The Engines Read My Site 265 Times. Reading Isn't Citing.

In Part 1 I published my own zero — known by name 98%, recommended to buyers 0%. One line stopped more readers than anything else in the post: the engines had crawled my site 265 times in a month and cited it none. This is the post about that line — why being read is not being cited, and why more content was never going to fix it.

What 0% means: category citation win on unbranded buyer questions (July 2026) — the engines did not yet recommend AEO Analyzers over competitors. It is not a product-quality score; branded retrievability in the same measurement was 98%.

In the thirty days to July 31, the crawlers run by the AI companies visited my site 265 times. They didn't skim. They took the whole thing — every page, every paragraph, every FAQ I'd ever written. And in that same window, across the buyer-style questions that matter commercially, they recommended my product exactly zero times.

Here is the part that reframed the whole problem for me. Those 265 visits were not 265 chances to be recommended. The large majority were training-tier crawls — bots ingesting the web to build the model's general memory, months before any specific buyer ever types a question. GPTBot alone accounted for 179 of them. Only about one visit in sixteen was a live, answer-time fetch — an engine reaching out to the web in the moment it was assembling an answer for a real person.

So the honest picture wasn't "the engines read me 265 times and snubbed me." It was quieter, and worse: they had absorbed everything I wrote into their general memory, and when the decisive moment came — a buyer asking a real question — my site was almost never the thing they reached for.

We are used to a web where being crawled leads to being ranked leads to being found. That chain is broken in the answer era, and the break is the whole story.

Reading is passive. A crawler takes your bytes; it costs the engine nothing and tells you nothing about whether you'll ever surface. Citing is active. It happens at the moment of the answer — the instant an engine, mid-response to a real buyer, decides which sources to name. Those are two different events, driven by two different levers, and almost everyone optimizes the first while believing they're improving the second.

I had spent years improving the first. More pages, more prose, more depth — all of it dutifully read, none of it cited. Volume was never the constraint. I could have tripled my word count and moved the number by nothing, because the engines already held more of my words than they could use. What they didn't have was a reason to reach for me when it counted.

When I looked at what the cited competitors had that I didn't, it wasn't more content. It was a first-party record shaped like the questions buyers actually ask — current, structured, extractable, and unmistakably about this one product rather than the stock tickers and browser extensions that share its letters.

That's when the three layers separated cleanly — and it's the model the whole tool is built on now:

Being read gets you to layer one and stops there. Everyone who tells you to "just publish more content" is selling you more layer one — and layer one was never my problem.

So I stopped writing more and started building the record the engines were missing: a corrected entity graph that separates my product from the tickers and extensions, a machine-readable statement of the facts, and an llms.txt that hands the engines the plain answers to the questions buyers ask — all structured so a passage can be lifted cleanly at answer-time.

Did it work? As of the stored baseline behind this post — the same July-31 measurement, same questions, same way — not yet. The citation-win number is still 0%. That's what the transcripts say, so that's what I'm telling you. The fixes are live; the proof isn't in.

And that's the deal I made in Part 1 and am keeping here: I re-measure monthly, publish whatever the number is, and back every claim with a transcript you can reproduce. If being read finally turns into being cited, you'll watch the number move. If it doesn't, you'll watch me find out why, in public.

If you've poured work into content and can't understand why the engines still don't name you, the odds are you're winning layer one and losing layer three — read, not cited. The good news: those are different problems with different fixes, and you can see which one is yours. The check is free and takes about ninety seconds: aeoanalyzers.com.

Being read felt like progress. Being cited is the only thing that pays.