We hit an unexplained ceiling on our own scoring tool. Here's the data.
Quick build-log update from a small, very specific experiment.
We make foglift.io, a scoring tool for AI search visibility. Part of our content strategy is shipping 8 listicles that AI engines might cite when someone asks "what are the best tools for X". 5 of 8 are live so far.
Each listicle uses an identical template:
- 10-vendor comparison
- 4-column table (tool, best for, free tier, starting price)
- Article + ItemList + FAQPage JSON-LD
- 7 FAQ entries
- Executable bash benchmark script
- 6-8 academic + industry citations
We ran our own scorer against each one as it shipped. 8 dimensions, 0-100 scale. Same crawler, same grading rules.
Results so far:
- Listicle #1 (developers): 86
- Listicle #2 (MCP integration): 86
- Listicle #3 (open source): 88
- Listicle #4 (AEO/GEO platforms 2026): 90
- Listicle #5 (ecommerce): 88
Listicles #1 and #2 had a 3-column table and no code block — that explains the gap to 88.
But here's the part I can't explain. Listicles #3 and #5 have the exact same upgraded structure as #4. Same column count. Same JSON-LD schemas. Same FAQ count. Same heading structure. Within a few hundred words of each other.
#4 is the only one that hit 90. It's also the only one whose topic literally matches the keyword the platform optimizes for ("AEO/GEO platforms 2026"). Which leads to a hypothesis I haven't tested yet: the scorer might be giving an extra Topical Authority bump when the page topic and the platform identity overlap exactly.
Re-scanning all 5 in a week to see if 90 holds or drifts back to 88. If it holds, the topic itself is the variable. If it drops, 90 was a transient crawler artifact.
Two questions for anyone who's built a content scorer or audited their own:
- How do you separate "stable score" from "transient crawler artifact" when re-scan windows are short?
- Has anyone seen topic-keyword overlap measurably shift score on their own scoring tool? Curious whether this is real or coincidence.
Will report back next week with the re-scan numbers.
I know a few people who build content scoring tools and audit their own content performance, they'd probably be willing to answer your questions if you want. It's tough to separate stable scores from transient artifacts. I'm happy to pass your questions along.
You probably didn’t hit a scoring ceiling.
You hit a naming ceiling.
When the page topic, platform identity, and query intent align exactly, the model has less ambiguity to resolve — so confidence goes up.
That usually gets misread as a content lift.
It’s often an identity lift.
“AEO/GEO platforms 2026” scored higher because the page was easier to classify, easier to trust, and easier to map to the query without interpretation overhead.
That’s the hidden tax in names like foglift.
The product may be sharp.
The name still makes the model do extra work deciding what you are.