I Did the Whole GEO Checklist on My Own Site. My Analytics Say It Did Almost Nothing.

TL;DR: I ran the standard on-site GEO checklist on my own site: schema, clean crawler access, answer-first blocks, the works. Then I opened my analytics. Almost all my AI referral traffic traced back to a single Reddit thread, not to any of the on-site work. Reddit drove roughly 28 times the AI traffic that my optimized pages did. The on-site checklist just flipped a switch. The Reddit thread cranked the handle. Here is the data and what I changed because of it.

What did I do on-site?

I did the full checklist everyone sells. Clean robots.txt so the AI crawlers could read me, valid schema, an llms.txt, answer-first blocks under question headings, tightened metadata. My crawler config was already correct, so the engines could see me fine. In these terms, the switch was already on. If on-site GEO were the crank, my AI traffic should have climbed from the pages I optimized.

What did the analytics actually say?

It said the pages barely mattered and one Reddit thread did the work. Looking at my referral sources, a single thread that hit 259 upvotes drove roughly 3,600 sessions, and my AI-assistant referrals sat around 126 for the same window. When I broke down where the AI traffic actually came from, Reddit was doing about 28 times the volume of my carefully optimized pages. The on-site work I had just finished was not what the assistants were reading. They were reading the Reddit thread, and the sites that quoted it.

Why did the checklist do so little?

Because on-site GEO is a switch, not a crank. Crawler access and markup decide whether an engine can read you. They are binary and cheap, and once you clear them, more polishing does nothing measurable. This matches the controlled evidence: Ahrefs found adding schema moved AI citations −4.6% on AI Overviews and about zero elsewhere, and Google says markup and llms.txt are not required for its AI search. What decides whether you get recommended is off-site: the mentions, threads, and reviews the engine retrieves. Across tens of thousands of brands, branded web mentions correlate 0.66 to 0.71 with AI visibility versus about 0.22 for backlinks. My own data is a sample size of one, but it points the same way as the big studies.

Can you delegate the crank?

This is the uncomfortable part, and I am not going to dress it up. The Reddit thread worked because it was me, with my history in that community, posting something people cared about. I cannot hand that exact move to a client and expect it to land. The crank is real, but it is not a button. For a local business the crank is not Reddit at all; it is their real-world footprint: reviews, citations, YouTube, being mentioned in the local roundups. The mechanism that demonstrably works is the one that takes actual effort, and the on-site mechanism that is easy to automate is the one that does not change the outcome. Anyone who tells you otherwise is selling the easy half.

What I would do differently

I would spend far less time on the on-site checklist and far more on getting mentioned. Flip the switch once, confirm the crawlers can read you, and stop. Then put the hours into the crank: get into the category roundups, earn reviews, show up on YouTube, and give people a reason to mention you where the engines are reading. That is the work I now do first, because my own analytics embarrassed me into it.

FAQ

Is on-site GEO a scam? No. It is necessary hygiene. It is just oversold as the thing that gets you recommended, when the evidence and my own traffic say it is not.

Isn’t one Reddit thread just luck? Partly. But the direction, off-site presence beating on-site markup, is confirmed by large studies, not just my one thread.

What if I have no audience to post to? Then the crank is other people’s audiences: roundups, reviewers, communities, and video. You do not need your own following to get mentioned in theirs.

Sources and method

The traffic numbers are from my own GA4 on a site whose crawler config was already clean, so the switch was already on before the Reddit thread ran. The correlation and schema figures are from Ahrefs’ brand-visibility and schema studies, and Google’s 2026 AI guidance. First-party analytics are a sample of one and correlational, so read them as a directional confirmation of the bigger studies, not proof on their own. If you opened your own analytics right now and filtered to AI-assistant referrals, would the traffic trace back to your optimized pages, or to somewhere you got mentioned?