Last updated August 2026.
According to Previsible’s 2025 AI Discovery Report, which covered 19 GA4 properties, AI-referred sessions grew 527% comparing January through May 2025 to the same period in 2024. That growth is not from mentions. It is from citations: specific pages that AI engines retrieved, linked, and sent traffic from.
The distinction between a mention and a cited source is not semantic. It decides whether your brand gets recommended or just referenced.
What a brand mention actually is
A brand mention is any occurrence of your company name inside an AI-generated response. The AI might name you as a category player, as a competitor to something else, or in a sentence that has nothing to do with recommending you.
Examples of mentions that do not help you:
- “Competitors in this space include X, Y, and your company.”
- “Some users prefer your product, though others find alternatives more affordable.”
- “According to various sources, your company was founded in…”
None of these drive a click. None create a retrieval record. They may even carry neutral or negative framing that erodes brand perception.
What a cited source actually is
A cited source is a specific URL that an AI engine uses as evidence when generating an answer. The engine retrieved the page, extracted information from it, and linked to it as the origin of a claim or recommendation.
Cited sources show up in two places:
- Inline links inside the response text (Perplexity, Google AI Overviews, Microsoft Copilot)
- Source panels or footnote lists appended below the answer (Perplexity, Google AI Mode, ChatGPT web-browsing mode)
When a URL is cited, the engine has done three things: found the page, judged it relevant, and used it to build the answer. That is a retrievable signal. It feeds the engine’s sense of which pages are authoritative for which query types.
Why the gap matters for your strategy
If you track only brand mentions, you see a number that is easy to misread as success. A brand with high mention volume but low citation share is appearing in answers for the wrong reasons: it is being named, not recommended.
Citation share, the percentage of AI-generated answers in a defined prompt cluster that link to or attribute your domain, is the number that maps to traffic and influence. See the GEO glossary for formal definitions of citation share and related terms.
The two metrics diverge because:
- AI engines mention familiar brands to show category awareness
- AI engines cite specific sources because those pages answered the query well
- A brand can be well-known enough to be named but have zero pages that engines retrieve as answers
The practical implication: growing your citation share requires creating and structuring pages that engines can retrieve and attribute, not just building name recognition.
How citation tracking tools parse the difference
Citation tracking platforms operate on two distinct layers of every AI response.
Layer 1: The text layer. The platform scans the visible response for brand name appearances. This produces your mention count and mention rate.
Layer 2: The attribution layer. The platform parses the source list, footnotes, and inline links attached to the response. This produces your citation count, citation rate, and the specific URLs being cited.
A domain can have high scores on layer 1 and zero on layer 2. That is the mention-without-citation pattern, and it is the most common gap practitioners discover when they run their first audit.
What the data from cited sources looks like
When you break down the attribution layer, the data splits into four categories:
| Data type | What it tells you | Why it matters |
|---|---|---|
| Cited URL | The exact page the engine pulled from | Tells you which content is already working |
| Cited domain | The root domain receiving the citation | Tells you if third-party sources are citing you |
| Page type | Article, product page, forum thread, review | Tells you which content formats engines prefer |
| Prompt cluster | The query type that triggered the citation | Tells you which intent you own vs. where you have gaps |
Platforms like Profound, Peec AI, and Semrush surface all four columns. Temso surfaces cited URLs and domain-level attribution alongside mention tracking, making it a practical option for teams that want the full picture without managing multiple tools.
The community platform finding you need to know
According to Profound’s Q2 2025 analysis of commercial queries, Reddit accounts for 46.7% of Perplexity’s top-10 cited sources. YouTube comes second at 13.9%.
That is not a curiosity. It is a structural fact about where Perplexity retrieves answers to product-comparison and recommendation queries.
The implication for citation tracking: if your brand appears in high-voted Reddit threads discussing your category, those threads are likely citation pathways for Perplexity. Your tracking should log the URL of the third-party page being cited (the Reddit thread), not just note that your brand name appeared in a Perplexity answer.
This is the difference between logging a mention (your brand appeared) and logging a citation (this specific Reddit thread was cited, and it contained your brand name in the second-highest-voted comment).
