Generative Engine Optimization (GEO): What It Is, What Works, and What's Hype

Generative engine optimization (GEO) is the practice of making content more likely to be cited, quoted or named inside answers written by AI systems such as ChatGPT, Perplexity, Gemini and Google's AI Overviews. Where traditional SEO competes for a position in a list of links, GEO competes for a mention inside a synthesised answer. The term comes from a 2023 research paper, and only a small part of what is now sold as GEO is supported by evidence.

Key takeaways

• GEO was named by an academic paper (Aggarwal et al.), which tested nine content changes on 10,000 queries. Adding quotations, statistics and cited sources raised visibility by roughly 27–41%. Keyword stuffing made it worse.
• Those gains were largest for pages that ranked poorly — and negative for pages already ranked first.
• Google's own May 2026 guidance says optimising for generative AI search "is optimizing for the search experience, and thus still SEO."
• llms.txt and special schema markup have no measured effect on AI citations. Google says its search doesn't use llms.txt at all.
• AI engines draw heavily on what other sites say about you. One study found 72.7–81.9% of sources cited by GPT were third-party "earned" media.
• AI recommendations are highly unstable. In repeated testing, the same list of brands appeared in fewer than 1% of runs. Any tool claiming to show your "AI ranking" is overstating what can be measured.

Short on time? If you want to know how AI assistants currently describe your business — or whether they mention you at all — we'll run the prompts and show you the real answers on a free strategy call. Book it here.

Where the term comes from

GEO was defined in a paper by Pranjal Aggarwal and colleagues, "GEO: Generative Engine Optimization" (arXiv 2311.09735), published in 2023 and presented at KDD 2024. It is, as of late 2026, still the only peer-reviewed study that directly tests content changes against generative-engine visibility.

The researchers built a benchmark of 10,000 queries (GEO-bench), applied nine different content modifications to source pages, and measured how much of each AI-generated answer was attributed to the modified source. Their headline claim — that GEO "can boost visibility by up to 40%" — is the origin of a number now quoted, usually without context, across a great deal of marketing material.

What the research actually measured

The table below shows the paper's results for its main metric, position-adjusted word count — roughly, how much of the generated answer came from your page, weighted by where in the answer it appeared.

Content change	Visibility change	What it means
Adding quotations from relevant sources	+41%	Strongest single effect measured
Adding statistics	+31%	Specific numbers rather than vague claims
Improving fluency	+28%	Clearer, better-written prose
Citing sources	+27%	Referencing where claims come from
Keyword stuffing	−8%	The classic SEO tactic made results worse

Source: Aggarwal et al., Table 1, arXiv 2311.09735. Percentages calculated from the paper's absolute scores.

The pattern is not mysterious. Quotations, statistics and citations are the hallmarks of well-sourced writing. A language model assembling an answer is more willing to lean on a passage that contains something specific and attributable than on one that only asserts.

The finding most GEO marketing leaves out

The paper also broke results down by how well each page already ranked. The effect reverses at the top.

Content change	Page ranked 1st	Page ranked 5th
Citing sources	−30.3%	+115.1%
Adding quotations	−22.9%	+99.7%
Adding statistics	−20.6%	+97.9%

Source: Aggarwal et al., Table 2.

In other words, these techniques roughly doubled the visibility of lower-ranked pages while reducing the visibility of pages already ranked first. GEO, as measured, is primarily a tool for pages that are not yet winning — which makes it more relevant to newer and smaller businesses than to established ones.

What this research does not prove

It is worth being exact about the limits, because the "40% boost" figure is routinely misused:

1. It was a simulated engine. The researchers built their own generative system and also tested on Perplexity. It was not a measurement of ChatGPT, Gemini or Google AI Overviews in production.
2. It measured share of an answer, not whether you get named. The metric captures how much of an answer is drawn from a page that was already retrieved. It does not show that these changes get a business mentioned when it otherwise wouldn't be.
3. It has not been replicated on 2026 production AI systems in any peer-reviewed study we could find.

So the defensible reading is narrow: writing that contains specific statistics, attributed quotations and cited sources is more quotable to an AI system, especially for pages that don't already rank first. That is useful. It is not a guarantee of anything.

