Third-party mentions: the 85% of AI visibility that is not your website
Brand mentions for AI visibility: where they live by business type, a 90-day mention plan, Reddit rules for honest participation, and how to measure growth.
In short
According to AirOps’ 2025 analysis of 21,311 brand mentions across ChatGPT, Claude and Perplexity, about 85 percent of a brand’s visibility in AI answers is formed off its own website: in media, directories, review platforms, comparison articles, forums and YouTube. Per an Ahrefs study of 75,000 brands, mentions on those sites correlate with appearing in Google AI Overviews roughly three times more strongly than backlinks (0.664 versus 0.218).
That means the biggest lever in generative engine optimization is not on-page work but a disciplined, honest program of earning mentions where your buyers’ questions are already being answered. This article maps mention sources for SaaS, clinics, HoReCa and B2B, lays out a 90-day plan, covers the Reddit rules after the bot cleanup, and shows how to measure mention growth next to Share of Voice.
The data: mentions beat links, and most of it happens off-site
Two numbers from 2025-2026 research change how you should split a GEO budget.
First, brand mentions on third-party sites correlate with appearing in Google AI Overviews roughly three times more strongly than backlinks. Assistants are not calculating PageRank. They summarize what many independent sources say about a category and pull the names that appear consistently and in a positive context. A single mention in a well-structured comparison article can do more than a dozen links from pages that never name you.
Second, about 85 percent of a brand’s AI visibility is formed away from its own domain. Your site tells the assistant what you claim about yourself. Everyone else tells it whether to believe you. When ChatGPT answers “best invoicing tool for EU freelancers,” it leans on review platforms, roundups, forum threads and video transcripts far more than on your pricing page.
The practical conclusion: if your GEO work is 80 percent on-page and 20 percent off-page, the ratio is backwards. Fix the crawler and extractability basics once. Per the Otterly.ai AI Citations Report 2026, 73 percent of sites have technical barriers to AI crawlers, including robots.txt blocks, CDN-level restrictions and JavaScript rendering problems, and part of that can be fixed in an hour. Then move most of the ongoing effort off-site.
Where mentions live, by business type
The right sources differ by category. The table below is a starting map; your own audit will show which of these are actually cited by assistants for your prompts, and those are where you start. The selection logic is covered in how ChatGPT chooses sources to cite.
| Business type | Highest-value mention sources | Why they get cited |
|---|---|---|
| B2B SaaS | Software review platforms (G2, Capterra and similar), independent comparisons and “alternatives to X” articles, podcasts with published transcripts, integration partner directories | Structured, comparative, frequently updated, written in the buyer’s language |
| Clinics and healthcare | Review platforms, doctor and clinic directories, local and city media, professional association listings, patient forums | Location and specialty are stated directly; reviews carry sentiment; directories confirm the entity |
| HoReCa | Google Maps and Apple Maps profiles, review sites, city guides and local media, food bloggers and video creators, booking platforms | Assistants answer “where should I eat” from maps and guides, not from venue websites |
| B2B services and manufacturing | Industry media and trade publications, case studies on client sites, event and conference pages, supplier directories, association member lists | Third-party confirmation of who you work with and where you appear as an expert |
Two patterns hold across all four rows. Pages that compare several options get cited more than pages about a single company. And pages with a visible date and a named author get cited more than anonymous, undated ones.
The 90-day mention plan
Mentions accumulate slowly, so plan in quarters. Each workstream below has an owner, a monthly quota and a place in the mention log.
Weeks 1-2: inventory and hygiene. Pull the citation column from your AI visibility audit. List every directory, review platform and media page that already mentions you. Fix the name, address, category, description and links on each one. Entity consistency matters more than volume (more on that below).
Month 1: PR and expert commentary. Pitch two to four story ideas a month to industry or local media where you have real data or a real point of view. Give expert commentary to journalists working on category roundups. The goal: placements that name the company and a person, and that will still load in two years. Skip paid advertorials on sites nobody cites.
Months 1-2: client case studies on THEIR sites. The most underused source in B2B. Your client publishes a case study or a “tools we use” page on their own domain and names you. It is an independent domain, it is factual, and it is exactly the kind of page an assistant uses to confirm who works with whom. Ask every satisfied client; offer to draft it.
Month 2: directory and review program. On each relevant platform, complete the profile and build a steady, honest flow of reviews from real customers. Reply to reviews. Do not incentivize ratings or buy reviews: platforms remove them, and assistants pick up the negative coverage that follows.
Months 2-3: YouTube. Assistants read video transcripts. Publish videos or appear in videos where the brand name is spoken and written in the title and description. Interviews on other creators’ channels count as third-party mentions; your own channel is an owned asset but still helps entity clarity.
Month 3: honest community participation. Find the forums, Slack and Discord groups, LinkedIn communities and subreddits where your buyers ask questions. Answer them under your own name. The mention comes when someone else recommends you later, not when you name yourself.
