One in seven online stores turned away a shopping robot in a test

A company that helps online shops sell more sent a piece of shopping software to 200 stores in the US and UK. Twenty-nine of them, or 14.5%, refused to let it in. The company, Blend Commerce, blames the security software at the store's front door, not the stores' own stated rules.

What the test did

Blend checked four things: could the software get into the site, find products, read the product details, and carry a basket to checkout. Each store got a score out of 100. Testing ran from September 8 to 15, 2026. Nothing was bought, and the journey stopped at the checkout page.

The software was Blend's own. It did not pretend to be ChatGPT, Gemini or any other known assistant. So the results show how stores treat unfamiliar automation, which is not quite the same as how they treat the big AI assistants.

What it found

A store counted as a refusal only if it sent back an error, a challenge page or an empty page on both tries. Large retailers refused at a 20% rate. Emerging ones refused at 2.5%. The average score across all 200 stores was 65.

• Four stores scored zero: Saks Fifth Avenue, Bergdorf Goodman, Anthropologie and JD Williams.
• Baby and nursery stores scored best, at an average of 88.
• Fashion scored worst, at an average of 49.

The gap in the rules

Most sites publish a file called robots.txt, which tells automated visitors what they may do. Blend could read it for 24 of the 29 refusing stores. All 24 welcome the search bots from ChatGPT, Perplexity and Claude. Blend says the refusals come from bot protection, software that judges visitors by how they behave.

Blend CEO Adam Pearce said most marketing teams probably do not know this is happening. For the other five stores, the test holds no evidence either way. Four of them are the zero scorers, which refused even the request for that file.

What to do

Blend's report says plainly that if an agent cannot load your product page, it cannot recommend what is on it. Pearce urged companies to test this themselves before the biggest shopping day of the year, and to get security and marketing teams in one room. The report also notes that bot protection exists for good reasons, so the aim is balance, not switching it off.

What is not clear

Blend sells services in this area, and the test used its own software. The report does not publish how much each check is worth, or how it scored checks it could not run. How the big AI assistants would fare is not shown.