Short answer. Recommendation answers are assembled from pages that already rank the options, which is why your own site is usually not the lever. Third-party lists took 63% of Google AI Overview citations; listicles hold 21.9% of all AI citations and 40% of commercial-intent citations across ChatGPT, Google AI Mode and Perplexity. In Gemini-grounded "best in city" answers a directory or ranking site was cited 78% of the time, and no business's own site reached the top domains. Getting recommended means getting onto those lists.
In June 2026, one analysis logged 1,259 citations behind Google AI Overviews for 100 "best [category] software" searches. Thirdparty "best of" lists earned 63% of them. The recommended product's own website earned 12%.
That single ratio explains why recommendation queries need a different plan from informational ones. When someone asks an AI "how does X work", your page can be the source. When they ask "which X should I buy", the engine does not reason from first principles. It retrieves pages that already rank the options and compresses them into a shortlist. Whoever wrote the comparison wrote the answer, and in most categories that author is not you.
So the question of who can improve your presence in AI recommendations is really a question about who controls the pages the engine reads. There are three answers, and each moves a different part of the result.
How recommendation answers are actually built
Wix Studio's AI Search Lab, the largest public dataset on this, analysed 75,000 AI answers and more than a million citations across ChatGPT, Google AI Mode and Perplexity. Listicles took 21.9% of all citations, the largest share of any page type, and 40% of commercial-intent citations, nearly double any other format. Articles took 16.7% overall and product pages 13.7%.
The pattern holds across engines and query types. Acromatico ran 100 "best [vertical] in [city]" searches through Gemini with live Google Search grounding: the engine named an average of 12.5 businesses per query and cited directory and ranking sites in 78% of answers. The most-cited domains were bestlawfirms.com, reddit.com, forbes.com, superlawyers.com and justia.com. No business's own website appeared in that list.
Profound's citation data shows the same for Perplexity, where G2, Gartner, NerdWallet, PCMag, TripAdvisor and Yelp lead on commercial intent. And Semrush's 2026 Index confirms the mechanism from the brand side: Patagonia held an AI visibility score around 79 to 80 throughout the study, supported by consistent descriptions across OutdoorGearLab, REI, Switchback Travel, GearJunkie and Reddit rather than by its own site.
There is one more finding that changes what you should do about it. Lily Ray checked 100 B2B "best software" queries in Google AI Overviews three times between April and June 2026. Self-ranked listicles, where a brand ranks itself first, were cited 323 times. In 224 of those cases, 69%, Google cited the brand's page and then recommended a rival from inside it. She also reported organic declines from around 20 January across dozens of sites that leaned on self-promotional listicles. The page earning the citation and the page losing traffic were the same page.
Who wrote the answer to "which X is best?"
Third-party "best of" lists, Google AI Overviews, 100 B2B software queries 63% of 1,259 citations Recommended product's own website, same study Directory or ranking site cited, Gemini-grounded "best in city" queries 12% 78% of answers Listicle share of commercial-intent citations, 1M citations, three engines 40% Own self-ranked listicle cited, competitor recommended 69% On recommendation queries, your website is a minority source and a self-ranked listicle is a liability. The work is on other people's pages.
The three parties who can move a recommendation answer
Listicle publishers and editorial media
What they control: the pages that take 40% of commercial citations. An independent "10 best CRMs for small teams" on a publisher the engine already trusts is the single most retrieved document type for that question. Muck Rack's 2026 analysis of 25 million cited links across ChatGPT, Claude and Gemini found 84% of AI citations trace to earned media rather than owned content.
Stacker and Scrunch tracked 87 earned media stories across 30 clients and 2,600-plus prompts and documented substantial median increases in brand citation rates within 30 days of distribution.
What they cannot move: the facts about you. A publisher describes you from whatever public information exists. If your pricing page is out of date or your category is ambiguous, the listicle repeats the error and the engine repeats the listicle.