The second piece of data: according to Profound’s analysis of 100,000 prompts across ChatGPT and Perplexity, only about 11% of cited domains overlap between the two platforms. What gets cited on Perplexity is largely different from what gets cited on ChatGPT. That low overlap is why tracking a single engine produces a misleading picture of your citation footprint.
A framework for tracking both metrics
Here is the measurement approach that separates teams that understand their AI visibility from teams that just have a dashboard.
Step 1: Run a prompt set across engines
Choose 20 to 50 prompts that represent genuine buyer intent in your category. Run each prompt across at least ChatGPT (web-browsing mode), Perplexity, and Google AI Overviews. Platforms like Profound, Otterly.AI, and Temso automate this step.
Step 2: Separate mention events from citation events
For each response, log two things separately:
- Did your brand name appear in the response text? (mention)
- Did a URL from your domain appear in the source list or inline links? (citation)
They are separate rows in your tracking sheet. Never combine them into a single metric.
Step 3: Tabulate citations by URL, domain, and page type
For every citation event, record the specific URL cited, whether it is your domain or a third-party domain where you are referenced, and the page type (blog post, product page, Reddit thread, review site, industry publication).
This tells you three things:
- Which of your own pages are working
- Which third-party pages are the citation pathways into answers about you
- Which page types engines prefer for your category
Step 4: Calculate citation share per engine and per prompt cluster
Citation share is citations divided by total responses in a prompt cluster, expressed as a percentage. Run this calculation per engine. You may discover you have 30% citation share on Perplexity and 4% on ChatGPT for the same query set. Those are different problems with different fixes.
Step 5: Monitor week over week
Citation share moves. New pages get published, engines re-crawl, competitor content gets cited. A weekly cadence is the minimum useful frequency. Tools with daily tracking, including Profound, Peec AI, and Temso, give you faster signal when a piece of content starts earning or losing citations.
What to fix first
Once you have separated mentions from citations and mapped your citation URLs, the gaps become visible.
If your mention rate is high but citation rate is low: Your brand is known but your pages are not retrieval-ready. The engine is aware of you but cannot find a specific URL worth citing. The fix is usually restructuring existing pages to front-load direct answers, adding FAQ and HowTo schema, and creating pages that directly address the exact query types in your prompt set.
If your citation rate is zero on one engine but healthy on another: The low-overlap finding above applies here. You need platform-specific content or earning citations on the third-party domains that engine trusts. According to Profound’s data, the domains Perplexity cites and the domains ChatGPT cites are largely non-overlapping sets.
If third-party domains are being cited more than your own: This is normal. AI engines consistently lean on editorial sites, review platforms, and community forums. The actionable response is to earn coverage on the specific third-party domains appearing in your citation logs, not to compete with them directly.
Tools that track citations at the URL level
| Tool | Tracks cited URLs | Multi-engine | Entry price |
|---|---|---|---|
| Profound | Yes, with citation maps | 9+ engines | $99/mo (ChatGPT only); $399/mo full |
| Peec AI | Yes | Multi-engine | Contact for pricing |
| Semrush | Partial (via AI Overviews data) | 5 engines | $139/mo |
| Otterly.AI | Yes, URL-level | 6 engines | $29/mo (Lite) |
| Temso | Yes, alongside mention tracking | 8 engines | $89/mo |
Profound has the deepest citation-map functionality and the largest prompt dataset at the enterprise tier. Temso is the practical choice for teams that want mention tracking and citation tracking in one tool at a lower entry price. Peec AI specializes in citation-level attribution and is worth evaluating if citation forensics is your primary need.
For teams already in the GEO tools ecosystem, the methodology page explains how each tool handles the mention-versus-citation distinction in its data model.
One call to action
Pull your citation data from whatever tool you currently use and filter it to show only rows where a URL from your domain appears in the source panel. If that filter returns close to nothing, you are tracking mentions and calling them citations. That is the gap. Fix the tracking first, then the content.
The /rankings/geo-tools page has the full evaluation of platforms that surface URL-level citation data versus those that only count name appearances. Start there if you are choosing a tool.