• AI visibility

Guide · AI search

What happens to brands that skip GEO when competitors adopt it?

In controlled tests they disappear: when rival brands used optimized product copy, brands that did nothing received zero recommendations across 4,745 trials. The same research shows the advantage of going first shrinks fast as rivals catch up, and many popular tactics do not work at all. Skipping generative engine optimization (GEO), the work of making a brand visible and accurately described in AI answers, is risky; copying rivals’ tricks is not a strategy either.

By Underneath Team Guide Updated

The short version

  1. In a test of three AI assistants, invented brands that did not optimize got zero recommendations across 4,745 trials once rivals did (Chu and Hou (opens in a new tab)).
  2. Going first paid off, but the gain fell from +0.802 to +0.007 on the authors’ scale once all nine rivals used the same tactic.
  3. In a separate benchmark, popular rewriting tactics produced a reliable gain in only three of 54 tests, and gains shrank as more sites adopted them (Puerto and colleagues (opens in a new tab)).
  4. Signs of optimization appeared on 8.90% of pages in Google and Gemini results for 1,000 real questions, and on 16.36% of pages updated in 2026 (Chu, Leng and colleagues (opens in a new tab)).
  5. On one live website, ChatGPT referrals grew 5.7 times, but untreated pages grew 3.5 times too, so most growth came from ChatGPT itself (Watanabe and Nakayashiki (opens in a new tab)).

What happened to brands that did nothing in the experiments?

They vanished: once rivals optimized, brands that stayed neutral were never recommended.

Chu and Hou (opens in a new tab), of Trine University and Texas A&M, built lists of ten skincare products: one real, well-known brand and nine invented challengers, all with identical specs. They then gave some challengers authority-style copy, such as clinical-trial claims, and asked GPT-4o-mini, Claude Sonnet and Gemini 3 Flash to recommend one.

With no optimized rivals, the famous brand won every time. A single optimized challenger cut the famous brand’s survival from 100% to 19.8%. And across 4,745 trials, invented brands without optimized copy received zero recommendations. The authors put it bluntly: any brand that does not optimize when competitors do becomes invisible.

Read the conditions closely. The brands were invented, the specs were identical, and the clinical claims were deliberately made up to find the upper limit. The finding holds inside that experiment; real markets have more ways to stand apart.

Does moving first keep you ahead?

Not for long: the first mover’s advantage shrank by half with each 1.4 rivals that copied it.

The first brand to optimize gained a lot, +0.802 on the authors’ payoff scale. As others copied the same tactic, the gain decayed, halving every 1.4 brands. With all nine challengers optimized, each one’s gain fell to +0.007, close to nothing. The famous brand recovered to 93.8% of recommendations, because the assistants ignored a signal everyone now sent.

The assistants differed at that end point. Claude and GPT-4o-mini largely returned to the famous brand, while Gemini kept some lasting effect, with the famous brand at 84.9%. The authors call this a prisoner’s dilemma: every brand is better off optimizing, yet when all do, nobody gains, and no one can afford to stop.

A separate benchmark, C-SEO Bench (opens in a new tab), from Parameter Lab, NAVER and university researchers, found the same pattern for the best two rewriting methods in retail and video games, tested on GPT-4o-mini: early adopters gained, and the gain fell steadily as more sites adopted the same method. The authors describe it as congested and zero-sum.

Is search ranking still part of it?

Yes: whether your page is found in the first place still outweighs most rewriting.

C-SEO Bench found that moving a page to the top of the material an assistant reads produced far bigger gains than any rewriting method. The authors conclude that traditional search optimization remains essential and that GEO complements it rather than replacing it.

Rankings are not the whole story, though. In our study of 80 US buyer questions, 8.3% of the pages ChatGPT cited ranked in Google’s top 10 for the question, against 25.7% for Claude. Chu and Hou also ran a small add-on test in which a basic search step chose which products the assistant saw. Against one optimized rival, the famous brand was then recommended only 5.0–6.7% of the time, because the search step did not favor familiar names. The authors treat that probe as directional.