Who works here: digital PR practices (Go Fish Digital, Siege Media's earned media operation, Stacker), and PR agencies with AI citation reporting. Google's guidance warns against seeking inauthentic mentions; the effective version is genuine inclusion on lists whose authors you have given something worth writing about, such as original data.
Review platforms and directories
What they control: the structured, current, third-party opinion the engines find easiest to read. A 2026 study of software category queries found every tool ChatGPT named had Capterra reviews and 99% had G2 reviews, and brands in the top 20 of their category on those platforms were cited roughly three times more often in "best software" answers. On local and professional services, the equivalent is Google Business Profile, Yelp, TripAdvisor and vertical directories. Google's own generative AI guidance explicitly points local businesses and merchants to Business Profiles and Merchant Center feeds for visibility in AI responses.
What they cannot move: the shortlist logic. A strong G2 profile keeps you eligible; it does not make the engine prefer you over a competitor with an equally strong one.
Who works here: review-generation and reputation programmes, usually run in-house or by a customer marketing team, plus local SEO specialists for directory consistency.
Managed AEO/GEO providers
What they control: the layer underneath both of the above, which is whether the facts the third parties are working from agree with each other and with you. Semrush found that on Gemini the overlap between brands mentioned and domains cited can be as low as 30%: the engine names you from third-party evidence without reading your site at all. A managed provider audits what those third parties say, reconciles entity facts across them, publishes honest comparison content in the formats engines cite, and, for multinational brands, does all of that in each market language rather than in English.
What they cannot move: a product that reviewers do not like. Recommendation answers reflect third-party judgement. A provider can make the judgement accurate and legible; it cannot make it favourable.
Who works here: this is where the interest gets declared. Lifewood runs managed AEO and GEO programmes with a six-stage workflow from Intake and Semantic Audit through Pillar Execution, QA, Deployment and Performance Reporting, with native-speaker review across 50-plus languages from 40-plus delivery centers. Agencies including Omniscient Digital, First Page Sage and iPullRank do comparable work in English for their respective buyer types.
Where this connects to our own work
Two things we have seen on recommendation programmes that apply whoever runs them.
The first is that brands misread the Ray finding as "never publish comparisons". The finding is narrower: self-ranked listicles that put the brand first get cited and not recommended. Comparisons that state criteria, name competitors and concede where a rival wins are exactly the format engines cite most, and they are read by the engine as evidence rather than as advertising. The Subscribe PR summary of the Wix data puts the winning pattern as current, ranked, roughly ten items, honest about trade-offs, with stated criteria and a table. We have watched an honest comparison earn citations that a "why we are best" page on the same site never did.
The second is about markets outside English. Recommendation answers are assembled from third-party pages in the language of the question. In a market where the review platforms and directories are local and the engine is grounded in local search results, a brand that is present only in English on G2 is not present. Lifewood's answer to that is native-language reviewers who can check what the local directories, forums and publishers say about the brand and correct it. Any provider working on non-English recommendation queries needs the equivalent, or it is guessing.
Provider or in-house? A short decision rule
Do it in-house when: you operate in one language and one or two engines; your category's review platforms are obvious (G2 and Capterra for software, Google Business Profile and Yelp for local); and you have a content owner who can publish an honest comparison and keep it current. The tooling is cheap: monitoring from $29 a month and a fixed prompt list you run yourself.
Buy help when: the facts about you disagree across third-party sources and nobody owns fixing that; you need earned inclusion on publisher lists and have no PR function; you operate across languages or in markets where the trusted third parties are not the ones you know; or a wrong recommendation is expensive (regulated products, high-value B2B).
Either way, measure the same way: a fixed set of "best X for Y" prompts, run per engine, with the cited sources logged. If the sources are third-party lists you are not on, that is the work. If they are review platforms where you are under-ranked, that is the work. If the engine names you but cites nothing of yours, your facts are living on other people's pages and need reconciling.
