An AI answer can name your brand while linking to someone else's website. Semrush's AI Visibility Index 2026 shows how often that happens, and why mentions and citations need to be measured as two separate metrics.
In AI search, a mention means an AI answer names a brand, and a citation means a domain is linked as a source. Semrush tracked both across ChatGPT, Google AI Overviews, Gemini and AI Mode in the UK and US from January to July 2026. They rarely line up: the top-100 mention and citation lists shared fewer than half their domains in every month measured.
Mentioned but not cited is common; the reverse is rare. Each AI platform cites a different set of sources, and categories differ more than platforms do. Measure mentions and citations as two metrics, at stated list depths, with normalized domains and recurrence counts.
Mentioned vs Cited: What Each Metric Records
The AI answer names your brand. It signals brand salience inside the answer, not a recommendation or positive sentiment.
Your domain appears as a source link. It shows your pages are used as evidence, not that the platform trusts or endorses you.
These are different positions with different causes and different fixes. For how citations compare with classic rankings, see Keyword Rankings vs. AI Citations.
Why a Single AI Visibility Score Breaks Your Audit
A blended score averages two signals that move independently. Ford shows how far apart they can sit. In the US in July, ford.com ranked between 14th and 23rd for mentions on every platform, and between 748th and 1,320th for citations.
| ford.com, US, July 2026 | Mention rank | Citation rank |
|---|---|---|
| ChatGPT | #18 | #1,055 |
| Gemini | #17 | #1,146 |
| AI Mode | #23 | #1,320 |
| AI Overviews | #14 | #748 |
A client with that profile looks healthy on a combined dashboard. Split the measures and you find a source problem that a brand campaign will not fix.
What the Semrush Data Shows
The Index covers 2.5 million UK and 25 million US prompts per month on ChatGPT, Gemini and AI Mode, plus separate keyword pools for AI Overviews (4.35 to 4.80 million UK, 48.34 to 54.60 million US), across 22 categories.
Different source positions
Tesco and Asda were top 100 for mentions everywhere. In July, ChatGPT cited Tesco 25th and Asda 606th; Gemini ranked them 29th and 60th.
Mentioned, not cited
Top 100 for mentions on all four platforms, outside the top 500 for citations. Chevrolet, Disney, Ford, Honda, Marvel, Nintendo and Toyota held it in the US every month.
Cited, not mentioned
Investing.com ranked 7th as a source and 50th as a brand in US Finance on ChatGPT. McMaster-Carr showed the same shape in Industrial.
Rare
Only Apple and YouTube stayed top 100 on both lists, every platform, every month, with a rank gap of 25 or less, in both markets.
Category examples make the pattern concrete. In UK Finance, Lloyds Bank was on all four mention lists in July but no source list, while MoneySavingExpert and Wise were cited on all four platforms every month. Real Estate has stable source cores in both markets; Industrial and Food and Drinks have none.
Before You Measure: Prompt Set and Crawler Access
Two prerequisites decide whether your mention and citation data means anything. Both have their own guides:
- A fixed prompt set built from real customer questions. See How to Build an AI Search Prompt Set You Can Actually Measure.
- Verified crawler access per user-agent, confirmed in server logs. See AI Crawlers vs Traditional Search Crawlers and How to Check if Google, Bing and AI Bots Can Crawl Your Website.
- No 403s for the bots you want. See Fix 403 Errors for AI Crawlers.
If a site is mentioned but never cited, rule out blocked access before you touch content.
How to Log Mentions and Citations as Two Metrics
Keep one log with two separate columns: whether the brand is mentioned, and which URLs are cited. A single "AI visibility" field cannot show the One-Way Door pattern. For the manual checking process, see How to Check If Your Site Is Being Cited by ChatGPT or Perplexity.
| Date | Platform | Prompt | Brand mentioned | Competitors mentioned | Cited URL | Cited domain (normalized) |
|---|---|---|---|---|---|---|
| 2026-10-19 | ChatGPT | best site audit tool for agencies | No | Tool A, Tool B | example.com/review | example.com |
Sort cited URLs into three buckets: your domain, third-party pages that mention you, and third-party pages that leave you out. The third bucket is your investigation list.
Normalize domains before you count
The Index merges bare and www forms of the same site but keeps country domains separate. In the UK, ford.com met the One-Way Door test in five months and ford.co.uk in the other two. Track only one address and the pattern disappears.
- Merge www and bare formsCount
www.example.comandexample.comas one record. - Keep ccTLDs separateRecord
.co.ukor.com.phon their own, then roll up at company level only in reporting. - Record subdomains separately
blog.,help.andshop.are often cited on their own.
