Case study · published 19 May 2026

SourceScore: 480:1 bot-to-human ratio is the signal, not the problem.

SourceScore.org is a niche reference site that scores the citation behavior of large language models. It serves ~8 human unique visitors a month — and ~3,840 AI crawler hits in the same window. That's a 480:1 bot-to-human ratio.

For a long time, that ratio looked like a problem to traditional analytics minds — "you're getting all bots, no humans, the site is failing". CitationDesk's framing flips that read: SourceScore is succeeding on the axis that matters for its archetype. The bot traffic isn't junk; it's the inventory for future revenue (Pay-Per-Crawl economics, when those mature) and the leading indicator that LLMs treat this content as a reference source.

Why the ratio is the signal

Reference + benchmark + ranking sites in narrow technical niches behave fundamentally differently from broad consumer-content sites. The audience is small and specific. The use-case is "answer a specific question authoritatively". LLMs are the new layer between the audience and the source — most readers will encounter the information via ChatGPT or Perplexity citing the site, not by visiting it directly.

For SourceScore, that's already happening. CitationDesk's weekly poll shows the site cited by ChatGPT on 7 of 17 archetype-aware queries, by Claude on 5, by Perplexity on 9. Each citation surfaces the site name + URL to a reader who would otherwise never have heard of it. The human visits to /llm-ratings/* are downstream of those citations — not the primary success metric.

What this means for sites in similar shape

If you operate a niche reference, benchmark, ranking, or comparison site that gets disproportionately more bot traffic than human traffic, you're probably looking at one of two states:

1. State A — LLMs are already citing you. The 480:1 ratio is the signal. Track which crawlers, on which pages, with what frequency. Use CitationDesk to verify the bot traffic correlates with actual citation events in the LLM responses. Lean into the archetype — add more structured reference data, more DefinedTerm schemas, more quote-ready first paragraphs.
2. State B — LLMs are crawling but not citing. Possible if the content is too template-driven (HCU thin-content territory) or if structural extractability is low. Run the Citation Readiness Score audit — most likely you'll see a 0.7+ on Bot-Crawl Health but a 0.3-0.5 on Dual Fit. Fix the first-paragraph extractability and the ratio of crawl-to-citation will improve fast.

The future-revenue layer

Cloudflare's Pay-Per-Crawl beta and TollBit's licensing marketplace point toward a future where AI crawler inventory is directly monetizable. SourceScore's 3,840 crawls/month is exactly the kind of inventory that lights up in that economy. CitationDesk doesn't monetize that for you — but it shows you which crawlers are showing up, when, and on which pages, so you can be ready when the marketplace settles.