Every month: log and measure. Every new mention goes into the log with URL, date, source type, a flag for whether the name is spelled correctly, and whether an assistant later cited it.
Reddit after the bot cleanup
Reddit is the most-cited domain in AI answers, according to Peec AI’s 2026 analysis of 30 million sources, and that is exactly why the platform is flooded with fake accounts. Reddit itself has reported removing an average of 100,000 automated accounts a day as part of its anti-bot enforcement, and individual communities permanently ban accounts for organized brand promotion attempts.
The rules worth following:
- Disclosed participation only. Your username or flair says who you are and where you work. Every time.
- Answer, do not pitch. Respond to a real question with a real answer. Mention your product only when it is the direct answer, and say so plainly.
- No sock puppets, no seeded questions, no “a friend of mine uses this.” These are exactly the patterns Reddit cleans out.
- No paid Redditors. Same category.
- Accept that you will not control the wording. Real mentions come from users describing you in their own words, sometimes with criticism. That is exactly what makes them credible to an assistant.
If an agency offers you “Reddit seeding” as a GEO tactic in late 2026, ask for a written description of the method. If it involves undisclosed accounts, walk away.
Consistency of the brand entity across sources
An assistant has to decide whether “Acme Clinic,” “Acme Dermatology,” “AKME (Kyiv)” and “akme-clinic.com” are one company or four. If the answer is unclear, mentions get split across entities and none of them reaches the threshold to be named in an answer.
Checklist for one clean entity:
- One canonical name, spelled the same everywhere you control, requested from every partner and journalist.
- One short description (one sentence), reused on every profile so the category is unambiguous.
- Same address, phone and email on every directory and profile.
- Same founder and team names, with one consistent transliteration if you operate in more than one language.
- Cross-references: your site links to your profiles, your profiles link back to your site, and case studies on client sites name the same canonical brand.
- Old names retired properly: if you rebranded, the old name should appear only in “formerly known as” context, with a redirect from the old domain.
This is unglamorous work, and it pays off more than most expect, because it makes every existing mention count toward one entity.
How to measure mention growth next to Share of Voice
Mentions are a leading indicator; Share of Voice in AI answers is a lagging one. Track both monthly, on the same page of the report.
| Metric | Type | How to collect |
|---|---|---|
| New third-party mentions this month | Leading | Mention log: URL, date, source type, name spelled correctly (yes/no) |
| Share of mentions on cited domains | Leading | Cross the mention log with the citation column from the prompt runs |
| Entity consistency score | Leading | Percentage of profiles and pages using the canonical name and description |
| Share of Voice by assistant and funnel stage | Lagging | Monthly run of a frozen set of 20-40 prompts, or a tracker such as Otterly (from $29/month), Peec (€89-199/month) or Profound (from $99/month) |
| Citation source split: own domain vs third party | Lagging | Same prompt runs |
| Accuracy of facts in mentions | Lagging | Same prompt runs, tagged accurate / outdated / wrong |
What to watch over a quarter: mentions on cited domains rise first, then the third-party share of citations rises, then Share of Voice follows. If mentions grow but Share of Voice does not, check whether the new mentions are on domains assistants actually read, and whether the entity is consistent.
Do not measure this with referral traffic. ChatGPT often names a brand without a link, so traffic badly undercounts the effect.
How MOVE runs mentions inside the retainer
At MOVE, third-party mentions are a standing workstream in the AI visibility retainer, not a one-off campaign. Each month has a quota of PR placements and expert commentary, a list of case studies to get published on client sites, a directory and review hygiene check, a YouTube slot, and disclosed community participation by a named person from the client’s team or ours.
Every placement goes into a shared mention log that is crossed with the monthly Share of Voice run, so the client sees which sources are being cited and which are not. Contracts fix Share of Voice and assisted conversions, never a position, because according to AirOps’ The 2026 State of AI Search report, only 30 percent of brands stay visible in the next answer to the same query.
Six mistakes that burn a mention budget
- Buying links and calling them mentions. A link from a page that never names your brand in context does almost nothing for AI visibility.
- Publishing case studies only on your own site. The version on the client’s independent domain is what counts as a third-party mention.
- Undisclosed activity on Reddit and forums. Risk of a community ban, and assistants pick up the “brand caught astroturfing” coverage that follows.
- Inconsistent naming. Three spellings of the brand name produce three weak entities instead of one strong one.
- Chasing volume on domains nobody cites. Check the citation column first; one mention on a cited domain is worth many on uncited ones.
- Measuring by traffic. Use Share of Voice and citation split instead; links are optional in AI answers.
Next step
If your GEO plan is mostly about your own site, the fastest improvement comes from moving effort to where 85 percent of visibility is formed. Order an AI visibility audit from MOVE: you get a map of the third-party sources assistants already use for your prompts, and a 90-day mention plan built on it. Details on the AI advertising and GEO service page, or go straight to the contact page with your market and three competitors.