How to Read Mention and Citation Data
The Semrush methodology is a good model for client reporting because it avoids the usual traps.
- Compare within a platformPrompt pools and answer formats differ, so use ranks, within-platform shares and overlap at stated list depths, never raw totals across platforms.
- State the list depthIn the July depth test, the US One-Way Door group grew from 7 domains at a top-50 cutoff to 136 at top 500. Always write the depth beside the number.
- Report recurrence, not one runOne appearance is a lead. Five of seven measurement periods starts to be a position.
- Benchmark the category firstThe spread between categories was wider than between platforms in every US month and five of seven UK months.
- Do not claim growth from recurrenceA range is the lowest and highest month, not a trend line, and a visibility change does not prove a revenue change.
Where Mentions and Citations Split, Platform by Platform
ChatGPT and AI Overviews were the least-overlapping source pair in every month, sharing about 42 to 44 of 100 leading domains. Each engine needs its own check.
| Platform | Source profile in the Index | Audit focus |
|---|---|---|
| ChatGPT | 12.5 to 22 source links per answer; news is the top source type (29.5% UK, 19.6% US mean) | Which publications it cites for your prompts; accuracy of third-party descriptions |
| AI Overviews | Streaming (mostly YouTube), news and medical reference lead; most single-platform domains | Classic ranking and indexing for the keyword form; video coverage; YMYL accuracy |
| Gemini | Most brands per answer, fewest sources (1.6 to 4.4); mix changes by market | Mention presence and description; do not judge it by citations alone |
| AI Mode | Top 10 domains hold 23% to 39% of citation weight; Facebook, Instagram, Pinterest and Quora in its top 100 every month | Short list of dominant sources; community and social profile coverage |
Because AI Mode concentrates citations, a top-10 source audit covers much of its activity. ChatGPT spreads citations wider, so review at least the top 100 sources for your category.
Diagnose the Gap: Pattern, Cause, Fix
| Pattern in your log | Likely cause | First fix to test | Owner |
|---|---|---|---|
| Mentioned, never cited, pages absent from logs | Crawler blocked at robots.txt, CDN or WAF | Fix access, then re-run the same prompt set | SEO and engineering |
| Mentioned, never cited, bots fetch your pages | Pages lack specific, current facts, or answers sit behind JavaScript | Add direct, crawlable answers; check raw HTML | Content and SEO |
| Cited, rarely mentioned | Cited page does not make the brand's role clear | Clarify entity naming and context on the page | Content |
| Competitors cited on third-party pages without you | Coverage gap on reviews, listicles or communities | Assess audience fit, then test outreach | PR and partnerships |
| Third-party page describes you inaccurately | Outdated or wrong source data | Request a factual correction | Communications |
| Results swing between reports | Domain forms or list depth not held constant | Fix normalization rules and stated cutoffs | Analytics and SEO |
Reporting Template for Clients
Report mentions and citations in separate columns, with the conditions stated beside them.
| Platform | Mention position | Citation position | Prompts tested | Measurement date | Next check |
|---|---|---|---|---|---|
| ChatGPT | |||||
| Gemini | |||||
| AI Mode | |||||
| AI Overviews |
Add a header line with market, language, category, prompt set version, list depths and domain normalization rules. Track business outcomes (AI referral sessions, branded search, conversions) in a separate table, so a visibility change is never presented as proof of a revenue change. For the business case to share with clients, see the remar.digital guide on AI visibility: cited vs mentioned.
FAQ
- What is the difference between cited and mentioned in AI search?
- A mention means an AI answer names a brand. A citation means a domain is linked as a source. Semrush's 2026 Index found fewer than half of top-mentioned brands also rank among top-cited domains.
- Why is my brand mentioned in AI answers but not cited?
- Usually one of three reasons: crawlers cannot fetch your pages, your pages lack the specific facts the answer needs, or third-party pages cover the topic better. The Index shows this pattern is common for well-known brands such as Ford, Toyota and Disney.
- What list depth should I use when reporting AI mentions and citations?
- State it every time. Semrush used top 100 for mentions and top 500 for citations in its One-Way Door test. Pick one depth per metric and keep it fixed between reports.
- Why do my AI visibility results change between reports?
- Usually because list depth, domain normalization or the prompt set changed. Hold those constant and report how many runs a finding holds.
Source: Semrush (An Adobe Company), AI Visibility Index 2026: Where the top brands stand in the UK and US, data January to July 2026, ai-visibility-index.semrush.com.