Being Read, But Not Recommended
There are two ways for an AI to not send you customers, and on a dashboard they look exactly the same.
In the first, the machines are reading your site constantly and never mentioning you when someone asks. In the second, they aren't reading you at all any more, but they're still quoting a copy from months ago. Both show up as a flat line. They call for opposite responses.
Telling them apart costs nothing.
Two instruments, and neither one is enough alone
Your website analytics, the kind that runs on a snippet of JavaScript, will never see a crawler. Crawlers don't run JavaScript. As far as your dashboard is concerned, they were never there.
Your server logs are the opposite. They record every request, including every crawler that announced itself, and they never see a click. A person reading three pages and a robot fetching three pages look similar in a log.
Neither half is the story. Put them side by side and you get four states instead of one number:
- Crawler visits, no referrals. You're being read and not recommended.
- Referrals, no recent crawling. You're being quoted from a stale copy.
- Both. The thing is working.
- Neither. You aren't in the conversation.
Most businesses already have both instruments. Almost nobody has put them next to each other. That's the whole technique, and it's free.
What happened when I ran it on myself
I have a fortnight of both, and here's the honest reading.
On the referral side: three referrers in two weeks. One from Google, one from Facebook, one from YouTube. No AI assistants at all — nothing from ChatGPT, nothing from Claude, nothing from Perplexity.
On the crawler side: about 60% of everything that touched my server was a declared crawler, and the biggest by a distance was Ahrefs, which is an SEO tool. Then Googlebot, then two more Google services.
So the first pass says: read heavily, referred not at all.
Except my instrument couldn't finish the sentence
Here's where it got interesting, and this is the actual reason I'm writing this one up.
To claim "AI systems are reading me and not citing me," I need to know that the AI crawlers specifically were among those visits. The ones with names: GPTBot, ClaudeBot, PerplexityBot, and the rest.
My summary named the top two crawlers and grouped everything else into a total. Those two account for about 700 of roughly 1,830 crawler visits. The other 1,100 are real, and I can't tell you what they were from the extract I pulled.
So the honest version is narrower than the headline. I know I'm getting no AI referrals. I know most of my server traffic is robots. I do not yet know whether the AI crawlers are among them, and until I do, "being read but not recommended" is a hypothesis about my own site rather than a finding.
That's a defect in how I pulled the data, not in the method. It's fixable in about ten minutes next time. But I'd rather tell you the version where the check came back incomplete, because you're going to run into the same thing the first time you try this.
Why the pairing still earns its keep
If I'd only looked at referrals I'd have concluded nothing is happening. If I'd only looked at server traffic I'd have concluded plenty is happening. The two together say something more specific and more useful: whatever is arriving isn't yet turning into a person, and the next question is which robots those are.
One number couldn't have told me that. Two cheap numbers did, and they were both already sitting there.
The version I'd suggest if you want to try it: pull a month of server log summary, note whether any crawler with an AI company's name in it appears, then check your analytics referrers for the assistants by name. Write down both. You don't need a tool and you don't need to buy anything.
Then do it again in a month, because one reading of anything is a snapshot, not a direction.
If you'd like a hand setting this up on your own site, text me at 503-664-0546 or email eog@ernestofgaia.xyz.
#ernestGoesToAI