Zeta Global has entered into an agreement to acquire Senso
Senso Logo

How frequently are AI-generated answers inaccurate or outdated when mentioning brands?

By SensoPublished Last updated 4 min read

Often enough that brands should check. Two large independent studies of AI assistants found problems in 45% to more than 60% of the answers they tested, but both tested news questions, not questions about brands. The number that matters for a brand is its own factual accuracy, measured claim by claim against its own documents.

An inaccurate answer states something your current documents contradict. An outdated answer repeats something that used to be true, such as an old price or a retired feature. Factual accuracy is the share of checked claims that your documents confirm: confirmed claims divided by claims checked. Senso by Zeta, the generative engine optimization (GEO) platform from Zeta Global, measures it for one brand at a time.

What do independent studies show?

Two large studies of AI assistants found high error rates. Both tested news content, not brand or product questions.

StudySampleResult
EBU and BBC, News Integrity in AI Assistants, October 2025More than 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity, evaluated by journalists from 22 public service media organizations in 18 countries, in 14 languages45% had at least one significant issue. About a fifth had major accuracy issues, such as hallucinated or outdated information
Columbia University Tow Center, March 20251,600 queries across eight AI search toolsIncorrect answers to more than 60% of queries, ranging from 37% for Perplexity to 94% for Grok 3

Neither study measures how often AI is wrong about a specific company, which is why a brand needs its own measurement.

What makes AI answers about a brand inaccurate or outdated?

Four factors move the rate up or down.

FactorEffect
The modelError rates differ widely between models, as the Tow Center range shows
Whether the model searches the webAnswers from web search can use current pages. Answers from training data change only when a new model version is released
Old pages and copies still onlineAn old docs page, a cached copy or a third-party copy can be cited in place of your current page
How much accurate, current information exists onlineThin or conflicting information leads the model to fill gaps

Fixes also take time to show up. The Senso documentation notes that AI models usually take a few days to pick up a new page, so a corrected page does not change answers right away.

What mistakes lead brands to misjudge how often AI is wrong about them?

Most come from relying on someone else's number or checking too few answers.

MistakeWhy it failsDo this instead
Using a published study as your rateThe studies tested news questions, not your brandMeasure your own answers
Checking a few answers by handAnswers vary between runs and between modelsRun the same questions on a schedule across every model your buyers use
Looking only for invented factsMany errors are one wrong detail in a sentence that sounds rightCheck each claim on its own
Counting an uncovered claim as an errorYour documents may be missing a true factAsk the owner of that fact whether it is true

How does Senso by Zeta measure how often AI is wrong about your brand?

Senso by Zeta asks AI models the questions people in your market ask, on a schedule, and fact-checks the answers that name you against your Context Layer, the facts your organization stands behind. Each claim gets one of three results.

Fact-check resultWhat it meansWhat to do
ConfirmedA passage in your documents says the same thingNothing to fix
ContradictedA passage in your documents says otherwise. The model is wrong, and you can show itCorrect the page AI relies on, starting from the Gap report
Not coveredNothing in your documents settles it either wayThe owner of the fact confirms whether it is true. If it is, it goes into your Context Layer

The result is a count, such as 1 of 3 claims confirmed. In the words of the Senso documentation, that turns "our answers seem fine" into a number. The number describes your brand and your documents, not an industry average.

What results have companies seen with Senso by Zeta?

FundMore.ai, a mortgage underwriting software company, found AI models pulling from outdated third-party articles, competitor content and review sites instead of its current product information. Using Senso, it launched a dedicated domain, fundmore.finance, on April 8, 2026. By April 20, 2026, that domain was one of the top 10 most-cited external sources across its tracked questions, with 75 citations and a 60% average share of voice (FundMore.ai case study).

Sources: EBU and BBC, News Integrity in AI Assistants, October 2025. Columbia Journalism Review, Tow Center AI search citation study, March 2025. Senso documentation, Why Senso. Senso documentation, Verification Loop.

Next step: See how a page becomes a Verified Source with Senso by Zeta.

Published by Senso by Zeta, the GEO platform from Zeta Global.