Mentioned but Not Cited: The Two Scoreboards of AI Visibility

Semrush’s 2026 AI Visibility Index (an analysis of 126 million US AI search prompts collected January through April 2026, covering more than 1,200 brands across 22 industries on ChatGPT, Gemini, Google AI Mode and Google AI Overviews) treats the two as structurally separate metrics for exactly this reason. A brand can be mentioned constantly and cited almost nowhere. It can also be the reverse: cited constantly while almost never being the brand the AI is actually discussing. The study’s cleanest example of that inverse is Wikipedia, quoted heavily as evidence yet rarely the subject of the conversation. Some sites push this to extremes: Medical News Today is cited 4.6 times for every time it’s mentioned by name, and Healthline runs at 4.2 to 1. These sites have become the infrastructure AI systems use to construct answers about everyone else.

If you sell something, you want both. Mentions are the recommendation; citations are the receipts. We covered how the engines decide which sources deserve those receipts in our piece on the trust signals AI uses to decide who to believe.

This piece is about the scoreboard those signals feed.

Why does AI mention your brand but never cite your website?

Because the two are earned through different mechanisms. Mentions come from brand authority and category fit: whether the model associates your name with the problem being asked about. Citations come from content depth: whether your pages contain clear, retrievable, quotable material the system can lean on as evidence. Strong brand, thin content gets you mentioned without citations. Deep content, weak brand gets you cited without mentions.

The Patagonia case in the Semrush study shows how far mention-side strength can be carried by other people’s websites. Patagonia held an AI visibility score of 79 or 80 in every month of the study, roughly 87,000 mentions across platforms, against just 8,300 cited pages. The mentions weren’t coming from patagonia.com. They were generated by a network of third-party outdoor-gear sources: 38,800 brand mentions on Reddit, 18,500 on OutdoorGearLab, 12,900 on REI. Six specialist sites collectively produced more than 65,000 Patagonia mentions, which the study notes exceeds the combined output of all the Tier 1 traditional media it tracked. AI systems recommend Patagonia because the review ecosystem talks about Patagonia, not because they’re quoting the brand’s own pages.

Shopify shows what balance looks like. Its visibility score held at 81 with 147,000 mentions against 42,100 cited pages, and on Google AI Overviews in April 2026 it was mentioned 45,098 times and cited 46,342 times: a near one-to-one ratio the study describes as rare. That balance is built, not lucky: review platforms like G2 and Capterra feeding the mention side, plus 222 individual documentation topics each scoring above 90 on the visibility scale feeding the citation side. The documentation is the part most brands skip.

How big is the gap between mentions and citations?

Big enough to change strategy, and it varies sharply by platform. Semrush measured what it calls mention-source overlap: how much the list of brands an AI names lines up with the list of domains it quotes. On Google AI Overviews the overlap ran as high as 64%; on Gemini it fell as low as 30%. That’s a 34-point spread. On AI Overviews, the brands being discussed are mostly also the brands being quoted. On Gemini, they’re largely different groups, which means a Gemini strategy built only on earning mentions leaves the entire evidence layer to someone else.

The citation math differs just as much. ChatGPT cites an average of 15.4 sources per response, the most of the four platforms, drawing heavily on Reddit and Wikipedia. Google AI Mode averages 11.4, AI Overviews 9.2, and Gemini just 3.3: a fraction of the citation slots, concentrated on a much smaller pool of sources. Mention slots run in the opposite direction: AI Mode names the most brands per answer at 5.2, followed by Gemini at 4.7, AI Overviews at 3.7 and ChatGPT at 3.3. So the platform with the most room to be cited has the least room to be mentioned, and vice versa.

None of this holds still, either. Similarweb research from November 2025 found the set of citations across AI platforms changes by roughly 50% month to month, with only 11% overlap in citations between the major platforms. And the Semrush study caught a warning sign even for the best-positioned source on the internet: Wikipedia’s mentions declined by about 50% between January and April 2026 while its citation volume held strong: a hint that AI systems are getting better at synthesizing information without naming where it came from.

How do brands measure visibility in AI answers?

