Claude cites about eight sources per answer, second only to Perplexity, and its five most-cited sites are all competitors' own domains. Among other sites it reads roundups and a larger share of articles than any other engine, cites no Reddit threads at all, and answered one question about AI visibility trackers with AI-text detectors.
This note looks at Claude alone; the cross-engine headline numbers are in what seven AI engines cite. Each number below says which of two lists it comes from:
- All hosts counts every cited site except cituna.com, including the domains of the competitors Cituna tracks and the engines' own properties.
- Target sites leaves out those two groups: what remains is publishers, directories, communities and the blogs of companies outside the tracked list.
How many sources Claude cites
Claude's 552 answers carried 4,377 third-party citations, or 7.9 per answer. That is second only to Perplexity (20.2); the other six engines average 8.0, a figure Perplexity pulls up. Anthropic says that when Claude uses the web, "it provides direct citations so you can easily fact check sources".
Claude cited 260 different target sites, second to Perplexity's 361 and above the other engines' average of 187. Between 63.8% and 67.4% of its third-party citations went to tracked competitors' domains or the engines' own properties.
The sites Claude cites most
Counting all hosts, the top five were aiclicks.io (171 citations), llmpulse.ai (170), rankability.com (116), geoptie.com (104) and getmint.ai (89). All five are tracked competitors' domains. Together they took 650 citations, 14.9% of the total, below the other engines' average of 17.9%.
| Site | Page kind | Citations |
|---|---|---|
| integrate.io | Roundup | 68 |
| writesonic.com | Roundup | 63 |
| blog.hubspot.com | Roundup | 43 |
| aiadvantageagency.com | Article | 37 |
| alhena.ai | Roundup | 37 |
| amplitude.com | Roundup | 37 |
| alternativeto.net | Review site | 35 |
| evertune.ai | Roundup | 34 |
| blog.timsoulo.com (tied) | Roundup | 32 |
| cmswire.com (tied) | Article | 32 |
| nohacks.co (tied) | Article | 32 |
The top 10 places took 418 citations: 9.5% of Claude's third-party total (the others average 15.2%) and 33.0% of the citations to its top 100 target sites, the lowest share of the seven (the others average 44.5%). By that measure, Claude spreads its citations to target sites more evenly than any other engine.
Two of the top 10, aiadvantageagency.com and alternativeto.net, were cited by no other engine. Across its top 100, the 29 sites only Claude cited took 17.5% of the citations, below the 21.3% average elsewhere.
What kinds of pages Claude reads
| Page kind | Sites | Citations | Share |
|---|---|---|---|
| Roundups | 61 | 889 | 70.2% |
| Articles | 31 | 279 | 22.0% |
| Review sites | 2 | 41 | 3.2% |
| Reference | 3 | 30 | 2.4% |
| Community | 3 | 27 | 2.1% |
| Video | 0 | 0 | 0.0% |
Claude gives articles, review sites and reference pages a larger share than any other engine: 22.0% for articles against a 15.1% average. The largest review and reference sites are alternativeto.net (35) and en.wikipedia.org (15). Its community share is 2.1%: business.quora.com (13), linkedin.com (10) and dev.to (4). Claude is the only engine of the seven that never cited reddit.com in the window. Reddit, YouTube and LinkedIn in AI answers has the comparison.
Where Claude misreads the question
| Subject of the cited pages | Question it answered | Sites | Citations |
|---|---|---|---|
| AI-text detectors | Small B2B SaaS tracker question | 7 | 24 |
| API, cost and LLM monitoring | AI visibility tool with an API for developers | 3 | 22 |
| How to cite ChatGPT | how to get cited by ChatGPT | 3 | 19 |
| What Grok itself costs | Small B2B SaaS tracker question | 2 | 6 |
That is 15 sites and 71 citations, 5.6% of the citations to Claude's top 100 target sites. Only Perplexity (9.7%) and Gemini (9.0%) have more; the other six engines average 3.3%.
