What Is GEO? What the Evidence Actually Says About AI Citations

1. Blog
2. Discoverability & Content

By Saroj Jha · July 28, 2026 · 12 min read

GEO is optimizing content to be cited by AI answer engines. But the famous ‘40% visibility lift’ measured something narrower than most people think. What 45 studies and Google’s own guidance actually establish, and what to do about it.

Recognition vs discovery

3.32%

discovery, against 99.4% brand recognition

Being known is not being recommended.

If you have read anything about GEO in the last two years, you have met the number: optimizing for AI increases visibility by up to 40%. It comes from a real, peer-reviewed paper. It also measured something far narrower than the way it gets quoted, and a July 2026 survey of 45 GEO studies, a preprint, files the general version of that claim under one verdict: rejected as a general claim ^[1].

This post is what the evidence actually supports, what it doesn’t, and what a Seed–Series B team should do on Monday.

Generative engine optimization is the practice of structuring content so AI answer engines such as ChatGPT, Perplexity, and Google’s AI Overviews and AI Mode cite it inside the answers they generate. The goal shifts from holding a position in a list of links to being named inside a synthesized answer.

Where the “40% more visibility” number came from

The foundational work is GEO: Generative Engine Optimization ^[5] by Aggarwal and colleagues, published at ACM KDD 2024. It’s a good paper. It named the problem and built the first benchmark, GEO-bench: 10,000 queries, with nine rewriting strategies tested against an unedited control.

Here is the experimental setup, which is the part that matters. For each query, the top five Google results were placed into the model’s context, one of them was rewritten under a given strategy, GPT-3.5 wrote the answer, and the paper measured how much attributed text each source received.

Adding quotations moved the position-adjusted word count, the source’s share of the answer weighted toward the top, from 19.3 to 27.2: a relative gain of about 41%. Adding statistics (+31%) and citing sources (+27%) produced smaller gains. Keyword stuffing made things worse, dropping the metric to 17.7 (−8%)^[5].

So the finding is real, and it is this: a document already sitting in the model’s context can be rewritten to claim a larger share of the answer. It is not a finding about whether your page gets retrieved in the first place. No clicks, referrals or conversions were observed anywhere in the study.

There is a second detail almost nobody quotes. The five sources’ shares sum to 100%, which makes the metric redistributive: one source’s gain is another’s loss. Under the cite-sources strategy, the fifth-ranked source gained 115% while the first-ranked lost 30%^[5]. GEO, measured this way, is closer to a zero-sum game than a rising tide.

Google’s position: it’s still SEO

On 15 May 2026, Google published its first official guidance on the subject, Optimizing your website for generative AI features on Google Search ^[6], announced^[7] by John Mueller on the Search Central blog. It addresses AEO and GEO by name and says, plainly, that from Google Search’s perspective optimizing for generative AI search is optimizing for the search experience, and thus still SEO.

The guide also quietly retires several tactics that GEO vendors sell^[6]:

• llms.txt: Google says Google Search ignores these files, so making one will neither help nor harm your visibility in Search.
• Content chunking: there is no need to break articles into tiny pieces. Google says its systems understand several topics on one page and show the relevant piece.
• Special schema or Markdown versions of pages: structured data isn’t required for generative AI features, and there is no special schema.org markup to add.
• Third-party AI-visibility tools: Google notes that no third-party tool has access to its internal ranking or AI systems.

None of this makes structured data worthless: Google still recommends it for rich results, other engines parse it, and it costs little. It does mean that if someone is selling you an AI-visibility program built on llms.txt and schema, the vendor with the most direct knowledge of the system has publicly said those aren’t the lever.

What actually holds up: relevance and position

Strip away the tactics and two factors dominate the literature.

A factorial experiment by Vishwakarma and colleagues at Sprinklr (What Gets Cited, peer-reviewed at ACM SIGIR 2026) ran 252,000 trials across six models and eighteen content factors ^[8]. Topical relevance and the source’s position in the list were the biggest drivers of which source gets cited first. Separate work found that moving a source higher in the context beats rewriting it: in C-SEO Bench, putting a page first in the model’s context lifted its rank by 0.7 to 2.8 places on average, while the best rewrite managed 0.36 at most^[2].

