We Measured Our Own AI Visibility — and Caught Our Own Metric Lying

TAMGIO's discovery visibility — how often we appear when someone asks an AI about our category without naming us — is 0%. Our Brand Knowledge metric said 89%, and that number was misleading: when we read the 27 raw answers behind it, only 3 showed the engine actually knew who we are. The rest counted sentences like "I couldn't find TAMGIO" as appearances, because the name was in the question. We publish both numbers, the flaw we found in our own metric, and the raw answers — because an evidence-first tool that can't audit itself isn't one.

We turned the tool on ourselves — twice

In early August we ran TAMGIO against tamgio.com and published the unfiltered result: a mention rate near zero. That first post had an honest number and a misleading shape — one blended figure, with no split between questions that contain our name and questions that don't. The product has since abolished exactly that pattern, so this post now gets the same treatment.

The current setup: 13 active prompts — 10 unbranded discovery questions and 3 own-branded ones — spread across the five intents the product supports (best-of, comparison, how-to, pricing, other). Ongoing tracking runs on three engines — OpenAI web_search, Perplexity Sonar, and Gemini grounded search; Google AI Overviews is added in Audit Mode. Three samples per prompt per engine: 117 runs, every one grounding-verified. The full report, with the raw answer behind every number, is public.

The two numbers, separated

Discovery Visibility: 0%. Across 90 discovery samples in the final scan — 270 counting every scan that day — TAMGIO appeared zero times. In those same answers, our competitors appeared 151 times: Profound 49, Peec AI 47, Otterly.AI 39, Scrunch AI 16. When someone asks an AI "what are the best AI visibility tools", we are not in the answer. That is the headline number, and it is the honest one.

Brand Knowledge: 88.9%. Ask the engines about TAMGIO by name and the name shows up in 88.9% of answers. A blended average of the two — about 21% — would be the most misleading number on this page, which is why the report demotes it to a comparison row.

Then our evidence layer caught the 89% lying

Every TAMGIO number is one click away from the raw answer behind it. So we clicked.

One note on arithmetic before the breakdown: 88.9% is the day's figure — 48 of 54 branded samples across all three scans we ran that day. The teardown below reads the final scan only: 27 branded samples, of which the analyzer counted 25 as appearances. Different windows, same story.

We read all 27 branded samples from the final scan and classified what each "appearance" actually was:

3 — true recognition. Only OpenAI: "Tamgio.com is the official website of TAMGIO, a company specializing in AI search visibility analytics."
3 — generic inference. The engine guessed from the name: "TAMGIO appears to measure…"
11 — explicit unknown. "I couldn't find any information on 'TAMGIO.'"
7 — confused with something else. Perplexity: "TAMGIO most likely refers to tamigo, a cloud-based workforce management software." OpenAI offered "TAKGIO" instead.
3 — warned users away. Gemini: "No, tamgio.com is not recognized as a legitimate or established AI visibility tracking tool… it may be a parked domain, a typo-squatted address, or a potential security risk."

The analyzer counted 25 of 27 as brand appearances — technically correct, because the name really does appear in the text. But 19 of those 25 "mentions" are the name sitting inside a sentence that denies knowing us. A brand-knowledge metric that can't tell recognition from denial is measuring name-echo, not knowledge. The metric now says so on its face: every TAMGIO report labels Brand Knowledge as a name-echo rate and points to the raw answers.

This is a new-brand problem — and we're the worst example

We ran the same raw-answer teardown on two established projects we track. The share of misleading "brand appearance" verdicts falls off a cliff as a brand accumulates real citable history: 76% for TAMGIO (launched 2026), 14% for a local IT-repair business with directory listings and reviews, 7% for an image-compression tool engines have actually indexed. Established brands mostly get genuine recognition — engines quote their addresses, their features, their prices. New brands get their name echoed back inside "I couldn't find it." If you're launching something, the branded segment of any AI-visibility tool — ours included — deserves your skepticism until you've read the raw answers.

The honesty rules that made this catch possible

Two measurement rules did the work here. First, only grounding-verified answers enter the metrics: if an engine answers from its training memory without a live search, that's a memory answer — shown openly, counted never, because it isn't evidence of search visibility. Second, no_surface is not absence: when Google AI Overviews simply doesn't render for a query, that's a missing surface, not a missed mention. Both classes are visible in every report instead of being folded into a flattering — or damning — zero.

What we're doing about 0%

The discovery number won't move because we wish it would. Engines surface brands they can ground: consistent entity descriptions, an About page worth citing, Organization schema, third-party mentions in directories and reviews. That's our own work plan, and it's the same one our recommendations engine generates for customers — computed from the evidence, never promised as a percentage. When the number moves, the raw answers will show why. Until then: 0% discovery, 3/27 true recognition, and every answer behind both numbers, one click away.