine optimization?

Current evidence says not much. Ahrefs log data, shared on X, found that 97% of valid llms.txt files got zero requests in a month, and Google has said it does not use such files. One developer reported Claude fetching theirs. It is cheap to add, but I would not spend real time on it.

Does schema markup help me get cited by AI?

Ahrefs, a vendor, reported that schema had no meaningful impact on citations. Schema still helps search engines read your business facts and can earn rich results, so keep it accurate. Just don't treat it as a citation lever.

Can anyone guarantee ChatGPT or Google AI will recommend my business?

No. Microsoft's Bing guidelines say GEO, like SEO, does not guarantee citations. Answers also change from one run to the next. I can't promise any placement, and you should be careful with anyone who does.

How do I check if AI tools recommend my business?

Write 20 to 50 questions your buyers actually ask. Run each one in ChatGPT, Claude, Gemini, Grok, Perplexity and Google AI Mode. Log who gets named and which sources get cited. Repeat the same prompts on a schedule, because a single screenshot is not a reliable result.

Generative Engine Optimization: What It Is and What Actually Works