PELORA Research · Evidence review
What gets a business named by AI: we graded 30 studies
Dozens of studies claim to explain what makes ChatGPT, Google's AI Overviews, Perplexity and Gemini mention a business. Most were published by companies selling the fix, and many contradict each other. We read 30 of them at the source, graded each one on its method and its commercial interest, and turned the evidence into one scorecard: what works, what is mixed, and what to skip.
The short answer
Across 30 studies, the strongest evidence points to three things: be findable, answer the exact question your customer asks, and get talked about on sites you do not own. Tricks aimed at AI, from llms.txt files to schema added for AI and hidden prompts, either showed no measurable benefit or amount to manipulation. Every AI cites differently and the answers change from one run to the next, so judge your visibility across many checks, not one.
Six findings worth knowing
Relevance
Peer-reviewed research found models rely heavily on how relevant a page is to the question, and that moving a page up the sources the AI reads beat methods built specifically for AI. [3] [2]
81.9%
of sources ChatGPT drew from earned media (other people's sites) in a US automotive test, against 45.1 percent for Google. [4]
No effect
Adding schema markup made no measurable difference to ChatGPT or AI Mode citations in the only controlled test we found. [22]
38%
of pages cited in Google's AI Overviews rank in its top 10. Ahrefs counted about 76 percent in July 2025, with an older counting method. [17]
13.7%
of citations are shared by Google's AI Mode and AI Overviews for the same queries. Each AI is its own channel. [19]
<1 in 100
chance an AI gives the same list of brands twice for the same question. One check tells you very little. [30]
The scorecard
Thirteen tactics, judged on the evidence
Required: the platform's own rules make it a condition, not a guarantee. Supported: several studies point the same way, mostly by correlation. Correlation only: a consistent pattern, but no test of cause. Mixed: credible studies disagree. No measurable effect or No effect: tested or ruled on, and nothing found. Skip it or Do not: works against you or is manipulation. Numbers in brackets link to the study below.
| Tactic | Verdict | What the evidence says |
|---|---|---|
| Let the AI search crawlers in, and get indexed | Required | Google says being indexed and eligible to show in Search is the requirement, and that indexing is not guaranteed. OpenAI says sites that block its search crawler are left out of ChatGPT's search answers. [P1] [P3] |
| Answer the exact question your customer asks | Supported | Peer-reviewed work found models lean heavily on how relevant a page is to the question, and that moving a page up the sources an AI reads beat AI-specific rewrites. Neither tests this tactic directly; this verdict is our reading of both. [3] [2] |
| Get talked about on sites you do not own | Supported | Earned media dominated AI sources in a controlled comparison, brand mentions correlate with AI visibility far more than backlinks, and best-of lists, including self-published ones, are the most cited page type in recommendation prompts. [4] [20] [23] |
| Be on YouTube | Correlation only | YouTube mentions had the strongest correlation with AI visibility of any factor Ahrefs measured, and YouTube is a top-five cited source on four of five engines in Peec's count. Both are correlations. [21] [27] |
| Keep important pages genuinely current | Supported | By last update, AI assistants cited pages 13.1 percent fresher than organic search, with Google's AI Overviews the exception; recently updated pages drew nearly double the citations in another study. Changing a date without changing the content is not the same thing. [24] [25] |
| Rank on Google's first page | Mixed | It still correlates with being mentioned, but only 38 percent of AI Overview citations come from the top 10, and around 8 percent for ChatGPT, Gemini and Copilot. [26] [17] [18] |
| Build backlinks | Mixed | Two studies found backlinks weak or neutral; a third ranked referring domains its top factor. The research disagrees. [20] [26] [25] |
| Add statistics, quotes and sources to your content | Mixed | One peer-reviewed study found gains of up to 40 percent in a benchmark; a later peer-reviewed benchmark found most AI-specific methods ineffective; a third found models largely ignore references. [1] [2] [3] |
| Add schema markup to win AI citations | No measurable effect | The only controlled test we found saw no measurable change on ChatGPT or AI Mode, and a small decline on AI Overviews it could not attribute to schema. Google says no special schema is needed. [22] [P1] |
| Add llms.txt or other AI-only files | No effect | Google says Search ignores them; a 129,000-domain study found negligible impact on ChatGPT citations. [P2] [25] |
| Plant prompts or text aimed at AI | Works in lab tests. Skip it. | Two academic teams pushed products up with injected or crafted text on the page. It is manipulation; one team warns it could disrupt fair competition and the other calls for defenses. [5] [6] |
