This business thought it was fine. The audit said otherwise.
A regional brand with real search equity, competing against national names. The owner had no idea anything was wrong, because nothing they could see looked wrong. Anonymized, but every number is real. Ask yourself on each finding: would I even know?
Strong on Google. Nearly invisible to AI.
Across 222 engine responses, the brand was named unprompted in just 9% of its own category searches, and 0% of the time on the two largest consumer AI apps. When it was named, the facts were 97% accurate. Accuracy was never the problem. Presence was. Their buyers were getting good answers that simply didn't include them.
What the audit uncovered, finding by finding
Their #1 Google ranking bought them nothing with AI
Top of organic search for several category terms. Unnamed by the major AI engines on those exact same questions. Years of SEO investment, zero transfer to the answer their buyers now read first.
top organic rank · 0 AI mentions on the same queryTheir AI visibility was one lost citation from zero
We ran every web-capable engine twice: web on, then web off. With the web off, accurate mentions collapsed toward nothing. Their entire accurate presence was rented. Nobody knew.
web-off mention rate fell to ~3%Recognized when asked. Never volunteered.
Ask AI about the brand directly: solid answers, 89% of the time. Ask an open category question, the kind real buyers actually ask: the brand surfaced just 9% of the time. Recognition without recommendation is losing politely.
89% aided · 9% unaidedOffline, AI described a company that no longer exists
With the web disabled, the model remembered a years-old version of the business. It recalled ~40% of the current portfolio, missed every flagship offering, and invented services the brand never sold. Confidently.
40% of offerings recalled · multiple fabricated linesA bank was answering for their name
Asked the bare brand name, the model talked about a bigger, same-named company in a different industry. Buyers asking about them were hearing about someone else entirely. Nobody had ever checked.
bare name → a same-named bank, not the brandEvery engine told a different story
One engine never named them at all. Another named them a third of the time. If they had spot-checked only ChatGPT, they would have walked away with the wrong conclusion in either direction.
per-engine unaided recall: 0% → 31%The audit turns all of this into a punch list
Every brand fact gets scored into one of five states by comparing what AI memorized against what it retrieves live. This is how "are we visible in AI?" stops being a vague worry and becomes a to-do list with priorities.
| State | What it means for you | The move | This business |
|---|---|---|---|
| Baked-in | Durable equity AI knows cold | Defend it | 61.5% |
| Retrieval-dependent | Visible today, gone if a citation drops | Harden it | 7.7% |
| Confusion-corrected | AI gets it wrong until the web corrects it | Reinforce it | 0.0% |
| Persistent gap | Invisible everywhere | Build it. Highest leverage. | 7.7% |
| Regression | AI knew it once and lost it | Recover it | 23.1% |
Nearly a quarter of their known facts had regressed: the AI knew them once, and live answers had dropped or garbled them. They weren't just failing to gain ground. They were losing it, month by month, silently.
The worst part: their marketing budget was aimed at the wrong target
Their hard-won search rankings sat in topics with almost no AI-search demand. The high-demand topics, the ones buyers actually ask AI about, were exactly where they were invisible. Share of voice in competitive category answers: 18%. National competitors owned the rest.
Read that again as a business owner: they were spending money getting better at the game their buyers were leaving. Nobody inside the company could see it. It took one audit to make it visible.
What AI "remembered" about them should worry every owner
It recalled a real event, backwards
The model's most recent "fact" was an expansion it reported as a closure. Right that something happened, wrong on direction and place. Imagine a buyer hearing your growth story as a shutdown story.
It didn't say "I don't know." It made things up.
On half-known facts, the model fabricated. Several "does this brand offer X?" probes came back "yes, with high confidence" for services that have never existed. Your customers can't tell the difference.
The danger zone is exactly where your brand lives
On true blanks the model admitted ignorance. The fiction clustered around partially-remembered facts. A recognizable-but-not-famous brand sits precisely in that zone. That's most businesses.
It wasn't one engine misbehaving
Multiple independent model families shared the same stale picture. Fixing this takes deliberate, broad authority-building. It starts with knowing exactly what's wrong, which is what the audit is for.
Four engines, four different verdicts on the same brand
| Engine | Responses | Named the brand unprompted | Cited sources | Tone |
|---|---|---|---|---|
| Engine A | 74 | 2% | 11% | neutral |
| Engine B | 74 | 10% | 35% | neutral |
| Engine C | 37 | 0% | 0% | positive |
| Engine D | 37 | 31% | 43% | positive |
The spread from 0% to 31% is the point: there is no single "AI" to check. Your buyers are spread across all of them.
Your move
Your numbers already exist. You just haven't seen them.
Every AI engine already has a verdict on your brand. Buyers are already hearing it. The only question is whether you find out before or after it costs you. The audit finds out in days, not quarters.