The first GEO piece on this site is a checklist and the second is a test log. Both are about doing. This one reads back what the log settled, because after a few months of re-asking the same three saved queries and watching the entries accumulate, the shape of my effort changed more than the tactics did.
The win lands in the wrong dashboard
AI referral traffic to this site is a rounding error: single visits on days when organic Google brings dozens. If I graded GEO by that referral column I would have quit in January. The movement shows up elsewhere. A citation in an answer engine shows up in my analytics a day later as a branded search, or a direct visit from someone who saw the site named and typed the domain. Both land in columns that read as brand, and neither credits the answer engine that caused them.
So I stopped grading GEO like a channel and started grading it like PR: track where the name appears, accept that the click gets attributed to something else, and stop expecting the referral report to justify the work. It will always under-count.
Ranking rewards a thousand words, quoting rewards one
A page earns a ranking with depth: coverage, structure, internal links. An answer engine quotes one sentence. Those are different contests, and the second one is winnable on purpose. The extraction test from the log made it concrete: paste the page into ChatGPT, ask what it says, watch which sentence leads.
Now every page here that wants to be cited carries one line near the top that survives being lifted out alone: subject, claim, number, in one sentence. The rest of the page argues for that line. When the liftable line is missing, the extractor promotes whatever sits on top, and “it depends on several factors” has never been quoted by anyone.
Two listicles beat my whole site
When I traced where the tools named in my tracked answers actually came from, seven of the eight traced back to two third-party listicles. My own pages, indexed and structured and current, supplied one. That reframed the effort. A quotable page on your own domain gets you into the game; the answers are being assembled from pages other people own. Getting named in the roundups, directories and comparison posts an engine already trusts moves an answer faster than another rewrite of your own page does. This is PR work in an SEO costume, and it keeps getting filed under the wrong job.
Freshness is a production problem
I made the freshness case in the GEO checklist post. What the log added: sixty pages is more than a person hand-refreshes past week three. Easy to nod at, brutal to operate. This site alone is roughly sixty pages, and a hand-refresh cadence across sixty pages is a part-time job that nobody keeps doing past week three. So freshness is a capacity problem. I solved it with a system. The content engines I run at work exist for exactly this reason. Agents draft and refresh on a rotation, a review gate keeps the judgment human, and the calendar stays full without anyone hand-cranking it.
Three things I stopped doing
I stopped chasing head terms with AI answers. The engines trigger on specific, question-shaped queries, and “CRM” was never going to cite anyone. I stopped tracking every model separately: three saved queries, re-asked monthly in two engines, tell me more than a dashboard of twelve would. And I stopped opening the AI-referral report weekly, for the reason above. The wins land in branded search, on a delay, wearing a different label.
The operating rules
Grade GEO like PR. Write the liftable sentence on purpose, one per page. Spend the outreach hours on other people's listicles. Solve freshness with a system. And keep a log, because most GEO advice arrives as a conclusion with the evidence missing, and the log is how you find out which conclusions survive contact with your own site.