This is the work we do. If your service pages are full of adjectives and short on specifics, that is usually the first thing to fix — and it helps human buyers as much as AI systems. See our GEO service.

What Google says about GEO

Google published guidance on AI search in May 2026. Its position is unambiguous: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO" (Google Search Central).

The same guide states plainly that you do not need special files or markup:

"You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." — Google Search Central

Google's AI Overviews and AI Mode are built on top of its core ranking systems, using a technique it calls "query fan-out" — generating a set of related searches behind the scenes and assembling an answer from what they return. That is why the foundations of SEO — being crawlable, indexed and genuinely useful — are also the foundations of appearing in Google's AI answers.

What does not work, despite being widely sold

llms.txt

llms.txt is a proposed plain-text file that summarises a website for AI systems. It is frequently sold as a GEO essential. The evidence says otherwise:

• Google: its search systems, including AI features, do not use it (Google Search Central).
• John Mueller of Google, June 2025: "FWIW no AI system currently uses llms.txt," noting that server logs show AI crawlers fetching pages but none fetching the file (SERoundtable).
• SE Ranking analysed roughly 300,000 domains and found no relationship between having an llms.txt file and how often a site was cited by AI. Removing the variable from their model actually improved its accuracy (SE Ranking).
• No AI company — OpenAI, Anthropic, Perplexity or Google — documents reading it in their crawler or publisher guidance.

Having one does no harm. Paying for one, or being told it is a meaningful lever, is a different matter.

Schema markup as an AI-citation lever

Structured data is valuable for some search features. As a way of getting cited by AI, the best available test found no benefit. Ahrefs compared 1,885 pages that added JSON-LD schema against roughly 4,000 matched pages that didn't, before and after. AI Overview citations went down 4.6%; AI Mode and ChatGPT moved within noise (summary of the Ahrefs study). Google's guidance agrees: "there's no special schema.org markup you need to add."

Keyword stuffing

The original GEO paper measured it directly: −8% visibility. Repeating a target phrase is a dead tactic in traditional search and actively counterproductive here.

"Your AI ranking"

This is the claim most worth being sceptical of. Researchers at SparkToro and Gumshoe ran 2,961 prompts repeatedly across ChatGPT, Claude and Google's AI — 60 to 100 runs per prompt. The same list of recommended brands appeared in fewer than 1% of runs, and the same list in the same order in fewer than 0.1% (Search Engine Journal). Rand Fishkin's summary: "any tool that gives a 'ranking position in AI' is full of baloney."

AI recommendations are not a stable ranked list. The only honest measurement is a rate: across many repeated runs of many relevant prompts, how often is your business mentioned, and is it described accurately?

What the evidence says actually matters

1. Being crawlable by the right bots

This is the one gate that is genuinely binary. AI companies separate their crawlers by role, and blocking the wrong one removes you from search answers entirely.

Company	Search crawler (allow this)	Training crawler (separate choice)
OpenAI	OAI-SearchBot	GPTBot
Anthropic	Claude-SearchBot, Claude-User	ClaudeBot
Perplexity	PerplexityBot	—
Google	Googlebot	Google-Extended

Per each company's own documentation, blocking the search crawler removes you from that assistant's search-based answers; blocking the training crawler is an opt-out from model training that, according to the vendors, does not affect search citation (OpenAI, Anthropic, Google).

Check yourdomain.com/robots.txt, and check your CDN or hosting provider's bot settings. Several providers ship "AI bot" blocking that some site owners have switched on without realising.

2. What other people say about you

This is the biggest lever and the one most businesses neglect, because it isn't on their own website.

• A University of Toronto preprint found AI search shows a "systematic and overwhelming bias" toward earned third-party media: 72.7–81.9% of sources cited by GPT were earned media, compared with 45–54% for Google (arXiv 2509.08919).
• Ahrefs analysed 75,000 brands and found branded web mentions correlated with AI Overview visibility at 0.664 — more than twice the correlation for raw backlinks (0.218) (Ahrefs). That study was limited to established domains, so treat it as directional for newer sites.