| Treat every AI as one channel | Do not | AI Mode and AI Overviews share 13.7 percent of their citations; ChatGPT leans on Wikipedia and Perplexity on Reddit; patterns swung sharply within weeks. [19] [28] [29] |
| Judge your visibility from one check | Do not | Ask the same question repeatedly and the list of brands almost never repeats. [30] |
What to do
In order of how strong the evidence is
- Make sure you can be found. Be indexed in Google and let OpenAI's search crawler in. This is the one step the platforms state as a rule, and it is a condition, not a guarantee. [P1] [P3]
- Answer the exact question your customer asks, plainly, on its own page. Relevance is what the peer-reviewed work says models lean on. [3] [2]
- Earn mentions on sites you do not own. Best-of lists, press, Reddit and YouTube. AI search draws heavily on third-party sources. [4] [20] [23] [27] [21]
- Keep your important pages genuinely current. Real updates, not new dates on old pages. [24] [25]
- Measure across many runs and every platform separately. The answers change constantly and each AI cites different sources. [30] [19] [28]
- Check what AI says about you, and correct the record where it is wrong. A peer-reviewed audit found AI answers often cite sources that do not support what they say. [7]
What to skip
llms.txt and other AI-only files [P2] [25], schema added in the hope of AI citations [22], and text or prompts planted for AI [5] [6]. The first two showed no measurable benefit; the third is manipulation.
For a plain-English guide to doing this, built from 20 sources, read How to get your business found on ChatGPT and AI search.
Where the research disagrees
Three open arguments
Do content rewrites for AI work?
The paper that named the field found adding quotations, statistics and sources lifted visibility by up to 40 percent in its benchmark [1]. A later peer-reviewed benchmark found most methods built for AI did not work and often hurt, while ordinary SEO did [2]. A third study found models largely ignore scientific references [3]. Our read: write for the person asking, and do not expect a formatting trick to do the work.
Do backlinks matter?
Two studies found backlinks weakly related or unrelated to AI mentions [20] [26]; a third ranked referring domains its top factor [25]. All three are correlations. Being linked to and being talked about tend to travel together, which may explain the split.
Is AI cutting clicks?
Pew found a click on a regular result in 8 percent of visits with an AI summary against 15 percent without one [8], and two vendor studies found large drops in click-through [11] [12]. Google says total clicks have been relatively stable year over year but published no numbers [16]. Pew measured clicks per visit, not total clicks over time, so the two are not a direct contradiction; only Pew's side can be checked.
The evidence
The 30 studies
Each study links to its publisher's own page. Every quotation was cut from a copy of that page saved on 4 October 2026, and a script stops the build if a quotation cannot be found in it. Grades: A 7, B 14, C 8, claim 1.
Academic research
Peer-reviewed papers and preprints. Most of the controlled experiments among the 30 are here.
GEO: Generative Engine Optimization
Rewriting content to add quotations, statistics or cited sources raised visibility in generative engine answers by up to 40 percent in their benchmark. Keyword stuffing gave little to no improvement. Results varied by topic.
What the source says
“Moreover, we show the efficacy of these strategies varies across domains, underscoring the need for domain-specific optimization methods.”
Read 4 October 2026.
C-SEO Bench: Does Conversational SEO Work?
Most methods built specifically to win AI answers did not work and often hurt. Ordinary SEO that improves a page's ranking worked better, and gains shrank as more competitors used the same tactics.
What the source says
“Our experiments reveal that most current C-SEO methods are not only largely ineffective but also frequently have a negative impact on document ranking, which is opposite to what is expected.”
Read 4 October 2026.
What Evidence Do Language Models Find Convincing?
When sources conflict, models lean on how relevant a page is to the question and largely ignore style signals people trust, such as scientific references or a neutral tone.
What the source says
“current models rely heavily on the relevance of a website to the query, while largely ignoring stylistic features that humans find important such as whether a text contains scientific references or is written with a neutral tone”
Read 4 October 2026.
Generative Engine Optimization: How to Dominate AI Search
AI search leaned heavily on earned media (third-party sources) over brands' own sites and social content. In a US automotive comparison, ChatGPT drew 81.9 percent of its sources from earned media, against 45.1 percent for Google. The authors also describe a big-brand bias.
What the source says
“AI search delivered 81.9 percent Earned and 18.1 percent Brand”
Read 4 October 2026.
