AI Mentions vs Citations: Semrush Finds Fewer Than Half of Top Brands Overlap

AI mentions vs citations is now a measurable split. Semrush's 2026 AI Visibility Index, released October 13, found that fewer than half of the 100 most-mentioned brands also rank among the 100 most-cited domains. That held in every month from January to July 2026, in both the US and the UK. In July, ford.com ranked 18th for ChatGPT mentions and 1,055th for ChatGPT citations. If your dashboard reports one visibility number, it hides which of two separate problems you have. This article covers what Semrush measured, why the gap exists, and how to diagnose yours.

What did Semrush's AI Visibility Index find about AI mentions vs citations?

Fewer than half of the top 100 mentioned brands appeared in the matching top 100 cited domains, in every month and both markets Semrush measured. The index covers seven monthly measurements from January to July 2026, across ChatGPT, Gemini, Google AI Mode, and Google AI Overviews, sorted into 22 industry categories.

Semrush ran fixed monthly prompt pools of 25 million US prompts and 2.5 million UK prompts through ChatGPT, Gemini, and AI Mode. For AI Overviews it used a separate, larger set of real search keywords: roughly 48 to 55 million US keywords per month and 4.4 to 4.8 million UK keywords.

The report defines a mention as a brand named in the answer text. A citation is a domain listed as a source. Neither one implies a recommendation.

UK overlap ranged from 36.7 to 45.3 of 100 depending on the month. US overlap ranged from 41.4 to 47.4. Track this metric for your brand → nobori.ai

How big is the mention and citation gap on each platform?

ChatGPT shows the widest gap. UK overlap on ChatGPT ran from 28.7 to 32.3 of 100, and US overlap from 32.8 to 35.8. Gemini in the UK landed between 33.4 and 36.9.

Google AI Mode and AI Overviews sit closer to the line. AI Mode fell below half in 5 of 7 months in each market. AI Overviews fell below half in 4 of 7 UK months and 2 of 7 US months.

The lesson for your reporting: a brand that looks strong on one engine can look average on another, and the cause differs by engine. Semrush also found that AI Overviews had the lowest brand-mention overlap with the other three platforms in all seven months. Its pairing with ChatGPT had the lowest source overlap of any platform pair in both markets.

Treat each engine as its own channel with its own mention rate and its own citation rate. Averaging them erases the signal you need to act on.

What does the Ford example show about AI citations?

Ford shows how far mentions and citations can drift apart. In July 2026, ford.com appeared in the top 100 mentions on all four platforms. Its citation ranks on those same platforms were far lower:

• ChatGPT: 18th for mentions, 1,055th for citations
• Gemini: 17th for mentions, 1,146th for citations
• Google AI Mode: 23rd for mentions, 1,320th for citations
• Google AI Overviews: 14th for mentions, 748th for citations

AI engines name Ford constantly and rarely list Ford's own site as a source. The supporting pages come from somewhere else, and the Semrush coverage points to news, information, and marketplace sites as heavy citation winners.

The UK shows the same pattern between two retailers. In July, Tesco ranked 25th among ChatGPT's cited UK domains and Asda ranked 606th. Both sat in the top 100 for mentions on all four platforms in every month. On Gemini, Tesco ranked 29th and Asda 60th for citations.

Same visibility in the answer text, very different visibility in the source list.

Why do AI engines mention brands they never cite?

Engines can mention brands from model memory and cite pages only through retrieval, and the two systems do not share a ranking. A brand with years of coverage across the web enters the model's training data and gets named without a lookup. A citation requires the engine to run a search, fetch a page, and decide that page supports a claim.

Ahrefs published data on October 12 that shows how often retrieval happens. In 66.6 million logged-out ChatGPT responses from August and September 2026, only 14.4% ran a web search. Searching responses averaged about 1.6 fan-out queries. Ahrefs notes the free model skips searches when it can answer from existing knowledge. That sample covers one free model, so paid and reasoning models likely search more.

