TL;DR: Around two thirds of Birmingham’s high street shops had at least one detail stated incorrectly by a leading chatbot, according to research from AI visibility platform Searchable. The study ran over 4,200 queries through ChatGPT, Gemini and Perplexity, comparing each reply against details the retailers themselves had confirmed.

The systems were near-universally willing to answer — 99% of questions about who a business is, where it sits and how to reach it drew a response. Roughly one reply in 16 was false.

Geography proved the weakest point. About one answer in ten put a shop under the wrong postcode, and that held even where the prompt had specified the city. Roughly one in 14 pointed customers at a website that was not the retailer’s: where a shop’s trading name diverged from the address it actually uses online, the models tended to supply a tidier-sounding domain instead. Roughly one reply in 54 credited the shop to another brand entirely.

The gap between tools is large

Which assistant a customer happens to use changes their odds considerably. Wrong answers came back 11% of the time from Perplexity, 5% from Gemini, and 3% from ChatGPT. Across the UK cities examined, Birmingham placed third for misrepresentation — though the study found only London standing clearly apart, with the rest bunched inside its margin of error.

Why the burden falls on small retailers

Searchable’s Chris Donnelly attributed the pattern to how much evidence a business leaves online. Independents typically have a site and a Google Business listing and little else; chains are described across directories, reviews and press coverage many times over. A model with a single thin source has nothing to check it against, so it fills the gap.

The commercial consequence is unfair by construction. Searchable points to a 2026 Rithium survey in which 90% of AI users said they research products through these tools, 53% picked a retailer on an assistant’s recommendation, and 58% said their trust in a brand drops when AI describes it wrongly. The shop absorbs the damage from a mistake it neither made nor can observe.

Looking forward

Donnelly’s remedies are ordinary maintenance: keep the website and directory entries accurate, and accumulate corroborating mentions through local media, trade bodies and reviews. That is advice from a firm selling visibility services, and worth weighing as such — but the mechanism behind it is sound enough. These models repeat whatever sparse record they can find.