I Ran My Own AI-Visibility Tool on My Own Site. It Scored 0%.
Corrections September 8 2026 — three, all found by our own checks before anyone outside saw a wrong number. We publish them rather than editing the post quietly.
Corrections, September 8 2026. Three, all found by our own checks before anyone outside saw a wrong number. We publish them here rather than editing the post quietly, because a measurement company that silently rewrites a published measurement has spent the thing it sells.
- The crawler window was wrong, and the corrected version reads better than what we published.This post originally said 265 crawler visits “in the thirty days to July 31”. The telemetry that produced them did not exist before July 22, so the real observation window was ten days, not thirty. That is a crawl rate roughly three times higher than the one we published. We got the window wrong. The count turned out to be wrong too — see correction 4.
- The same telemetry was later found to drop writes,so we called 265 a floor. Correction 4 supersedes this: the larger problem was not dropped writes but counted scanners.
- The 98% branded figure is unverifiable, and we are leaving it visible rather than deleting it.It was recorded before we retained transcripts, so the raw answers behind it no longer exist and it cannot be re-derived. In September we also fixed a scoring rule: an engine that runs a search, fails to find the site and says so had been counted as a mention. Under the corrected rule our September branded figure is 80% of 40 runs, not 82% — the correction lowered our own number. Every figure published from August 1 onward has its transcripts stored, and a publication gate now blocks any measured figure that cannot be recomputed from them at publish time.
- The crawler count was mostly not crawlers.We logged bots by the name they gave us. Scanners give false names, and 235 of the 271 July hits were scanners probing for credentials while calling themselves GPTBot. Counted against pages that exist, the honest figure is 36, not 265 — and we stored no IP addresses, so even 36 is unverified. The instrument is fixed; the raw log is kept. The one finding that survives: Claude’s search crawler has fetched a page here twice, ever. (Added September 23 2026.)
What the 0% means: category citation win on unbranded buyer questions in July 2026 — the engines did not yet recommend AEO Analyzers over competitors. It is not a product-quality score. Branded retrievability in the same measurement: 98%.
Updated 15 October 2026: Pro and Business sweeps now ask five engines, adding Grok. The method is otherwise unchanged.
I build AEO Analyzers — a tool that measures what ChatGPT, Claude, Perplexity, Gemini and Grok actually say when buyers ask about a business. In late July I pointed it at the one site I know best: my own.
The branded results looked great. Ask any of the four engines about AEO Analyzers by name and they find it almost every time — 98% branded retrievability across the stored runs. And I wasn’t happy. Do you know why?
Note added November 1st, 2026: the ledger restated this period under the stricter rule of September 2nd, 2026 (an answer counts as finding us only when it reaches our own site). See the ledger.
Because in business — and honestly, in life — what determines your outcomes isn’t just the answers you have. It’s the questions being asked. Right now, your buyers are asking AI engines the questions your business exists to answer — and the engines are answering them with or without you. The brands that win aren’t the ones with the best answers sitting unread on their websites; they’re the ones present at the moment the right question gets asked. That conviction is what AEO Analyzers is built around: it works out and asks the hard, specific questions that matter for your brand, your business, and your category — across the major answer engines, repeatedly — with every question and every answer on the record.
So I asked mine. The unbranded, buyer-style questions — “what tools measure AI search visibility?” — measured five times per query, across all four engines. My category citation win: 0%. Zero. High confidence. Not once, on any question that mattered commercially, did any engine recommend my product. They recommended my competitors, by name, while my site sat unmentioned.
Note added November 1st, 2026: the ledger restated this period under the stricter rule of September 2nd, 2026 (an answer counts as finding us only when it reaches our own site). See the ledger.
The engines had already read everything I wrote
My crawler telemetry showed the strangest finding of all: in the ten days to July 31, requests carrying AI companies’ crawler names had fetched a page from my site 36 times (we first published 265; most of those were scanners — see the corrections above). Their crawlers had repeatedly retrieved material across the site. They just never reached for it when it counted.
