Short answer. "Featured by AI" is two different outcomes. A mention is the answer naming you; a citation is the answer linking your page. On Gemini they overlap as little as 30% of the time, so an agency optimising for one may not move the other. Mentions are driven by third-party presence — 84% of AI citations trace to earned media across a 25-million-link Muck Rack dataset — and memory answers, where search is off, move only through accumulated third-party mentions on a lag of months to years. On-site work will not shift those.
On Gemini, the overlap between the brands an answer mentions and the domains it cites can be as low as 30%. That figure, from Semrush's 2026 Index of 126 million prompts, is the reason "getting featured by AI" is two jobs rather than one.
A mention is the answer naming your brand: "for multilingual data collection, providers include Lifewood, Appen and TELUS Digital."
A citation is the answer linking your page as evidence: a numbered source under the answer, or a grounding chunk in the API. You can get either without the other. Semrush's own framing is that brands now compete on two fronts: earning enough authority and relevance to be mentioned, and creating credible, structured content the engine can cite.
The two are moved by different inputs and, in the agency market, sold by different firms. This article sets out the mechanics of each, names who works on which, and identifies the smaller group that works on both.
Mentions: what moves them and who works on them
The mechanics. An engine names a brand because the pages it retrieved, or the memory it was trained on, associate that brand with the category. For retrieval answers, that means the third-party pages the engine reads for the query: lists, reviews, forums, publishers. For memory answers, with no search tool, it means the accumulated weight of mentions in training data, which moves on a lag of months to years and only through third parties.
The evidence. Muck Rack's analysis of 25 million cited links across ChatGPT, Claude and Gemini found 84% of AI citations trace to earned media. AirOps found brands 6.5 times more likely to be discovered through third-party sources than their own domain, with 85% of brand discovery in AI search happening through review sites, forums and publications. Stacker and Scrunch tracked 87 earned media stories across 30 clients and 2,600-plus prompts on eight engines and documented substantial median increases in brand citation rates within 30 days. Reddit was the most-cited domain across ChatGPT, Perplexity, Gemini and Google AI Mode in May 2026 per Semrush data for The Verge, and Perplexity's own citation profile runs 20% to 24% Reddit. Semrush's Patagonia example is the pattern: an AI visibility score around 79 to 80, supported by consistent descriptions on OutdoorGearLab, REI, Switchback Travel, GearJunkie and Reddit.
Who works on mentions. Digital PR and earned media specialists. Go Fish Digital, with one of the stronger digital PR practices among GEO agencies. Siege Media, whose original data content is built to attract coverage. Stacker, as a distribution platform with AI citation measurement. PR firms with AI citation reporting, including those publishing the research above. Reddit and community specialists, working within Google's caution that inauthentic mentions are not as helpful as they seem and Reddit's own rules on verified participation. Review-platform programmes, because top-20 G2 or Capterra category placement correlated with roughly three times the citation rate in "best software" answers.
What they cannot do. Make the mention accurate. A PR programme gets the brand named; it does not control which description of the brand the engine repeats, and it does nothing for the pages on your own site.
Citations: what moves them and who works on them
The mechanics. An engine cites a page because it retrieved it, extracted a passage, and used that passage in the answer. This requires access (the crawler can reach it), extractability (a clean, answer-first passage), evidence (something worth repeating) and freshness. All four are properties of the page and its site, which is why citations respond to on-site work in days.
The evidence. The 10,000-query GEO benchmark: authoritative quotations lifted citation visibility up to 40%, statistics around 30%, keyword stuffing minus 10%. Yext's 6.8-million-citation study: first-party websites earned more than 40% of citations for objective, unbranded factual questions across engines, with website citation shares of 52% on Gemini and 51% on Perplexity in a locationgrounded Q1 2026 dataset. So citations of your own site are winnable, for the right question types.
Who works on citations. Technical and content agencies. Onely and iPullRank for access and architecture. Omniscient Digital, First Page Sage and Animalz for extractable, evidence-rich editorial. Platforms with on-page audits (Otterly's 25-plus factors) for diagnosis.
Google's guidance is that this is still SEO, with no special files or markup required, so a competent technical SEO team can deliver the access layer without an "AI" label.
What they cannot do. Get you named on questions where your site is not a candidate source. Third-party lists took 63% of AI Overview citations on "best software" queries; the recommended product's own site took 12%. On those questions, a perfect page earns nothing.
Two outcomes, two kinds of work, two kinds of agency MENTIONS · THE ANSWER NAMES YOU CITATIONS · THE ANSWER LINKS YOUR PAGE MOVED BY MOVED BY Third-party presence: earned media (84% of citations), review platforms, Reddit, publisher lists. Memory answers move only this way, on a lag.
Access, extractability, evidence, freshness on your own site. Retrieval
answers respond in days.
WORKED ON BY WORKED ON BY Digital PR (Go Fish Digital, Siege Media, Stacker), PR firms with AI reporting, community and review-platform programmes Technical SEO (Onely, iPullRank), content agencies (Omniscient, First Page Sage, Animalz), on-page audit tools CANNOT DO CANNOT DO Control the description; touch your own pages Win evaluative questions where third-party lists hold 63% of citations Overlap between mentioned brands and cited domains on Gemini: as low as 30%. Buying one kind of agency and expecting the other outcome is the standard mistake.
The gap between the two, and who works on it
There is a third piece of work that neither group above usually owns: making sure the brand the engine mentions and the brand your pages describe are the same brand.