Ranking Manipulation for Conversational Search Engines
Prompt-injection text added to a product page reliably pushed low-ranked products up, and the attack carried over to Perplexity. It works in the lab, it is manipulation, and the authors call for defenses against it.
What the source says
“a tree-of-attacks-based jailbreaking technique which reliably promotes low-ranked products”
Read 4 October 2026.
Manipulating Large Language Models to Increase Product Visibility
A crafted block of text added to a product page made an AI far more likely to recommend that product first. The authors warn this could disrupt fair competition.
What the source says
“We demonstrate that adding a strategic text sequence (STS) -- a carefully crafted message -- to a product's information page can significantly increase its likelihood of being listed as the LLM's top recommendation.”
Read 4 October 2026.
Evaluating Verifiability in Generative Search Engines
Only about half of AI answer sentences were fully supported by their citations, and about a quarter of citations did not support the sentence they were attached to. The engines tested are older, but the lesson stands: check what AI says about you.
What the source says
“a mere 51.5% of generated sentences are fully supported by citations and only 74.5% of citations support their associated sentence”
Read 4 October 2026.
What AI search is doing to traffic
How often people click, and how much traffic AI assistants send.
Google users are less likely to click on links when an AI summary appears in the results
People clicked a regular result in 8 percent of visits with an AI summary, against 15 percent without one, and clicked a link inside the summary in 1 percent of visits.
What the source says
“Users who encountered an AI summary clicked on a traditional search result link in 8% of all visits.”
Read 4 October 2026.
Americans and AI, 2026
49 percent of US adults use AI chatbots, up from 33 percent in 2024 (Pew notes the 2024 question was worded differently). Search and work are the most common uses, and a majority say they read the AI summaries at the top of search results.
What the source says
“A majority of Americans say they read AI summaries at the top of search results”
Read 4 October 2026.
ChatGPT Is Not Replacing Google, It's Expanding Search [Study]
People who started using ChatGPT did not search Google less. ChatGPT added to search rather than replacing it.
What the source says
“Based on the data, ChatGPT adoption did not reduce Google Search usage.”
Read 4 October 2026.
AI Overviews Reduce Clicks by 34.5%
The top-ranking page got a 34.5 percent lower click-through rate when an AI Overview appeared. Ahrefs has since re-run the study for 2026.
What the source says
“We analyzed 300,000 keywords and found that the presence of an AI Overview in the search results correlated with a 34.5% lower average clickthrough rate (CTR) for the top-ranking page, compared to similar informational keywords without an AI Overview.”
Read 4 October 2026.
AIO Impact on Google CTR: September 2025 Update
Organic click-through on informational queries with an AI Overview fell 61 percent between June 2024 and September 2025, and queries without one fell too. Brands cited inside the AI Overview got 35 percent more organic clicks than brands that were not. That is a comparison, not proof the citation caused it. Seer has since published a third version covering January 2024 to February 2026.
What the source says
“When you are cited in an AIO, you get 35% more organic clicks and 91% more paid clicks compared to when you are not cited at all.”
Read 4 October 2026.
Google AI Overviews: New CTR Study Reveals How to Navigate Negative SERP Impact
Keywords that started showing AI Overviews lost 15.49 percent of click-through on average. Branded keywords gained 18.68 percent; non-branded lost 19.98 percent.
What the source says
“Overall, keywords that now trigger AI Overviews saw a CTR decline of -15.49% on average.”
Read 4 October 2026.
What our AI Mode user behavior study reveals about the future of search
In Google's AI Mode, the typical task ended without a single click to an outside site, and 77.6 percent of sessions had no external visit. A small study, but it watched real people rather than estimating.
What the source says
“The median number of external clicks per task was zero.”
Read 4 October 2026.
The 2026 AEO / GEO Benchmarks Report
AI assistants sent 1.08 percent of all website traffic across 10 industries, and 87.4 percent of that AI traffic came from ChatGPT. Small today, and growing.
What the source says
“AI referral traffic accounts for 1.08% of all website traffic for these 10 key industries.”
Read 4 October 2026.
AI in Search is driving more queries and higher quality clicks
Google says total organic clicks have been relatively stable and that click quality has risen. It published no numbers to check this against, so we grade it as a claim, not a study.
What the source says
“Overall, total organic click volume from Google Search to websites has been relatively stable year-over-year.”
Read 4 October 2026.
What gets cited
Large studies of which pages and brands AI answers name, and what those pages have in common.