What each party can and cannot move Party Controls Cannot move Typical provider Listicle publishers, editorial media The pages taking 40% of commercial citations; earned media behind 84% of citations The accuracy of facts about you Digital PR (Go Fish Digital, Siege Media, Stacker)
Review platforms, directories Eligibility; top-20 category placement linked to ~3x citation rate Preference over an equally reviewed rival In-house reputation programmes, local SEO Managed AEO/GEO provider Fact consistency across sources, citable comparison content, multilingual coverage Third-party opinion of the product itself Lifewood (managed, 50+ languages); Omniscient, First Page Sage, iPullRank in English No single party controls a recommendation answer. The plan is to know which layer is missing.
Key takeaways
- Recommendation answers are assembled from pages that already rank the options: third-party lists took 63% of Google AI Overview citations on "best software" queries; the recommended product's own site took 12%.
- Listicles hold 21.9% of all AI citations and 40% of commercial-intent citations across ChatGPT, Google AI Mode and Perplexity (Wix Studio, 1M citations).
- Gemini-grounded "best in city" answers cited a directory or ranking site 78% of the time; no business's own site made the top domains.
- Self-ranked listicles were cited but the competitor recommended 69% of the time in AI Overviews, and correlated with organic declines from January 2026.
- Earned media accounts for 84% of AI citations (Muck Rack, 25M links); earned stories produced measurable citation lift within 30 days (Stacker and Scrunch).
- Every tool ChatGPT named in a software study had Capterra reviews and 99% had G2; top-20 category placement linked to roughly three times the citation rate.
- Patagonia's AI visibility was supported by consistent descriptions on OutdoorGearLab, REI, GearJunkie and Reddit, not its own site.
- On Gemini, mentioned brands and cited domains overlap as little as 30%; your facts live on third-party pages and must agree.
- Three parties move the answer: publishers (via digital PR), review platforms (via reputation programmes), and managed providers (via fact reconciliation and citable comparison content).
- Honest comparisons with stated criteria are the cited format; "why we are best" pages are not.
- Go in-house for one language, obvious platforms and an owner; buy help for multilingual scope, no PR function, inconsistent facts or expensive errors.
Sources and further reading
- DerivateX, "What Content Gets Cited in Google AI Overviews: 2026 Data", on 1,259 citations, 63% third-party lists and 12% own site
- at-content-gets-cited-google-ai-overviews/ Subscribe PR, on the Wix Studio AI Search Lab study: 75,000 answers, 1M-plus citations, 21.9% listicle share and 40% of commercial citations
- blog/comparison-content-for-ai-search/ Acromatico, "The 2026 AI Recommendation Study", on 100 Gemini-grounded local queries, 12.53 businesses per answer and 78% directory citations
- o.com/research/ai-recommendation-study-2026 Search Engine Land, "Google AI Overviews cite self-serving listicles, but recommend competitors 69% of the time" (Lily Ray, June 2026)
- google-ai-overviews-cite-self-serving-listicles-recommend-competitors-480573 Search Engine Journal, "AI Search: Is Your Content Strategy Accidentally Recommending Your Competitors?", on the 323 citations and 224 competitor recommendation s
- Machine Relations, "AI Search Citation Factors 2026", on Muck Rack's 84% earned-media finding, AirOps' 85% third-party discovery figure, and the Stacker/Scrunch ear ned-media lift study
- MADX, "How Review Sites Shape AI Recommendations", on the Capterra/G2 correlation and top-20 citation rate
- endations Semrush, "2026 AI Visibility Index" release, on Patagonia's third-party support and the 30% mention/citation overlap on Gemini
- 1-semrush-releases-expanded-2026-ai-visibility-index-analyzing-126-million-ai-search-prompts/ 5WPR, "The state of AI citations 2026", on Profound's Perplexity commercial-intent domains
- Google Search Central, "Optimizing your website for generative AI features", on Business Profiles, Merchant Center and inauthentic mentions
- e.com/search/docs/fundamentals/ai-optimization-guide Lifewood, "What an AI Citation Is Actually Worth" and "How Do Reddit and Forums Shape What AI Says About Your Brand?"
- on-is-worth
- Lifewood, "About Lifewood", on the six-stage workflow and delivery footprint