Your own site is one source, and a self-interested one. Independent reviews, directory profiles, press mentions, community discussion and being included in genuine comparisons are what AI systems weigh most.

3. Quotable, specific, sourced content

This is where the GEO paper's findings apply directly. A page that says "we deliver great results" gives an AI system nothing to cite. A page that says a client's cost per lead fell from $716.65 to $90.23, and explains how, gives it something concrete.

In practice:

• Lead every important section with a self-contained answer that makes sense out of context.
• Replace adjectives with figures, and say where each figure came from.
• Quote and link to authoritative sources rather than paraphrasing them vaguely.
• Write so any paragraph stands alone, because an AI system will often read it in isolation.

Want this applied to your site? A free strategy call includes a look at how quotable your current pages are, and what an AI assistant says when asked about your category. Book it.

4. Consistency across every source

AI systems reconcile information from many places. A business described differently on its website, its Google Business Profile, its directory listings and its social profiles is harder to treat as a single entity than one described identically everywhere. Use the same name, the same one-line description and the same contact details wherever the business appears.

How GEO relates to SEO and AEO

	SEO	AEO	GEO
Competes for	A position in a list of links	Being the extracted answer	A mention inside a synthesised answer
Main surfaces	Google results pages	Featured snippets, AI Overviews	ChatGPT, Perplexity, Gemini, AI Overviews
Measured by	Rank tracking	Snippet and AI Overview appearances	Mention rate across repeated prompts
Precision of measurement	High	Medium	Low — sampling only

They are layers, not alternatives. Google's guidance treats GEO as part of SEO, and the evidence supports that: a page that cannot be crawled, indexed and trusted has nothing for an AI system to cite. For a fuller comparison, see SEO vs AEO vs GEO. For the practical steps to getting recommended by name, see how to get recommended by ChatGPT and Perplexity.

How to measure GEO honestly

Because AI answers vary so much from run to run, measurement has to be statistical:

1. Write down 20–30 prompts a real buyer in your category might use.
2. Run each prompt several times across ChatGPT, Perplexity, Gemini and Claude.
3. Record, for each run: were you named, how were you described, and was the description accurate?
4. Report the mention rate, not a position — and repeat monthly.

Google also launched generative AI performance reporting in Search Console in June 2026, showing impressions from AI features for sites where it has rolled out. It does not report clicks, and it does not cover other AI assistants.

Frequently asked questions

What is generative engine optimization in simple terms?

Generative engine optimization is the work of making your business more likely to be mentioned or quoted when someone asks an AI assistant a question. Instead of ranking in a list of links, the goal is to appear inside the answer the AI writes. It overlaps heavily with good SEO and good writing.

Is GEO different from SEO?

Partly. Google's own guidance says optimising for its generative AI search "is still SEO", because AI Overviews are built on the same ranking systems. GEO adds emphasis on quotable, specific content and on what third-party sources say about you, and it is measured by sampling AI answers rather than by rank tracking.

Does llms.txt help with GEO?

There is no evidence that it does. Google says its search systems don't use it, a Google search advocate said no AI system currently uses it, and a study of around 300,000 domains found no relationship between having the file and being cited by AI. It does no harm, but it is not a lever.

How long does GEO take to work?

Expect months rather than weeks. AI systems rely on what is already indexed and on what other sites say about you, and both take time to build. Crawler access fixes can have an effect quickly; earned mentions and content improvements accumulate slowly.

Can GEO guarantee my business gets recommended by ChatGPT?

No, and anyone offering a guarantee is overstating what is possible. AI answers vary from one run to the next — in one large test the same brand list repeated in fewer than 1% of runs. What can be improved is the likelihood of being mentioned, measured as a rate across many prompts.

Is GEO worth it for a small business?

It depends on how your customers find you. If they research and compare providers before getting in touch, a growing share of that research happens in AI assistants and GEO is worth the effort. If most customers arrive through local map searches or referrals, local SEO and your Google Business Profile usually deserve attention first.

---

Want to know what AI assistants say about your business right now? We'll run the prompts and show you the real answers on a free strategy call — including when the honest answer is that you don't come up yet, and what would change that.

Book a free strategy call · Our GEO service · Our AEO service