By tracking four things separately, per platform: how often you’re mentioned, how often you’re cited, how much the two overlap, and how consistent all of it is across platforms. That’s the framework the Semrush study lands on, and its companion survey of 481 marketers explains why it’s worth spelling out: 45% said they cannot properly measure their brand’s visibility in AI-generated answers at all, and only 9% can track all four. Adobe’s own April 2026 survey of more than 500 marketers was bleaker still: 98% lacked a clear, documented roadmap for AI search optimization.

A workable measurement loop looks like this:

1. Build a fixed prompt set. The questions your buyers actually ask, run the same way every time, so movement means the answers changed — not your questions.
2. Log mentions and citations as separate columns. A run where you’re named twice and quoted zero times is a different problem from the reverse, and averaging them into one “visibility score” hides which problem you have.
3. Track per platform, not in aggregate. With overlap ranging from 64% to 30% and citation density from 15.4 to 3.3, a blended number across ChatGPT and Gemini is describing a platform that doesn’t exist.
4. Watch trend, not snapshot. If citation sets are churning roughly 50% monthly, a single measurement is a photograph of weather. The monthly series is the climate.

One further reason to keep the columns separate: the failure modes compound. A brand that’s mentioned but never cited has no control over what the evidence says about it, which is exactly the condition under which AI systems get your brand wrong and nobody notices until a prospect repeats the error back to you.

How do you earn more brand mentions in AI?

Mentions are won off your website more than on it. The model names your brand because the sources it trusts keep associating you with the category, so the work is seeding accurate, consistent descriptions of what you do across the third-party surfaces AI systems actually read: review platforms, community discussion, comparison sites, industry publications.

The study’s platform source-mix data tells you where that effort lands. ChatGPT leans on community and reference platforms: Reddit and Wikipedia above all. Google AI Mode leans social and local, with Facebook, Instagram and Yelp prominent; Yelp reached sixth place in AI Mode’s most-mentioned brands, the only local-discovery platform to rank that high anywhere in the study. AI Overviews is dominated by YouTube, its single largest source at 35.2 million mentions. If your category gets discussed on review sites, that’s where your mention rate lives. If it gets demonstrated on YouTube, video is your mention channel: a dynamic we unpacked in how YouTube videos earn AI citations.

Two cautions from the data. First, consistency across sources matters as much as volume: the study frames Patagonia’s steadiness as the product of near-identical descriptions across a half-dozen independent sites, and the brands that hold visibility everywhere (only 36 of 1,200+ managed it on all four platforms, every month) share broad reach and a clear, stable role in their category. Second, concentration is a risk: Cleveland Clinic draws 86% of its AI mentions from Google AI Overviews alone. One platform re-weighting its health sources could move most of that brand’s AI presence overnight.

How do you earn more AI citations of your website?

Citations are won on your website. The system quotes pages that answer a specific question cleanly, structured so the answer is easy to lift: a direct claim up front, supporting detail underneath, headings that match how people ask, and depth across the topic rather than one page trying to carry everything.

Shopify’s 222 documentation topics scoring above 90 are the pattern to copy at whatever scale you operate: many pages, each owning one narrow question, rather than a handful of broad ones. The Source Surplus league table: Medical News Today at 4.6x, Wikipedia 4.3x, Healthline 4.2x, WebMD 4.0x: is a list of sites built exactly this way: comprehensive, plainly structured reference content on tightly scoped questions. None of them got there with clever copy. They got there with coverage and clarity.

Match the effort to the platform economics. ChatGPT’s 15.4 citation slots per answer make it the most winnable citation surface for a mid-sized brand; Gemini’s 3.3 slots, concentrated on YouTube, Wikipedia and major commerce platforms, are close to spoken-for. And keep the integration lesson from the survey data in view: 81% of teams that run SEO and AI visibility as one workflow reported more traffic or leads from AI platforms, against 36% of teams running them separately. Citation-earning content is not a new discipline bolted onto the side: it’s the same structured, answer-first work that’s been winning search features all along, pointed at a new reader. That’s the thesis of our whole AI search practice: one workflow, two scoreboards.