Meta Reads More Than Everyone Else Combined. None of It Was a Question.

September 18, 2026

One company reads this site more than every other AI company put together. In thirty days, not one of those requests came from a person asking a question.

I have been measuring a small WordPress site for a few months now, recording only AI crawlers and sorting each request by what the crawler is for. Over the thirty days ending September 11, it was visited 11,818 times, and 8,979 of those landed on pages with writing on them.

Here is who came.

The first column is the one people quote. The second column is the one that decides whether any of it can ever reach you.

Three crawlers, three different jobs

Meta does not run one crawler. It runs several, and it says plainly what each is for. Three of them are AI crawlers, and this instrument watches all three.

• meta-externalagent — Meta's documentation says it “crawls the web for use cases such as training foundation AI models or improving products by indexing content directly.”
• meta-webindexer — it “navigates the web to improve Meta AI search result quality for users,” which is the crawler whose job includes getting citations right.
• meta-externalfetcher — it fetches individual links at a user's request. This is the one that means somebody asked something a few seconds ago.

Over thirty days, the split between them was not close.

Thirty-nine requests collecting material for a model, for every one request improving the thing that could cite you. And the third row is not small. It is empty.

The path back exists. It is barely used.

This is what makes the ratio worth looking at rather than just complaining about. Meta is not a company without a route from its answers to your site. It documents one. meta-webindexer exists precisely so that Meta AI can search well and attribute accurately.

On this site, that crawler did 155 requests while the training collector did 6,027. The machinery that might send a reader back is running at about two percent of the machinery that takes material away.

Compare the shape of that with the two vendors at the other end. Perplexity's traffic here is 93% user-initiated: almost everything it does on this site is a person, right now, waiting for an answer. OpenAI is 21%. Meta is zero.

That is not a moral ranking. A training crawler is not doing anything wrong by collecting training data; that is its job, and Meta says so in the open. But the mix tells you what a given company currently wants from your writing, and the mixes are not remotely alike.

An awkward footnote: I cannot prove any of this was Meta

Everything above rests on 6,182 requests that said they were Meta. Meta publishes no IP ranges and no reverse-DNS scheme for these crawlers, so there is no way to confirm that any single one of them came from Meta. My dashboard marks the entire vendor with a dash where other vendors have a percentage.

So the largest reader of this site is also the one I can least verify. I wrote about that problem a week ago and it has not improved: most requests on this site still arrive with no proof of who sent them, and none of Meta's ever can.

I do not think these requests are forged. The volume is steady, the behaviour is consistent, and there is no obvious reason to impersonate a crawler that most site owners have never heard of. But I am reporting a number I am not able to check, and that should be said out loud rather than buried.

Which brings us back to the two

In the first piece I wrote, this site had been read about ten thousand times in a month and had received two human arrivals, one from ChatGPT and one from DuckDuckGo.

I did not know it at the time, but the single largest contributor to that reading number was the one vendor structurally least likely to appear in the arrival number. The gap between “read” and “visited” was not evenly caused. It had a majority shareholder.

What this does not mean

• It does not mean Meta never fetches on request. It means that on this site, in these thirty days, its fetcher did not appear. One site is not a census.
• It does not mean Meta never cites anyone. Meta runs a crawler whose documented job is citation quality. It ran here 155 times. What I can say is the ratio, not the policy behind it.
• It does not mean training collection is theft. That argument exists and it is not this article. This article is about what the numbers say a company is doing, which is a smaller and more checkable claim.
• 52.3% is of the crawlers this instrument tracks. Meta also runs crawlers for ads and for link previews. Those are not AI crawlers and are not counted here.

What this instrument cannot see

• Cached pages are invisible. When a page cache answers, PHP never runs and nothing is recorded. Every count here is a lower bound.
• One site, one month, one language. This is a Japanese site about a city in the Philippines. A different subject, a different language, a different publishing rhythm would draw a different table.
• Arrivals are undercounted by design. AI clients running as phone or desktop apps send no referrer, so a reader who does follow a link can land in your analytics as direct traffic. The arrival number is a floor.

How to measure your own site

The instrument is a WordPress plugin called AILYS Lens, and it is free. I build it, which you should factor in. Every calculation runs on your own server, nothing is transmitted anywhere, and human visitors are never recorded. The vendor table described here is the main dashboard, and the split by purpose — training, answer indexing, live fetch — is the column worth looking at second.

You may find the same shape. You may find the opposite. The point is that this is a question with a factual answer about your own site, and it takes a week to get.

Illustration generated with AI and selected by the author.