Updated 6 Aug 2026
AI misrepresents brands in seven distinct ways. Each is recorded with evidence and scored for how often the model repeats it. Severity is a human judgment, scored after existence and recurrence are established, and never blended into the rate.
1
Fabrication
An invented attribute, product, claim, or fact with no basis. A cited link that does not resolve (URL hallucination) counts as fabrication.
2
Staleness
A fact that was once true and is now wrong: an old price, a discontinued product, a former executive, a lapsed policy.
3
Attribute error
A wrong specification: an ingredient, a spec, a size, a material, a certification status.
4
Conflation
Mixing the brand with another entity, product, or competitor, including confusing a brand with a dupe or lookalike.
5
Material omission
Leaving out something that changes the read: a required certification, a safety warning, a real differentiator.
6
Unsupported claim
A claim stated with confidence that has no evidentiary basis behind it.
7
Framing error
True facts arranged in a way that misleads: negative framing, a false equivalence, or the wrong category placement.
The findings below come from the Maison Aubrenne report, an illustrative read AIVO produced on a fictional prestige skincare house to demonstrate the method end to end. The brand is invented. The structure, the severity scoring, and the evidence handling are exactly what a client receives. That read recorded 281 accuracy errors across three markets, and two error types accounted for 62% of them.
Critical: would change a purchase decision, misstates a certified claim, or creates regulatory exposure. Moderate: affects consideration without changing the decision. Minor: wrong, with no consequence. Instance counts are confirmed and carry no range. Recurrence rates are estimated from samples and carry a 95% range.
LAYER ONE: EXISTENCE
If a model fabricates a specific claim in a specific response, that is a deterministic fact about that one response. AIVO records it with the prompt, the response, and the supporting evidence. No confidence interval applies to something that has already happened.
LAYER TWO: RECURRENCE
How reliably does the model reproduce that same error when asked again? That is a rate, the error occurred in k of n repeated runs, scored with a 95% confidence range (calculated using the Wilson score interval).
EXAMPLE FINDING
Layer one: This fabrication is confirmed.
Layer two: It recurs in an estimated 90% of runs, with a 95% confidence range of 68% to 97%.
Illustrative example. Actual findings are drawn from the specific brand's measurement.
How much a given error matters to the brand is a human judgment call, scored after existence and recurrence are established, not blended into the rate.
AIVO is an independent AI brand representation firm. We measure how AI platforms describe a brand across presence, comparison, recommendation, accuracy, sentiment, sources, access, and audiences. Every rate is reported as a range at 95% confidence, per platform and per market, and every finding traces back to the prompts, responses, and citations behind it. We do not sell content, PR, technical remediation, or any other implementation, and we take no referral fee from the partners we recommend. The AI Brand Representation Report starts at $4,000 for one market, one language, and one core offering, delivered in 7 to 14 business days.
Independent measurement of how AI represents your brand. One market, one language, delivered in 7 to 14 business days.
30-minute scoping call · no commitment