AI Citations Hit 265: What I Actually Discovered in the Data

Last week I shared that my AI citations had jumped from 18 to 163. I checked again this week and the number is now 265. The climb is continuing.

When I posted the original number, some of the replies raised a fair question: what does that figure actually represent, and can you trust a single dashboard to tell you? That's the kind of question worth taking seriously rather than waving away, so instead of defending the number I went looking for how it's actually produced, and whether I could see the same activity from a completely different vantage point. The result is a much more detailed picture than any one dashboard gives, and it's the most useful thing I've learned in weeks.

This issue is that investigation. Three data sources, each one blind in its own way, and what they show when you stack them.

Source one: what the citation dashboard is actually counting

The 265 comes from Bing Webmaster Tools, in a section called AI Performance. Before trusting it, I read what it says about itself, and three things stood out.

It tells you plainly that the data is a sample of overall activity, refined as more data comes in. So it's an estimate, not a precise meter.

It only covers one ecosystem. The sources are listed as Microsoft Copilot and partners, which means citations from ChatGPT, Perplexity, Claude, and Gemini are entirely invisible here. When I say 265, the honest version is 265 sampled citations from one provider's world, almost certainly an undercount of the real total, and silent on every other assistant people use.

And it records the question behind each citation. That tells you something real about the mechanism: the assistant searched, pulled my page, read it, and grounded its answer on it, live, at the moment someone asked. That's retrieval, not the model reciting something from training. It's the part you can actually influence by how you build a page.

The distribution also changed in a way that matters more than the total. Two weeks ago, one page carried almost everything: 91 of 163 citations from a single "best newsletters" page. Now that page sits at 103, but it's no longer most of the total. Citations have spread across sixteen pages and a wide range of questions: structured side projects for professionals, side projects for experienced consultants, remote project ideas for engineers leaving full-time roles, best developer marketing newsletters. That's a sturdier signal than the headline number, because a count spread across sixteen pages doesn't collapse if one ranking shifts.

So source one is real, but partial: a true signal, sampled, from one corner of the market.

Source two: the crawler logs, which see what the dashboard can't

A citation can only happen if the engine fetched the page first. The fetch is the prerequisite. So to see the engines Bing can't report on, I looked at who is actually crawling the site at the CDN layer, where every bot identifies itself.

Over the last period (24h increments), Cloudflare logged 185 AI crawler requests, up about 15%, with 168 returning a clean success response and none failing. The crawlers can reach the pages, which is the necessary first step and the thing I'd quietly broken earlier this year without realizing.

The breakdown is where it gets interesting, because it shows the assistants that have no citation dashboard at all:

OpenAI's retrieval crawler hit the site 37 times, up 270% on the prior period. Separately, the ChatGPT-User agent, which fetches a page because a human in ChatGPT asked something that pulled it in live, appeared 16 times. Anthropic's ClaudeBot came 15 times. Perplexity, 4, up 100%. Apple's crawler, 26. Google's, 42. And ByteDance's Bytespider led the raw count at 45, though that one is mostly training collection rather than live retrieval, so I weight it lightly.

Here's the number that reframes everything. Those 265 measured citations trace back to Bing's crawler, which hit the site just 9 times in this window. Meanwhile OpenAI's retrieval bot hit it 37 times and a human-triggered ChatGPT fetch happened 16 times, with no dashboard anywhere to count what came of it. The citations I can measure are the small, visible corner of a much larger surface I can only infer from the supply side.

The crawl doesn't prove a citation. But a retrieval crawler fetching your pages dozens of times is strong evidence you're being read and used by engines that will never send you a report, and it widens the picture far beyond the one ecosystem the dashboard covers.

Source three: did a human actually arrive

Citations and crawls are both upstream of the only thing that matters in the end, which is a person landing on the page. So the third source is my own analytics: referrals from the assistant domains.

I'll be straight about this one. The citation count is large, the crawler activity across engines is real and growing, and the human referral traffic from AI assistants is still modest. Most of the human activity I gained over the past two weeks came from organic sources (social posts, conversations, and the like), not from citations directly. The citation number and the subscriber number are two different stories, and stapling them together would be the exact kind of metric theater this whole exercise is meant to avoid.

What the three say together

Each source is blind in a different way. The citation dashboard sees usage, but only inside one ecosystem and only as a sample. The CDN logs see access across every engine, but can't tell you whether a crawl became a citation. Analytics sees real humans, but not what happened upstream to send them.

Stacked, they tell a coherent story. Retrieval crawlers from multiple engines are fetching the pages successfully, so the content is reachable and clearly of interest to them. Inside the one ecosystem I can measure, that interest is converting into citations at a rising and broadening rate. And the downstream human traffic, while real, is still small enough that I'd call AI citations a promising leading indicator rather than a traffic channel to bank on yet.

That's a less tidy story than "265 and climbing." It's also the true one, and the true one is more useful, because it tells me exactly where to look next instead of just feeling good about a number.

The One Thing

What I'm Thinking About This Week

The gap between a crawl and a citation, and between a citation and a click. I can now see OpenAI and Anthropic and Perplexity fetching the pages, and I can see Bing citing them, but I can't yet draw a clean line from a fetch to a citation to a visit. So the next experiment is to instrument that chain: watch which crawlers hit which pages, cross-reference against which pages get cited, and tag the referral paths to see what actually arrives and does something once it lands. If the chain holds, it tells me where to spend effort. If it breaks at the click, citations are a flattering number and I'll treat them that way.

Reply to this email with one of:

• Whether you've ever checked your own CDN or server logs for AI crawler traffic, and which bots you found

• Which of the three sources you'd trust most, and why

• A metric in your own work you suspect is softer than you treat it

P.S. The playbook that kicked all this off is here: How to Get Cited by AI. The rest of the toolkit, in the order I'd use it: Visibility Audit → SEO Strategy Playbook → Title & Meta Audit → Sovereign Idea Workflow → Business Model Canvas. All free.

P.P.S. Know someone who takes a single dashboard number at face value? This is the issue for them. Forward it; they can subscribe at thesovereigntechnologist.com.

That’s all for this week.

See you next Thursday.