How we measure what AI says about your business.

1. The questions

There are two kinds of questions:

• Discovery questions, without the business name, like someone who doesn't know it yet: “What is the best dentist in Leeds?”, “Recommend a trustworthy garage in Dublin.” These are the ones that make up the score.
• Brand questions, with the business name: “What do you know about Smile Dental in Leeds?” They show whether AI knows the business and what it says about it.

On the paid plans, every question is asked twice to each assistant, because the same question can get different answers.

Paid-plan customers can add questions of their own (3 on Essential, 5 on Professional). We ask them the same way and show the answers in the report, but they don't count towards the score, so it stays comparable between months and between businesses.

The questions are written the way customers in the business's country talk. For the most common business types in Portugal we use fixed questions, reviewed by hand. For every other type and country, an AI model writes the questions in the country's language once, and every business of the same type in the same city gets exactly the same ones, so they can be compared.

2. The assistants

We ask through each assistant's official API, with web search on and set to the business's country, as when a customer asks on their phone. We use no history, account or anyone's preferences: it is the answer a new customer would get.

The same question is asked to the same assistant at most once a week. Two businesses of the same type in the same city share that week's answers to the discovery questions, which also makes them comparable.

3. Reading the answers

An AI model reads every answer and lists, in order, the specific businesses it recommends. It ignores platforms and directories (Google Maps, TripAdvisor, Yelp) and businesses the answer says are in another city.

We count an answer as mentioning your business when it gives the same website, or when the name has the same distinctive words. “Smith Auto Ltd” and “Smith Auto Repairs” count as the same business; “Station Car Wash” and “Clean Station Car” do not.

4. The score from 0 to 100

Each discovery answer earns points depending on where the business appears:

The score is the total points divided by the maximum possible (3 points in every answer), on a scale of 0 to 100. A business that always appears first scores 100; one that never appears scores 0.

Competitors are scored the same way, in the same answers, so the scores compare directly. We also calculate a score per assistant, using only that assistant's answers.

5. The margin of error

AI answers vary from one request to the next. That is why we show the score's margin of error, at 95% confidence, calculated from how much the points vary between answers. With more questions and repetitions, as on the paid plans, the margin gets smaller.

6. What AI knows about the business

Each answer to a brand question is classed as “yes” (describes the business with concrete facts), “partly” (vague or uncertain information) or “no” (doesn't know it or talks about another one). The result is “yes” when at least half the answers are, and “no” when they all are.

7. The website check

When you give us your website, we visit the home page the way an AI assistant does and check:

• whether the site opens and uses a secure connection (HTTPS);
• whether it lets AI assistants' crawlers in, in robots.txt and in practice;
• whether the content can be read without JavaScript, which assistants usually don't run;
• whether it has the business markup in schema.org (LocalBusiness), with address, phone and opening hours;
• whether it has an llms.txt file summarising the site;
• whether it mentions the city and shows the phone and address as text;
• whether the page has a title and a description.

Limitations

• It is a sample: it shows the trend, not a guarantee of what each customer will see.
• Assistants change models and sources often; that is why the paid plans repeat the measurement every month.