Update: 38% of AI Overview Citations Pull From The Top 10
Only 38 percent of pages cited in AI Overviews ranked in Google's top 10 for that search. Ahrefs counted about 76 percent in July 2025, but improved its parsing method in between, so the two are not like-for-like.
What the source says
“Update: 38% of AI Overview Citations Pull From The Top 10”
Read 4 October 2026.
Only 12% of AI Cited URLs Rank in Google's Top 10 for the Original Prompt
Across the AI assistants it checked, only 12 percent of cited links were in Google's top 10 for the same prompt. Perplexity was the exception at about 1 in 3; for the others it was around 8 percent.
What the source says
“Only 12% of AI Cited URLs Rank in Google”
Read 4 October 2026.
Are AI Mode and AI Overviews Just Different Versions of the Same Answer? (730K Responses Studied)
Google's AI Mode and AI Overviews said much the same thing (86 percent semantic similarity) but shared only 13.7 percent of their citations.
What the source says
“AI Mode and AI Overviews reach very similar conclusions (86% semantic similarity) while citing different sources (only 13.7% citation overlap)”
Read 4 October 2026.
An Analysis of AI Overview Brand Visibility Factors (75K Brands Studied)
Brand mentions across the web had the strongest link to appearing in AI Overviews (0.664). Backlinks were weak (0.218). Ahrefs stresses this is correlation, not cause.
What the source says
“Brand web mentions show the strongest correlation (0.664) with AI Overview brand visibility.”
Read 4 October 2026.
Top brand visibility factors in ChatGPT, AI Mode and AI Overviews
YouTube mentions had the strongest link to AI visibility of anything measured (about 0.737), ahead of web mentions, and YouTube is ChatGPT's sixth most-cited domain. Correlation, not proof of cause.
What the source says
“YouTube mentions show the strongest correlation with AI visibility (~0.737”
Read 4 October 2026.
We Tracked 1,885 Pages Adding Schema. AI Citations Barely Moved.
Adding schema markup made no measurable difference to ChatGPT or AI Mode citations. AI Overviews showed a small but statistically significant 4.6 percent decline that Ahrefs could not attribute to schema. The pages tested were already cited often, so it cannot speak to brand-new pages.
What the source says
“Statistically indistinguishable from zero”
Read 4 October 2026.
Do self-promotional best lists boost ChatGPT visibility?
Across 750 best-of recommendation prompts, recently updated best-of lists were the most common page type ChatGPT cited, at 43.8 percent of source pages, including lists where the publisher ranked itself first.
What the source says
“"best X" blog lists represented 43.8% of all page types”
Read 4 October 2026.
AI assistants prefer to cite fresher content
By last update, pages cited by AI assistants were 13.1 percent fresher than pages ranking in organic search (25.7 percent by publish date, averaging ChatGPT, Perplexity, Copilot and Gemini). Google's AI Overviews were the exception, citing slightly older pages. Ahrefs warns against changing dates without changing content.
What the source says
“25.7% “fresher””
Read 4 October 2026.
How to optimize for ChatGPT: LLMs.txt doesn't matter but brand mentions on Quora and Reddit do
Referring domains topped its list of factors, content updated in the past three months drew nearly double the citations (6.0 against 3.6), and llms.txt files made no meaningful difference.
What the source says
“has shown negligible impact on ChatGPT citation likelihood.”
Read 4 October 2026.
What drives brand mentions in AI answers?
Brands on Google's first page had a strong link to being mentioned by GPT-4o, queried through OpenAI's API rather than ChatGPT search (about 0.65). Backlinks were weak or neutral.
What the source says
“Brands ranking on page 1 of Google showed a strong correlation (~0.65) with LLM mentions.”
Read 4 October 2026.
Top domains cited by AI search: analysis based on 30M sources
Reddit was the most cited domain overall, followed by YouTube, LinkedIn, Wikipedia and Forbes. YouTube was a top-five source on four of the five engines; on ChatGPT it ranked sixth in a separate Ahrefs count.
What the source says
“Reddit and YouTube appear across all five platforms, which reinforces the overall ranking.”
Read 4 October 2026.
AI platform citation patterns
Each AI leans on different sources. Within each platform's top 10 sources, Wikipedia was 47.9 percent of ChatGPT's and Reddit was 46.7 percent of Perplexity's.
What the source says
“Within ChatGPT's top 10 most-cited sources, Wikipedia accounts for nearly half (47.9%) of citations among these leading sources.”
Read 4 October 2026.