An answer that skips retrieval can name brands but cannot cite pages. That is one plausible source of the gap. Semrush does not test this cause directly.

Track this metric for your brand → nobori.ai

Which sources do AI engines cite instead of brand sites?

News and information sites take a large share of citations, especially on ChatGPT. Among ChatGPT's top 100 cited domains, news sites made up 29.5% of citations in the UK and 19.6% in the US. On Gemini in the UK, news was the leading category every month, at 16.6% to 23.0%.

After news, ChatGPT's UK list favored trip-planning sites and online marketplaces. In the US it favored online learning and streaming platforms. Search Engine Land's coverage adds that AI Overviews leaned on streaming, business software, online learning, and medical-reference sites in the US.

Source concentration also differs. In the UK, the top 10 domains took 23.2% to 35.8% of AI Mode citations among its 10,000 most-cited domains. On ChatGPT the top 10 took 12.8% to 15.4%.

AI Mode rewards a small set of dominant domains. ChatGPT spreads citations across a wider field, which leaves more room for a third-party page about you to get picked up. Our earlier post on third-party mentions and AI citations covers how to earn those placements.

How much do the four AI platforms disagree on brands?

They agree on a core and diverge everywhere else. Any two platforms shared 63 to 70 of their top 100 mentioned brands in the UK and 70 to 78 in the US, on average.

The agreement shrinks when you require all four. Semrush calls brands that sit in the top 100 mentions on all four platforms in every month the Universal Core. It holds 24 domain records in the UK and 36 in the US. Fifteen appear in both markets, including Adobe, Apple, Amazon, Google, LinkedIn, Microsoft, Reddit, TikTok, and YouTube.

Gemini names more distinct brands per answer than ChatGPT. ChatGPT lists the most source entries per answer, averaging 12.5 to 22 in the UK, against 1.6 to 4.4 for Gemini. Those counts are URL entries, not unique sites, and they include answers with no citations.

If you are outside the Universal Core, you compete for a different set of slots on each engine. Our post on cross-engine citation overlap shows the same split from another dataset.

Does the mention and citation gap apply to B2B brands?

Possibly, but Semrush's named examples come from consumer categories. Ford, Tesco, Asda, and Investing.com are not B2B software vendors, and the coverage we reviewed publishes no B2B-only breakdown. Treat the direction as a hypothesis for your category and the exact percentages as category-specific.

The test is simple. Ask ChatGPT for a shortlist in your category, then compare the vendors named in the answer text with the domains listed as sources. If review sites, trade publications, or Wikipedia fill the source list while your site is absent, you have the same gap Ford has.

Our post on ChatGPT's shrinking B2B brand shortlists shows why a mention matters: fewer slots means each named brand carries more weight.

Run your own category through both lists before you assume Semrush's numbers apply to you. Track this metric for your brand → nobori.ai

What are the three positions a brand can hold in AI answers?

Search Engine Land's coverage of the index frames three positions: you are mentioned, your own site is cited, or third-party sites carry the citation. Each position needs a different fix, so locate yours first.

• Mentioned and cited: the engine names you and links your site. Protect it with fresh, structured pages.
• Mentioned, not cited: the engine knows you but sources its claims elsewhere. Ford and Asda sit here.
• Cited, not mentioned: the engine uses your page as evidence but names someone else. Investing.com in US Finance on ChatGPT sits here, ranked 7th among cited domains and 50th among mentioned brands.

A fourth position, neither mentioned nor cited, needs the same diagnosis plus a longer timeline.

The same brand can hold a different position on each of the four platforms. Score each platform separately.

How do you diagnose your own mention and citation gap?

Pull the same prompt set across all four engines and record two fields for each answer: whether your brand appears in the text, and whether your domain appears in the sources. Then compare the lists.