The biggest misread is Claude's own. The question reads: "I run marketing at a small B2B SaaS. Which AI visibility tracker covers ChatGPT, Claude and Grok without enterprise pricing?" It drew 25 listings in Claude's top 100, more than any other question, and 10 of them are articles. Nine of those 10 are off the subject:
- Seven AI-text detectors: aidetector.us.com, gptcleanup.com and quillbot.com (four citations each), aichecker.ai, pangram.com, phrasly.ai and verva.com (three each). aichecker.ai, pangram.com and aidetector.us.com describe themselves on their home pages as AI-text detectors; the other four were cited for pages at /ai-detector or /ai-content-detector.
- Two pages on what Grok itself costs: mem0.ai (4) and aibusinessweekly.net (2).
The other misreads share words with the question but not its subject. For the API question, levo.ai (16) was cited for pages on API visibility and API monitoring, holori.com (3) for "AI cost visibility tools" and confident-ai.com (3) for LLM observability tools. For "how to get cited by ChatGPT", apastyle.apa.org (11) was cited for its page at /blog/how-to-cite-chatgpt, scribbr.com (5) for a guide to citing ChatGPT in APA, MLA and Chicago style, and a library FAQ titled "How do I cite ChatGPT or AI?" (3).
One mixed site is left out of the count: alternativeto.net (35), whose sampled pages are two Otter.ai listings and one for Profound.
Which questions Claude answers from which sites
Each site in our data lists up to three of the tracked questions it was cited for. Claude's top 100 target sites carry 155 such listings:
- The six "alternative to X" questions: 29 listings, 25 roundups and three review-site listings.
- Tool-choice questions (best tracker, price, Search Console, API, audits): 96 listings, 64 roundups and 24 articles.
- How-to and definition questions: 30 listings, 15 roundups and 11 articles.
The share-of-voice question ("how do I measure my brand's AI share of voice") is answered only from articles: five sites, all articles. The cheapest-tool question draws 21 listings, 18 of them roundups.
How Claude compares with the other six engines
| Measure | Claude | Other six, average |
|---|---|---|
| Third-party citations per answer | 7.9 | 8.0 |
| Target sites cited | 260 | 187 |
| Top 5 of all hosts, share of third-party citations | 14.9% | 17.9% |
| Top 10 target sites, share of third-party citations | 9.5% | 15.2% |
| Top 10 target sites, share of top-100 target citations | 33.0% | 44.5% |
| Articles, share of top-100 target citations | 22.0% | 15.1% |
| Community and video, share of top-100 target citations | 2.1% | 16.5% |
| Top-100 target citations to sites no other engine cited | 17.5% | 21.3% |
| Top-100 target citations to misread pages | 5.6% | 3.3% |
For a brand in this category, Claude is the engine where articles and reference pages take the largest share, and where Reddit threads took none (how to get cited by Claude). Its misreads also suggest phrasing tracked questions carefully: a question that stacks several engine names and a price condition drew the most off-topic pages (how to choose the prompts to track).
Method and limits
What was counted. A citation is one source URL shown in one answer, grouped by site. Third-party figures leave out cituna.com. The target list also leaves out the tracked competitors' own domains and the engines' own properties. Claude's file ranks its 100 most-cited target sites, with ties at the cutoff of two citations broken arbitrarily; the 63.8% to 67.4% range comes from the 160 target sites outside it having between one and two citations each. Page kinds are the labels in Cituna's data. Each site lists up to three questions and up to three sample page addresses, and misreads were judged from those addresses, so the misread count is a floor.
The window. 33 scans in the 30 days to 22 September 2026, which produced 552 Claude answers. "Other six, average" is the plain average of the six other engines' figures.
The questions. They are about one software category, AI visibility tools, and they are the questions Cituna tracks for itself, listed in the cross-engine note.
What the numbers can and cannot show. They show which sites Claude cited for these questions in this window. They cannot show why it chose them, how it behaves in other categories or markets, or what any one user sees, because answers vary by account, location and time. Site-level counts cannot separate one page from another on the same site. The full method is on the methodology page.