Skip to content

AI hallucination

An AI hallucination is a statement from an AI model that sounds plausible but is false, such as a made-up measurement, feature or guarantee.

Also known as
  • Hallucination

What it means

Google’s Machine Learning Glossary describes a hallucination as “plausible-seeming but factually incorrect output by a generative AI model”.1 The model does not flag it as a guess. It can read just like the true sentences around it.

It happens because a language model writes what is likely, not what it has checked. Asked to describe a product with gaps in its data, it may fill them with details typical of similar products rather than true of this one.

In product copy the risk is a figure or feature that reads as fact: a capacity, a guarantee, a compatible model. Google’s rule for product details suits all generated copy: “Only provide an attribute name and value when the value is confirmed”.2

Think of it like this

It is like a confident stranger giving directions to a street they have never visited. The directions sound sure, use real street names, and still send you the wrong way.

For example

A bath’s row holds its type, size, colour and capacity, but no weight. Asked for a full description, a model writes “weighs just 28 kg”. Nothing in the row says so: the figure is a hallucination, however plausible.

Why it matters

An invented detail in your copy becomes an invented detail in an assistant’s answer, and can mean a return when the product arrives. UK law counts “the provision of false or misleading information relating to a product” as a misleading action.3 That is unfair if it is likely to lead the average consumer to a decision they would not otherwise have taken.4

Not to be confused with

Filling a gap is not always a hallucination. Product data enrichment fills a missing value from a source you can point to; a hallucination fills it from nowhere.

Sources

  1. Google for Developers. Machine Learning Glossary.
  2. Google Merchant Center Help. Product detail [product_detail].
  3. legislation.gov.uk. Digital Markets, Competition and Consumers Act 2024, section 226 (Misleading actions).
  4. legislation.gov.uk. Digital Markets, Competition and Consumers Act 2024, section 225 (Prohibition of unfair commercial practices).

Now try it onyour own file

Nothing goes live without your sign-offEvery value keeps its sourceYour data is never shared

Send us one supplier spreadsheet exactly as it arrived. We will show you the same products before and after, with a source on every value we add.

One email with the upload link. Nothing else lands in your inbox.