Read that again, because it reorders the whole to-do list. The two strongest levers are being relevant to the question and being retrieved in the first place. A perfectly optimized page that never enters the model’s context contributes nothing at all.

Which makes the unglamorous work decisive: can an engine crawl your page, does the content exist in the raw HTML without JavaScript executing, is the page actually indexed, and does it directly answer a question someone actually asks? Plenty of sites that “don’t show up in AI answers” fail at one of those four before any content tactic becomes relevant.

The finding that should stop you rewriting pages for AI

Here is the most useful result of the last year, and the one least likely to appear in a GEO sales deck.

The SAGEO Arena ^[3] study (Kim and colleagues at Yonsei University, peer-reviewed at ACM KDD 2026) rebuilt the full pipeline, from a crawled corpus through retrieval and reranking to generation, across 171,003 documents and 2,700 queries in nine domains. When pages were optimized for citation in the body only, average presence in the top 20 fell by about 9%, top-10 presence after reranking fell 16%, and final citation fell 6%.

Rewriting a page to be more citable made it harder to retrieve. And since retrieval comes first, the total effect was negative. Rewriting only the title, meta description, headings and schema went the other way: top-20 presence rose 22%, though final citation rose just 2%^[3].

This resolves an apparent contradiction in the field. Fixed-context studies measure what happens once you’re already in the answer. End-to-end studies measure what happens across the whole pipeline. An intervention can win the first and lose the second.

The generalizability evidence points the same way. C-SEO Bench ^[2] (Puerto and colleagues, peer-reviewed in the NeurIPS 2025 Datasets & Benchmarks proceedings) tested ten rewriting methods, eight of them from the GEO paper, across two tasks and six domains. On GPT-4o-mini, only three of 54 method–domain tests significantly improved a page’s rank, and none in question answering. Several transformations reduced rank. Gains also shrank as more competitors adopted the same tactics, which is what you would expect from a redistributive metric.

There is no single AI ranking

Audits consistently find that engines cite different source ecosystems entirely.

• Nearly 30% of the domains cited by Google’s AI Overviews do not appear in the first-page results shown alongside them, which points to a selection step separate from Google’s own ranking. That comes from 55,393 trending queries over 40 days in spring 2026^[9].
• In a benchmark of 11,500 real user queries collected in December 2025, the sources retrieved by organic Google, AI Overviews and Gemini 2.5 Flash overlapped at an average Jaccard similarity below 0.2 for every pair^[10].
• Vendor data points the same way. Ahrefs matched about 4 million AI Overview citations to Google rankings for 863,000 keywords: only 38% of cited pages ranked in the top 10 for that keyword, against 76% in its July 2025 run, though Ahrefs changed its parsing in between^[11].

There is no global “AI ranking” to optimize toward. Visibility is indexed by engine, surface, query phrasing and date. Any dashboard that blends engines into one score is averaging away the only thing you could act on.

Being known is not being recommended

This is the finding that matters most for a startup, and it deserves its own heading.

A January 2026 preprint, The Discovery Gap by Amit Prakash Sharma, not peer reviewed, tested 112 Product Hunt startups with 2,240 queries to two models in December 2025^[4]. When products were named directly, OpenAI’s gpt-4o-mini, answering from its training data through the API, recognized 99.4% of them. When the same products were sought through generic category discovery queries, the way a buyer searches before they know you exist, it surfaced them in 3.32% of cases. Perplexity’s Sonar, which searches the web live, fell from 94.3% recognition to 8.29% discovery. On Perplexity, discovery tracked referring domains, a classic SEO signal; the paper’s GEO content score predicted nothing on either model.

The model knowing your company exists tells you almost nothing about whether it will recommend you to someone who doesn’t. This is a single preprint using two models, so hold it loosely, but the distinction it draws is fundamental, and it’s the distinction most brand-monitoring dashboards quietly ignore. Tracking “does ChatGPT know us” while your buyers search “best tools for X” measures the wrong thing entirely.

