Notes

Working notes on measuring AI answers

Problems people run into when AI assistants describe their organisation, and what can honestly be done about them. Every number here came from a recorded run.

  • Can AI-answer measurement tools evaluate regional, multilingual or locally hosted models?

    They can, provided the tool can actually call the model and keep country, language and hosting conditions in separate samples. A regional, multilingual or locally hosted result is a different test from a public English-language assistant, and averaging those results hides the market that is failing.

  • How do I get my business to show up in AI answers?

    Make your business easy to identify, connect it clearly to the customer needs it serves and publish those facts in crawlable, consistent pages. Then test repeated AI answers to learn whether the remaining problem is access, selection, accuracy or genuine fit, because no website change can guarantee a mention.

  • How long does it take for a website change to show up in AI answers?

    You changed your website and you want to know when AI assistants will reflect that change. The frustrating answer is that a website change can appear in a web-grounded answer within days while the model’s stored understanding takes months to change.

  • Measuring AI visibility when the answer changes every time

    You are trying to measure AI visibility, but the answer changes every time you ask. The problem is not only inconvenient variation, because a single response can make a weak presence look successful or make a real pattern look absent.

  • What to do when an AI describes your company as something it is not

    The AI describes your company as something it is not and customers may receive the wrong category before they ever reach your website. The difficult part is identifying whether the error comes from unclear public information, a retrieval failure or model memory that has not caught up.