Most sites missing from Claude citations are failing at one of six checkpoints: crawler access, Brave Search indexing, content tone, authorship signals, content depth, and, most often, simply not knowing whether they're already being cited, since Claude offers no native tracking at all. Claude's web backend runs on Brave Search rather than Google or Bing, and its citation preferences lean more heavily on tone and sourcing than either Google or ChatGPT. That means some of the fixes here are different from the first two posts in this series, even where the checkpoints look similar.
Same crawler confusion, different bots
If you read our ChatGPT diagnostic post, this will look familiar. Anthropic runs the same kind of split OpenAI does: separate bots for training versus citation, with similar names and very different jobs. Per Anthropic's own documentation, ClaudeBot collects content for model training. Claude-User fetches a specific page when a person asks Claude to look at something. Claude-SearchBot is the one that indexes your content for citation-relevant search results, block it, and Anthropic says explicitly that it "may reduce your site's visibility and accuracy in user search results."
The same July 2026 Inspeccia audit referenced in our ChatGPT post found the identical pattern here, slightly more pronounced: of the 100 sites (out of 1,567 audited) that fully blocked ClaudeBot, 96 of them (96%) never touched Claude-SearchBot at all. They opted out of training and left citation eligibility completely untouched, almost certainly without realizing there were two separate decisions to make (Inspeccia).
One genuine difference worth knowing: Anthropic states that all three of its bots, including the live-fetch Claude-User, honor robots.txt. That's not universal: OpenAI and Perplexity both note that their user-initiated fetchers don't reliably respect it. So a correctly configured robots.txt carries more weight with Anthropic's bots than it does elsewhere. Check yours for Claude-SearchBot specifically; if it's absent, you're allowed in. If Anthropic's crawlers are still not reaching you after that, cross-reference your source IPs against Anthropic's published crawler IP list, a firewall or CDN can block a bot with a clean robots.txt sitting right next to it.
1. Is your content indexed by Brave Search at all?
Claude's web search runs on Brave Search, not Google or Bing, confirmed via Anthropic's own subprocessor disclosures (reported by TechCrunch, March 2025) and already covered in our AI search optimization guide. This is the step most teams skip entirely, because Brave Search doesn't offer anything like Google Search Console or Bing Webmaster Tools: there's no dashboard to check. The practical workaround: search site:yourdomain.com directly in Brave Search to see what's already indexed, and use Brave's Submit URL page to hand it your homepage (Brave's crawler then discovers further pages by following your internal links, so a clean sitemap referenced in robots.txt still helps even without a formal submission tool for it).
2. Does your content read like a source, or like a pitch?
This is the sharpest divergence from the first two posts in this series. Overtly promotional language measurably reduces Claude's confidence in a page as a neutral reference: it's built to write in an informative, encyclopedic register itself, and it appears to weigh content the same way. A page that reads like marketing copy, even if every claim on it is accurate, is a weaker citation candidate than the same information written in plain, balanced prose. If your target page was written by (or optimized primarily for) a sales or conversion goal, that's worth a rewrite pass before touching anything technical.
3. Are your claims attributed to named, verifiable sources?
Claude places real weight on claims backed by named, credible sources: quote primary documents directly and link to them rather than paraphrasing secondhand summaries. The same goes for your own byline: content from a named author with visible credentials reads as more trustworthy than unattributed or anonymously published pages. If your target page has no author, no date, and no linked sources for its factual claims, it's asking Claude to trust it on faith alone.
4. Is there anything here Claude hasn't already seen a hundred times?
Claude is comparatively good at identifying thin or templated content, including AI-generated filler that restates common knowledge without adding anything new. Every paragraph on the page should carry a specific, verifiable claim or an original point, not just be present to hit a word count. This is also where structured data won't save you: a cross-platform test by searchVIU checking whether ChatGPT, Claude, Perplexity, Gemini, and Google AI Mode use JSON-LD schema during live retrieval found that none of them do: all five extracted only the visible HTML content on the page. Whatever structure or signal you want Claude to pick up on has to be in the text a person would actually read, not hidden in markup.
5. Are you checking whether any of this is working?
This is the one place Claude is genuinely harder to measure than the other two platforms. Google now has a (limited) native impressions report, and ChatGPT appends a UTM parameter to its outbound links. Claude does neither: there's no consistent referrer data and no first-party reporting on citations at all. The only ways to know whether Claude is citing you are running your own manual test queries against Claude directly, or using a monitoring tool that does that systematically across your target prompts. If you've never checked, the honest starting assumption isn't "Claude doesn't cite us," it's "we don't actually know," which is a different problem with a different fix.
That measurement gap is the specific thing Citation Hawk runs against every week: your tracked prompts, tested against Claude (and Google AI Overviews and ChatGPT), with a report on whether you were cited, mentioned by name without a link, or absent entirely. It doesn't replace fixing the five issues above: it's the feedback loop that tells you whether they worked.
Putting it together
Start with the crawler check, since it's fast and rarely the actual problem, but confirm it anyway, since the failure mode (blocking training, forgetting search) is common and easy to miss. Then work outward: Brave indexing, tone, sourcing, depth, and finally, whether you're measuring any of it at all.
For the full technical checklist (covering Claude, Google AI Overviews, and ChatGPT Search in one place), see our guide to getting your website into AI search results.
Tom Eastwood is the founder of Citation Hawk, a tool that monitors AI citation rates across ChatGPT, Claude, and Google AI Overviews.