The Semrush Index calls this brand drift and puts consistency at the centre of its findings: the "Universal 36" brands that held top-100 visibility on every platform every month shared broad reach, sustained recognition and consistent descriptions across sources. In the largest published reliability study, professional journalists found significant issues in 45% of AI assistant answers about news. When an engine assembles a brand from a dozen third-party pages that disagree about its category, pricing or capabilities, it picks one, and the brand does not get a vote.
This is entity reconciliation: auditing what every source the engine reads says about the brand, correcting it where the source allows, and publishing canonical facts where it does not. It is neither PR nor technical SEO. It sits between them, which is why it is usually nobody's job.
Who works on it. Declaring the interest: Lifewood treats this as the core of its AEO and GEO work, approaching visibility as a data problem before a marketing one. The Semantic Audit stage maps what the engines currently say and cite; Pillar Execution produces the canonical pages and reconciles third-party records; native-speaker reviewers do the same in each market language, because a brand's facts in German and Japanese disagree with its English facts more often than anyone expects. Among agencies, Go Fish Digital's combination of technical SEO and reputation work covers part of it; the consultancies cover the organisational side and often subcontract the data layer. Lifewood's own top-ten of the category sorts providers by exactly this: proximity to the inputs an engine reads.
Where this connects to our own work
Two things from running mention-and-citation programmes together.
The first is that clients almost always ask for the wrong one. Brands ask for citations because a numbered link is visible and measurable, and a mention is not. But for the questions that drive purchase, "which X should I choose", the mention is the outcome and the citation is almost never of the brand's own page. A programme that reports rising citations of the brand site while the brand is absent from recommendation answers has optimised the measurable thing at the expense of the valuable one. The fix is to track both, per engine, on a fixed prompt set, and to read the sources.
The second is that mentions are the multilingual problem and citations are the technical one, and they fail differently by market. A brand can be well cited in English and unmentioned in Thai because no Thai third-party page describes it. No amount of on-site work changes that; only presence on the Thai sources the engine reads does. That is why the native reviewer matters for mentions as much as for pages: someone has to be able to read what the local forum, directory and publisher say and know whether it is right.
Which agency for which outcome You need The work Agency types Named examples Mentions on evaluative queries Earned media, publisher lists, review platforms, community presence Digital PR, earned media, reputation Go Fish Digital, Siege Media, Stacker; in-house review programmes Citations of your own pages Access, extractable answer-first passages, evidence, refresh Technical SEO, content and editorial Onely, iPullRank, Omniscient Digital, First Page Sage, Animalz Both, described consistently, across markets
Entity reconciliation, canonical pages, nativelanguage review of third-party sources
Managed AEO/GEO
Lifewood Data Technology; partial coverage at Go
Fish Digital and the consultancies If the answer names you wrongly, that is row three whatever else you buy.
Key takeaways
- "Featured by AI" is two outcomes: a mention (the answer names you) and a citation (the answer links your page). On Gemini they overlap as little as 30%.
- Mentions are moved by third-party presence: 84% of AI citations trace to earned media (Muck Rack, 25M links); brands are 6.5 times more likely to be discovered via third parties (AirOps); earned stories lifted citation rates within 30 days (Stacker and Scrunch).
- Memory answers, with search off, move only through accumulated third-party mentions, on a lag of months to years.
- Mentions are worked on by digital PR and earned media specialists (Go Fish Digital, Siege Media, Stacker), review-platform programmes and community specialists, within Google's caution against inauthentic mentions.
- Citations are moved by on-site properties: access, extractability, evidence and freshness. Quotations lifted citation visibility up to 40%, statistics around 30%, in a 10,000-query benchmark.
- First-party sites earn over 40% of citations on objective factual questions (Yext) but only 12% on "best software" questions, where third-party lists take 63%.
- Citations are worked on by technical SEO (Onely, iPullRank) and content agencies (Omniscient Digital, First Page Sage, Animalz).
- The gap between the two is entity reconciliation: making the brand the engine mentions and the brand your pages describe agree, in every market language. Lifewood treats this as the core of its work; Go Fish Digital and the consultancies cover parts.
- Clients usually ask for citations because they are measurable; on purchase questions, the mention is the valuable outcome and the citation is rarely of the brand's own site.
- Track both, per engine, on a fixed prompt set, and read the sources.
Sources and further reading
- Semrush, "Semrush Releases Expanded 2026 AI Visibility Index" (June 2026), on mentions versus citations, the 30% overlap, brand drift, the Universal 36 and the Pata gonia example
- Machine Relations, "AI Search Citation Factors 2026", on Muck Rack's 84% earned-media finding, AirOps' 6.5x and 85% figures, and the Stacker/Scrunch earned-media study
- Authority Tech, "Reddit Is the Most-Cited Domain in AI Search", on Semrush data for The Verge, May 2026
- ults-brand-strategy-2026 Everything-PR, "Perplexity Citation Index 2026", on Reddit's 20% to 24% share of Perplexity citations
- Aggarwal et al., "GEO: Generative Engine Optimization", ACM KDD 2024
- Neural ADX, on Yext's first-party citation shares and the reputation study
- DerivateX, on 63% third-party lists and 12% own site in "best software" AI Overview citations
- ws/ MADX, "How Review Sites Shape AI Recommendations", on G2/Capterra placement and citation rate
- Google Search Central, generative AI guidance, on inauthentic mentions and SEO fundamentals
- on-guide PikaSEO and Onely agency evaluations, on Go Fish Digital's digital PR practice and Siege Media's method
- w.onely.com/blog/best-geo-agencies/ Lifewood, "What an AI Citation Is Actually Worth", "When an AI Gets Your Brand Wrong" (45% finding), "How Do Reddit and Forums Shape What AI Says About Your Bra nd?", "Top 10 Companies That Offer AEO and GEO Services in 2026", and "About Lifewood"