How an Aivosity report is made

We start with three things: your business name, your industry, and where you work. From those, our question generator writes thirty questions in the shape people actually use with an assistant — early research ("what should I look for in…") through to the last step ("which one should I book…").

We vary the wording and the intent on purpose, and we never put your name in a question. The whole point is to see whether a model brings you up unprompted, the way it would for someone who had never heard of you.

• Which physiotherapy clinics in Seville are best for sports injuries?
• I run a 12-room hotel in Girona. Which channel manager should I use?
• Recommend an English-speaking immigration lawyer in Alicante.
• What's a reasonable price for solar panel installation in Murcia, and who is reliable?
• Best coworking space in Valencia for a remote team of five?
• Is there an accountant in Barcelona who specialises in freelancers (autónomos)?

Every question goes to ChatGPT, Gemini, Claude and Perplexity through their official APIs. We leave the models in their normal, consumer-style mode rather than coaxing them with a clever prompt, and where a model can ground its answer in live web results, that is the answer we record — it is the one a buyer would get.

For every response we store the full text, the businesses named and the order they appear in, the claims made about each business, and any sources or citations the model provides. Responses are captured within a single window so the four models are compared on equal footing.

Answers wobble from one run to the next, so each question is asked three times per model. Your mention rate comes from all of those runs together — 360 answers in total — not from one lucky or unlucky reply.

Structured-data analysis turns hundreds of free-text answers into comparable measurements. We identify every business mentioned and resolve variations of the same name, then compute:

• Mention rate: the share of question-runs where a model names your business, overall and per model.
• Position: where you appear when you are mentioned, since "first choice" and "also consider" are very different outcomes.
• Competitor ranking: the businesses each model recommends most often for your questions, whether or not they are on your own competitor list.
• Sentiment and framing: whether you are described as premium, budget, specialist, generalist, and whether that matches how you position yourself.

We take every claim a model makes about you and hold it against the facts you gave us and what your own site says — hours, services, prices, locations, certifications, who works there. Anything that does not line up is flagged, with the sentence quoted in full so you can judge it yourself.

Then we look at where the information came from. Cited sources are collected and grouped — your own site, directories, review platforms, press, the odd blog — and when a model cites nothing, we crawl the public web to find the likeliest origin of a stale claim. Fixing the source is nearly always faster than arguing with the model.

Last, the crawler reads your site the way a model would and scores what it can actually use. Does the page say plainly what you do, for whom and where? Is the structured data (Organization, LocalBusiness, Service, FAQ) there and does it match the visible text? Are your hours and prices trapped in an image? Do your robots rules or load times keep crawlers out? We only fetch pages that are public and we respect robots directives.

All of it lands in one report: the numbers, the competitor table, the error list, the source map, the readiness score and a prioritised plan. You get a PDF for sharing and a private web version with the full response tables, and you can book a call if you would rather we walked you through it.