How many competitors are already optimizing for AI?

A growing minority, by one estimate: about one page in eleven in real search results shows signs of it.

Chu, Leng and colleagues (opens in a new tab), led from the CISPA Helmholtz Center for Information Security, built a detector for pages optimized for AI search. They ran it on pages returned by Google Search and Gemini for 1,000 real user questions. It flagged 8.90% of 10,095 pages. Among pages with a stated update date, the share rose from 7.02% for 2024 to 16.36% for 2026. The authors warn that the detector makes mistakes and that the rise is a trend in pages, not proof of rising adoption.

Our own data shows one visible form of it. In our best-of lists study, 24.2% of numbered “best X” lists cited by AI engines, where the publisher could be identified, ranked their own publisher first.

How big is the real-world payoff?

Smaller than headline case studies suggest, according to the most careful field study we found.

Watanabe and Nakayashiki (opens in a new tab) work at Glasp, the website they studied, so this is a vendor’s own account of one domain. In January 2026 they applied a bundle of changes, mainly rewritten titles and summaries, to one large section of their site and used the rest of the site as a comparison. Total ChatGPT referrals grew 5.7 times, but untreated pages grew 3.5 times over the same window. Once that rising tide was removed, the optimized section showed a step up of about 1.82 times. A stricter test left that result suggestive, not conclusive.

The lesson for a brand deciding whether to act is about measurement. Raw growth in AI traffic mostly reflects AI assistants getting bigger, so it cannot tell you whether your rivals’ work, or yours, is paying off.

What should you do about it?

Do not sit out, but compete on substance and measure honestly rather than copying rivals’ tactics.

  1. Start with being findable. Pages that search systems retrieve beat clever rewrites of pages they never see.
  2. Differentiate with real facts. Specific, verifiable evidence lasts longer than a tactic every rival can copy.
  3. Skip invented claims and keyword tricks. They can be flagged, and in benchmarks many tactics lowered rankings.
  4. Watch what rivals are doing. Check which competitors appear in AI answers to your buyers’ questions, and which pages and lists those answers cite.
  5. Measure against a baseline. Compare optimized pages with untreated ones, and judge visibility across many repeated questions. In our consistency study, pinning one brand’s visibility near 50% within 10 points would take about 97 runs.

If you want a structured way to do this, see our generative engine optimization service.

What does the research not tell us yet?

No study has followed real competing brands in live AI assistants over time, so the strongest results are from simulations.

  • The “zero recommendations” finding used invented brands, identical specs and deliberately made-up claims, in one product category.
  • Benchmarks used older assistant versions, often with the material handed to the assistant rather than found by search.
  • The prevalence estimate depends on an imperfect detector and covers one set of 1,000 questions.
  • The only field study covers one website, run and reported by its owner.
  • Nobody has measured what skipping GEO costs in sales.

Frequently asked questions

Is GEO necessary if my competitors are doing it?

The controlled evidence says ignoring it is risky. In one test, brands that did not optimize while rivals did received zero recommendations across 4,745 trials, though the brands and claims were invented.

Does being first to do GEO give a lasting advantage?

No. In the same test the first mover’s gain halved with every 1.4 rivals that copied it and fell to almost nothing when all nine did.

Do common GEO tactics like adding statistics work?

Often not. In one benchmark only three of 54 cases tested showed a reliable gain, and adding statistics lowered rankings in 19 of 24 settings.

How much web content is already optimized for AI search?

One detector flagged 8.90% of pages in Google and Gemini results for 1,000 real questions as optimized, rising to 16.36% among pages updated in 2026.

Sources

Free strategy call

Some questions are easier to answer about your own business.

Bring the one that matters most. On a free 30-minute call we’ll take a first look at it and send you a short written read afterward.

Ask about my businessHow we work