When does it count? At the moment of the answer — the instant a buyer asks and an engine assembles its recommendation from the sources it trusts. Being read is not being cited. The crawlers had my prose; what they hadn’t found was a first-party record shaped like the questions buyers actually ask — current, structured, extractable, and unmistakably about this product rather than the stock tickers and browser extensions that share its letters.
And there’s a deeper reason, the one that convinced me this category matters: you cannot search for what you don’t know exists. Radio waves filled the sky for all of human history before anyone thought to ask about them. A buyer who doesn’t know AI-visibility measurement exists will never type it into a search box — but they will ask an engine “how do I show up in ChatGPT’s answers?” The engines get asked the question before the category has a name. Either you’re in that answer, or you’re invisible — not because you’re bad, but because nobody knew to look.
What the zero taught me — and what I built because of it
I could have quietly fixed things and announced a success story later. Instead I’m publishing the zero, because the zero is the honest starting line — and because everything I learned fixing my instrument became the product itself.
- Measurement integrityI found my own tool truncating one engine’s answers and silently counting them as losses, and counting answers where the engine never searched the web at all. Now truncated runs are excluded and flagged, search-grounded and memory-only answers are scored separately, and every number ships with its sample size and confidence printed next to it.
- FidelityOne engine had, at some point, invented a co-founder for my company — a person who does not exist. Others confused my product with unrelated things that share its letters: stock tickers, browser extensions, similarly named tools. That embarrassment became a detector — every branded answer is now checked against the site’s own record and classified as accurate, drifted, or confused with another entity, with the colliding entities named.
- RecommendationsPlenty of AI-visibility advice ends the same way: become more authoritative, get referenced on Wikipedia, and the engines will trust you. For most businesses it is a goose chase. I have watched real companies spend real budget paying intermediaries to engineer a Wikipedia presence — months and money spent tilting at windmills that were never theirs to reach. So my tool refuses to assign that homework. Every suggested source is tiered by attainability: what you can do yourself this week, what is genuinely pitchable, and what is aspirational — labelled as aspirational, not disguised as a to-do.
Your time is the scarcest input in your business; an action plan should spend it where movement is actually possible.
Where the number stands
My site’s own fixes are live too: corrected structure, an entity graph that separates me from the tickers, a machine-readable record of the facts. Did it work? As of today’s stored baseline: not yet. The number is still 0%. That is what the transcripts say, so that is what I’m telling you.
Here’s the commitment: I re-measure monthly, the same questions, the same way, and I will publish the number each time — whatever it is. If the fixes work, you’ll watch it move. If they don’t, you’ll watch me figure out why, in public. Every claim I ever make about this product will be backed by a transcript you can reproduce yourself.
If you want to know what the engines say about your business — before a prospect finds out for you — the check is free and takes about ninety seconds: aeoanalyzers.com. Whatever your number is, at least you’ll know it.
Mine’s zero. For now.
Where the number stands now
This post is the first zero. The measurement did not stop here — it runs every month, the same twelve questions, the same five runs each, across the same four engines. Here is the most recent one.
November 2026, measured 1 Nov 2026. Found by name: 78% in November, 88% in October, 80% in September, 55% in August. Recommended to buyers did not move at all. Those are two different problems, so they are never blended into one score. The word “aeoanalyzers” appears in none of the 200 category answers — absent, not ranked last. On 13 October we re-measured on the same four engines and again with a fifth added; the figures, one stored answer in full and what they mean are in We built an AI visibility tool. Then we tested ourselves.
Every month, whatever it says
Every month so far, each with its own note on what changed and what we learned, lives in one place.
See the full ledger →Updated October 2026: Pro and Business ask five answer engines, now including Grok. See We added a fifth AI engine.
AEO Analyzers (aeoanalyzers.com) was created and is solely maintained by Lindsay Hiebert, founder of PIGENAI LLC. It is unaffiliated with any similarly named browser extension, plugin, or tool.