How the bot checks you

1. We read your website

You give us your business name, website, country and, optionally, competitors. The bot reads your site to work out what kind of business you are, what customers buy from you, any niches you focus on, and where you serve customers: a neighbourhood, village, town, city, county, region, state or the whole country.

It also runs a website readiness audit: whether AI crawlers are allowed in, and whether your pages are easy for a machine to understand.

2. We write the questions your customers ask

The questions are written blind. The question writer only gets a plain description of your market, such as "café in Ranelagh, Dublin", and what customers buy there. It never sees your website, your name, your selling points or your competitors, so every business in the same market gets the same kind of test.

Questions are unbranded. They never contain your name, the names of your own venues or products, or a competitor's name, because real customers usually ask for a recommendation, not about you. Any question that names one is dropped, and an incomplete question plan is regenerated or the scan fails openly.

Places follow how shoppers search: a neighbourhood, village, town, city, county, region, state or the whole country, never a street or address. If you sell online or across the country, the questions are asked nationally.

• Research questions: a customer early in their search asking for a few names worth a look
• Comparison questions: a customer weighing options
• Buying decision questions: a customer ready to choose

Every question asks the assistant to name businesses. The full check uses 25 questions (8 research, 8 comparison, 9 buying decision). The free check uses 5 (2 research, 2 comparison, 1 buying decision).

If your business has niches some shoppers look for on purpose, such as vegan brunch or dog-friendly rooms, 6 of the 25 full-check questions test them. They are written separately and shown under "Your specialties", so your AI score only reflects the general questions anyone shopping in your market asks.

3. We ask the assistants

Free check: the 5 questions are asked live on ChatGPT through Bright Data's ChatGPT scraper, giving 5 answers.

Full check: all 25 questions are asked to ChatGPT, Gemini and Claude through their APIs with web search turned on, giving 75 answers. Model versions are pinned so rescans are comparable.

API answers with search are not guaranteed to match what someone sees in a consumer chat app. Logged-in history, location, memory and app features can change what a person sees.

4. We score what came back

Whether you are mentioned is decided by a plain text match on your business name and the names of your own venues, branches and products, not by another AI's opinion. Several of your names in one answer count as one mention. A separate AI reader lists the real alternatives named, leaving out landmarks, attractions, review sites and booking platforms, and notes the tone of each answer.

Your score uses the general market questions only. Visibility is the share of successfully collected answers that mention you. The overall AI score combines visibility (40%), first picks (20%), share of brand mentions (15%, scaled up to 100), citations to your site (10%) and website readiness (15%). Failed answers are excluded. These are prompt samples, not market share or a ranking guarantee.

5. You get the report

Every question and every real answer is in the report, alongside competitors picked instead, sources the assistants relied on, the website audit and an action plan. Free reports show the score and a locked preview of the full details.

The report also names the market that was tested, for example "Café · Ranelagh, Dublin · Ireland", so you can see exactly where you were checked.

A scan usually takes several minutes, and can take longer when assistants are busy. Rate limits are waited out and failed questions get a second pass.

What this cannot tell you

• AI answers vary. The same question asked twice can name different businesses.
• Results reflect one country and one moment in time.
• Perplexity and other assistants are not checked.
• Following the action plan, or publishing an llms.txt file, does not guarantee recommendations. llms.txt is an experimental convention with no evidence that assistants rely on it.