The most-cited domains in AI: a 3-month study
Citation patterns swing fast: in September 2025 ChatGPT abruptly cited Reddit and Wikipedia far less, while on AI Mode the biggest growth came from YouTube, Reddit and Facebook.
What the source says
“In September 2025, ChatGPT suddenly began citing Reddit and Wikipedia”
Read 4 October 2026.
How to measure it
Whether a single check of AI visibility means anything.
AIs are highly inconsistent when recommending brands or products
Ask an AI the same question many times and you will rarely get the same list of brands twice. A single check of where you appear tells you very little.
What the source says
“AIs rarely give the same list of brands or recommendations twice (<1 in 100 times, no matter the question)”
Read 4 October 2026.
The platforms' own rules
What Google and OpenAI say, in their words
- P1 Google Search Central, AI features and your website. No special steps are needed to appear in AI Overviews or AI Mode, and no special files or schema.
- P2 Google Search Central, AI optimization guide. Optimizing for AI search is still SEO; llms.txt and AI-only markup are ignored by Google Search.
- P3 OpenAI, Overview of OpenAI crawlers. Block OpenAI's search crawler and you are left out of ChatGPT's search answers.
Method
How the studies were chosen and graded
- Included. Studies from 2023 onward that state a sample and a method and measure what is linked to, or causes, a page or business appearing in ChatGPT, Google AI Overviews or AI Mode, Perplexity, Gemini or Copilot, plus studies of what AI search is doing to clicks. The platforms' own documentation is listed separately as rules, not studies.
- Read at the source. Every page was fetched and saved on 4 October 2026. Where a page did not open or a figure appeared only inside a chart, the figure was left out; we did not use a copy from any other site. Quotations come from the saved source page, or from the paper's full text where the figure is not in its summary page.
- Graded on method. A: peer-reviewed research or independent measurement of real behaviour. B: a large sample with a published method, or a preprint. C: a ranking of factors by correlation, a small sample, or a method only partly stated (such as vendor tallies that do not say how prompts were chosen). Claim: a statement with no data published.
- Interest named. 21 of the 30 were published by companies or consultants that sell search software or services, or by the company that runs the search engine. That does not make them wrong; it is stated on every study so you can weigh it. We have one too: PELORA is a marketing agency, and AI search optimization is one of the services we sell, so weigh this review the same way.
- Correlation is labelled. Most industry studies show what visible brands have in common, not what made them visible. Only the academic experiments and one vendor test [22] test cause and effect.
Considered and left out
- SparkToro and Datos, AI tools' share of searches (3.2 percent). The figure appears only inside a chart on the page we could read, so we could not confirm it word for word.
- SparkToro and Similarweb, zero-click searches in 2026 (68.01 percent). The primary page did not open for us and the coverage page returned an error. We do not cite figures we could not read at source.
- Seer Interactive, SearchGPT citations matching Bing (87 percent). We could not locate the original study page.
- BrightEdge, AI platforms disagree on brand recommendations (61.9 percent). No sample or method stated.
- Profound, 11 percent of domains cited by both ChatGPT and Perplexity. The figure circulates with different sample sizes in different reports; we could not tie it to one method.
Cite or embed
Citation: PELORA Research (2026). What gets a business named by AI: we graded 30 studies. PELORA Marketing, published 4 October 2026. https://peloramarketing.com/research/ai-visibility-evidence-review-2026
Citing one study: cite the original study and its date, and this page as where you found it graded.
Embed the scorecard: free to use with the credit link.
<a href="https://peloramarketing.com/research/ai-visibility-evidence-review-2026"><img src="https://peloramarketing.com/images/ai-visibility-scorecard-2026.png" alt="What gets a business named by AI: scorecard of 30 studies, PELORA Research 2026" width="1600" style="max-width:100%;height:auto"></a>
<p>Source: <a href="https://peloramarketing.com/research/ai-visibility-evidence-review-2026">PELORA Research, What gets a business named by AI (2026)</a></p>View the scorecard image · Download the CSV
Corrections: if we have described your study wrongly, or it has been updated, write to hello@peloramarketing.com. Corrections are made on the page and logged here.
Update log
- 4 October 2026. First published: 30 studies and 3 platform rules, all read on 4 October 2026. Updated quarterly.
Want it applied to your business?
We read the research so you do not have to
A free Growth Check is one page on your website, your search and AI visibility, and the ads you run. You keep it either way. No agency controls what an AI says. Be wary of anyone who guarantees a ChatGPT ranking.