1. Build 30 to 50 prompts that match how buyers search your category, split across informational, comparison, and "best for" intent.
2. Run every prompt on ChatGPT, Gemini, Google AI Mode, and AI Overviews.
3. For each answer, log mention (yes or no), citation of your domain (yes or no), and the three most-cited third-party domains.
4. Calculate your mention rate and citation rate per engine.
5. List the third-party domains that appear most often when you are mentioned but not cited.

The gap between your two rates is the number to watch. A mention rate of 40% with a citation rate of 5% points to an off-site problem. A citation rate that exceeds your mention rate points to an entity problem.

Repeat monthly. Semrush measured seven separate months because the numbers move.

What should you do when you win the mention but lose the citation?

Make the third-party pages that engines already cite say accurate things about you, and publish pages worth citing on your own domain. Engines already cite those outside sources more often than your homepage.

Start with the domain list from your diagnosis. For each frequently cited third-party page, check whether it describes your product correctly, lists current pricing, and includes the use cases you want associated with you. Request corrections or updates where it does not.

On your own site, publish content that answers the specific claims engines make about you: pricing pages with real numbers, comparison pages with named competitors, integration documentation, and original data. Engines cite pages that contain a checkable fact.

Ahrefs' October 12 study found ChatGPT citations tilting toward primary sources, with PubMed rising from 64th to 4th among cited domains between March and September 2026. Original data on your own domain fits that pattern.

Your branded search results deserve the same audit, since Wikipedia, YouTube, and LinkedIn often fill the source list there.

What should you do when you win the citation but lose the mention?

Strengthen your entity signals so the engine connects your cited page to your brand name. A page that answers the question but never states who published it gives the model a fact without an attribution.

• Put your brand name in the title, first paragraph, and author byline of cited pages.
• Add Organization and Person schema with consistent names across your site and profiles.
• Keep your name, category description, and product naming identical on LinkedIn, Crunchbase, G2, and Wikipedia where you have a presence.
• Write the category claim in a sentence the model can copy: "[Brand] is a [category] for [audience] that does [specific job]."

Our guide to entity optimization for AI search walks through the full checklist.

Start here if the diagnosis shows this position, because the engine already uses your page. You are closing an attribution gap.

What are the limits of the Semrush mention and citation data?

Read the index as a measure of recurrence, not of commercial results. Semrush states it does not link visibility to revenue, does not attribute changes to model updates, and weights the four platforms equally rather than by usage.

Four more limits apply:

• The data ends in July 2026, so it predates the September changes in ChatGPT citation behavior that Ahrefs documented.
• The index excludes Claude, Perplexity, and Microsoft Copilot.
• It measures top-100 lists, so a brand ranked 101st is invisible in the overlap math.
• Semrush sells AI visibility and GEO tools, which gives it a commercial interest in the framing.

The release also repeats an Adobe claim that AI-referred visitors convert 60% better. Coverage of the report notes that neither document states the source of that figure, so we do not rely on it here.

Use the index as a prompt to measure your own numbers. Do not use it as a benchmark for your category. Our comparison of AI citation monitoring tools covers which products can report both metrics per engine.

What should you do this week about AI mentions vs citations?

Run the diagnosis, pick one gap, and fix one source. Here is a five-day plan.

1. Monday: build your 30 to 50 prompt set and run it on all four engines.
2. Tuesday: log mention and citation for each answer and calculate both rates per engine.
3. Wednesday: list the top ten third-party domains cited in answers where you appear.
4. Thursday: audit the three most important of those pages for accuracy and request updates.
5. Friday: publish or refresh one page on your own domain that states a checkable fact about your product.

Then set a monthly reminder to rerun the same prompts. Your baseline is the only way to see whether your changes worked.

One number will not tell you where you stand. Two numbers per engine will.

Get daily AI visibility alerts for your brand → nobori.ai

Sources: Semrush, 2026 AI Visibility Index release (October 13, 2026); Search Engine Land, "Your brand can win the mention and lose the citation" (October 14, 2026); Relevant Audience coverage of the Semrush index (October 15, 2026); Ahrefs ChatGPT fan-out and citation study (October 12, 2026).