One delivery question. Two different AI answers.

The buying problem

For a local retailer, a delivery promise can determine whether a buyer proceeds. We checked both non-brand discovery questions and questions about the retailer's advertised service conditions.

What we checked

Six questions were checked in fresh, anonymous ChatGPT and Gemini sessions, once per platform. The review retained 12 full answer records and compared relevant statements with website policies.

This was an English-language diagnostic. Model versions were not fully exposed, location was not controlled, and no repeat round or real order was completed.

What the evidence showed

On one delivery question, one assistant rejected same-day service while the other described a prepaid same-day option and a slower general policy. Different website pages advertised different conditions, so the real policy needed confirmation.

The brand or associated candidates appeared in several discovery answers. A claim that the business was completely invisible would have misrepresented the evidence.

The recommended change

Make one service-policy table the reference for every relevant page, then rerun the original questions.

• Confirm coverage by postcode, service level, payment method and order cutoff.
• Reconcile general policies, local pages and FAQs against that confirmed table.
• Repeat the same questions in two further rounds and record both accuracy and mentions.

Where the work stands

The diagnosis and delivery draft were completed. Service-policy confirmation, website implementation and repeat testing were still pending in the source record. No increase in traffic, orders or AI recommendations has been measured.