Field notes · SEO & Content · Apr 14, 2026 · 5 min

Two listicles beat my whole site: notes from a GEO log

Running a GEO log for months taught me which of my own conclusions were wrong. Here is what the tests settled: where the win shows up in analytics, and why quoting and ranking reward opposite things.

7 of 8tools in answers traced to two listicles
1Perplexity citation earned so far

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, I changed where I spent the hours more than I changed the tactics.

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. It 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. I track where the name appears and let something else take credit for the click. The referral report will always under-count.

Ranking rewards a thousand words, quoting rewards one

A page ranks by going deep on coverage, structure and internal links. An answer engine quotes one sentence. Those are different contests, and the second one is winnable on purpose. The log made it concrete. I paste the page into ChatGPT, ask what it says, and 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.

Where credit landsgraded like PRAI referral columnsingle visits, a rounding errorBranded searcha day later, reads as brandDirect visitsomeone typed the domainthe operating rulesGrade GEO like PRThree saved queries, re-asked monthly in two enginesLIFTOne liftable linesubject, claim, number, one lineSOURCEOther people's lists7 of 8 tools traced to twoFRESHA production systemagents draft, review gate holdsSCOPEThree saved queriesre-asked monthly, two enginesLOGKeep the logwhich conclusions surviveThree things I stoppedChasing head termsTracking every modelOpening the referral report
A citation pays out in columns that read as brand, on a delay, so I work from the rules on the right. Seven of the eight tools named in tracked answers traced back to two listicles somebody else owns.

Two listicles beat my whole site

When I traced where the tools named in my tracked answers came from, seven of the eight traced back to two listicles. My own pages, indexed and structured and current, supplied one. That moved my hours. I started out wrong: I thought rewriting my own pages was the work. A quotable page on your own domain gets you into the game, and engines assemble the answers from pages other people own. Getting named in the roundups, directories and comparison posts an engine already trusts should move an answer faster than rewriting my own page again. That is the one rule the log has not tested, because I have not pitched a list yet and that is the next step.

Freshness is a production problem

I argued for freshness in the GEO checklist post. What the log added: this site alone is roughly sixty pages, and refreshing sixty pages by hand is a part-time job nobody keeps past week three. So I built a system. Agents draft and refresh on a rotation, and I approve every draft. The content engines I run at ThinkingAI are that system at work.

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 ChatGPT and Perplexity, tell me more than a dashboard of twelve would. And I stopped opening the AI-referral report weekly, for the reason above.

The operating rules

The rules above came out of this log, which holds three saved queries, re-asked monthly, and one Perplexity citation that turned up on a monthly re-ask. Most GEO advice hands you the conclusion and skips the evidence, and the log is how I find out which conclusions survive contact with my own site.

Suggested posts