Measuring it without fooling yourself

Answer engines are non-deterministic, and the measurement problem is genuinely harder than the optimization problem.

Tracking ChatGPT, Gemini, Google AI Mode and Perplexity daily for 45 days in early 2026, across four industries, Don’t Measure Once (Schulte, Bleeker and Kaufmann, University of St. Gallen; an April 2026 preprint, not peer reviewed) found day-to-day overlap in cited sources at a Jaccard of roughly 0.34 to 0.42 ^[12]: most of what you see on any given day differs from the day before. Their conclusion is that visibility is a distribution, so a single reading is one draw from a range, and every prompt needs repeating.

Then there’s the denominator problem. Engines don’t always search: in the same study, 57.8% of ChatGPT’s runs returned no citations at all^[12]. A dashboard that calculates your “share of citations” only among answers that contained citations, and reports that as visibility, is quietly excluding those runs.

Query phrasing changes everything too. Across 55,393 trending queries over 40 days, AI Overviews activated on 13.7% overall, but on 64.7% of queries phrased as questions^[9].

And a citation is not an endorsement. Stanford’s 2023 human audit of four generative search engines (Bing Chat, NeevaAI, perplexity.ai and YouChat)^[13] found only 51.5% of generated sentences were fully supported by their cited sources. A 2026 audit of AI Overviews found the problem persists: 11.0% of 98,020 claims were not supported by the pages cited for them^[9]. You can be cited for a claim your page doesn’t make.

Do this

Repeat every prompt several times: the St. Gallen team found estimates steadied only at seven to eight runs per prompt per day^[12]. Freeze your prompt set between runs, report per engine, keep presence and citation as separate numbers, count the runs where nothing was retrieved, and treat small movements as noise until they survive repetition. For Google, add Search Console’s generative AI performance report, which since June 2026 counts how often your pages are shown as sources in AI Overviews and AI Mode; it reports impressions, not clicks^[14].

Free tool

Test whether a change in AI visibility is real

The free AI Visibility Significance Test sizes your prompt set, scores presence and citation separately, and tells you whether a change between runs is signal or noise.

What the evidence actually supports doing

Nothing here is exotic. That’s rather the point.

1. Be retrievable first. Crawlable, indexed, and present in the raw HTML without JavaScript. This is the precondition for everything else, and it’s where I would look first.
2. Answer real questions directly. Relevance is the single most reproducible lever in the literature. Write the page that answers the question, not the page that games the answer.
3. Include verifiable, extractable evidence: dates, prices, definitions, specific figures with sources. Support is moderate and conditional, and the constraint matters: fabricated statistics can increase reuse while degrading everything that makes reuse worth having^[1].
4. Don’t keyword stuff. It cost a source 8% of its share of the answer in the original benchmark^[5].
5. Don’t rewrite whole page bodies purely for citability. The end-to-end evidence says that can cost you more in retrieval than it gains in citation.
6. Measure per engine, with repetitions, against a frozen prompt set. Distinguish presence from citation, and category discovery from brand recognition.
7. Publish an llms.txt if you like: it takes ten minutes, and Google says Google Search ignores it. Do it after everything above, and don’t call it a strategy.

The honest summary of where the field stands: there is strong evidence that already-retrieved content can be rewritten to claim more of an answer; moderate evidence that extractable, well-structured evidence helps; and, in the reviewed literature, no technique demonstrating a stable, cross-platform causal effect on organic discoverability or on downstream clicks and conversions. ^[1]

That isn’t an argument for ignoring AI search. Buyers are genuinely researching there, and absence shows up in clicks: Seer Interactive, an agency, found organic click-through of 0.70% for brands an AI Overview cited against 0.52% for brands it left out (3,119 queries, 42 organizations, third quarter of 2025; a correlation, not proof of cause)^[15]. It’s an argument for treating it as good information engineering with honest measurement attached, and for being skeptical of anyone selling a number. For the craft side of this, using AI in your marketing without flattening your voice, see Practical Ways to Use AI in Your Marketing Without Losing Your Voice.

If you want me to look at your own funnel, positioning and channels, my Growth Audit ends with three prioritized fixes and a 30-day action plan.