My AI visibility went from zero to 26% in six weeks. The one category still at zero is the one this post is about.
Six weeks after publishing a zero-appearance rate across every AI platform, most categories moved. The one still at zero is AI Search Visibility itself.
Ben Rees - 13 August 2026
Six weeks ago I published the data: across ChatGPT, Gemini, Copilot, and Perplexity, over ten weeks of tracking, my appearance rate for non-branded queries was zero. Not low. Zero.
That post is no longer accurate, and the way it stopped being accurate is more useful than the original zero.
What actually changed
Two of the four platforms I was tracking in June are no longer part of the system at all. Copilot and Perplexity both started blocking automated queries with Cloudflare protection in late April, and rather than fight an arms race against a CDN, I suspended both. Gemini still works, but only with a human present to click through it, so it's no longer part of the overnight run. What's left running automatically, every night, is Google's regular search and its AI Mode summaries.
That sounds like a smaller system. It measures more accurately, because I stopped conflating "the platform stopped answering" with "I stopped appearing."
The gap left by dropping three platforms got filled by something new: as of the 9th of August, I'm also running raw API calls to Claude and ChatGPT with no browsing and no retrieval tools enabled, which isolates what's actually baked into each model's training data from what live search surfaces. The first reading: parametric memory citing me 6% of the time, live retrieval citing me 20% of the time. Petroni et al. (2019) is the reason that gap exists: what a model "knows" is frozen at training time, and my content is recent enough that most of it hasn't been absorbed into that frozen layer yet, even where retrieval already picks it up.
The number that moved
My most recent capture run, from the 11th of August: 25.7% overall appearance rate. Branded search, unsurprisingly, is at 100%. Scaling B2B Marketing content is at 57%. Marketing Team Leadership, a category that looked stuck at 3% as recently as a month ago, is now running 20 to 40% per capture, once I checked the underlying pages and found the content already existed, it just wasn't titled to match how people actually ask the question.
None of that happened because the platforms got friendlier. It happened because retitling and interlinking specific pages, page by page, moved specific queries.
The category that didn't move
AI Search Visibility, the category that covers "how do you measure whether AI cites you," is still at zero. Not diluted-by-old-data zero. Zero in the 9th of August reading. Zero in the 11th of August reading. The specific query "AI visibility measurement framework for B2B marketing" has never once returned bjrees.com, across every platform, in every run since tracking began. Google Search Console shows why organic interest isn't the problem: "geo brand tracking" gets 36 impressions and sits at position 24.8. People are searching adjacent to this exact topic. They're just not finding me when they do.
That's not a failure of the measurement system. It's the same lesson the system taught in June, just narrower and more specific: zero, once you've ruled out a broken query set or a suspended platform, is real information. The AI Search Visibility category is the one closest to what I actually do for a living, and it's the one still returning nothing.
What this means for how you measure your own visibility
Build for platform failure, not platform stability. Two of my four channels stopped working within weeks of launch, for reasons that had nothing to do with content quality. A measurement system that assumes every platform stays queryable will silently mislead you the moment one of them doesn't.
Separate parametric from retrieval, and expect them to disagree. A model's training-time prior and what live search surfaces are two different signals moving on two different timescales. Conflating them, especially early, produces false confidence or false alarm depending on which one happens to be moving.
Recheck zero before you accept it. The Marketing Team Leadership fix wasn't new content, it was retitling existing content to match real query phrasing. Before assuming a category needs a new asset, check whether the content already exists and just isn't discoverable under the words people actually use.
The measurement system is more honest now than it was in June, not because the numbers look better, but because I know exactly which parts of it I can trust.
Related reading
I am invisible on every AI platform. Here is the data.
For ten weeks I ran an AI visibility measurement system across ChatGPT, Gemini, Copilot, and Perplexity. My appearance rate for non-branded queries is zero. Here is what the data actually shows.
Making decisions in a Bayesian world
Most marketing decisions can't be A/B tested. Bayesian logic, combining prior knowledge with whatever data you do have, fills the gap.
A Weighted Sum of Everything Ever Written About You
What Bayesian linear regression, copied by hand from a textbook I've owned since January 2007, tells us about getting cited by AI.
Your brand doesn't get its own space inside an AI model
Language models pack far more concepts than they have space for, a phenomenon called superposition. Generic content competes for that same overcrowded space as everyone else's, and usually loses.