Chancela
When someone asks an AI for the best clinic in their city, a real business either appears in the answer or it does not. Chancela measures which, and the free audit runs live, question by question, in front of you.
Who has this problem
Search is turning into a single answer, and an answer names a few businesses instead of listing many. A local business that is not named can disappear from that answer. Measuring it requires a clean instrument that repeats the same question and records which source responded, without relying on the business owner’s personalized account.
My role
Chancela product architecture, measurement engine, reporting workflow and web application.
How to try it
- Open the product and fill in the business name, city and category. Competitors are optional.
- The free audit runs live, question by question, and shows where the business appeared and which sources the AI cited.
Usage conditions
- One engine, GPT with web search, and five questions, with a hard cost ceiling per audit.
- Five audits per connection per hour, and a daily cap across everyone.
- A bot check and a honeypot run before anything is spent.
- The full audit, with the other engines and every question of the category, is not what this button runs.
Public access has usage limits to keep AI service costs under control.
Decisions I made
- Business mentions use exact matches against normalized names and declared aliases. Domain association is checked separately by label boundary, so odonto.com.br does not match superodonto.com.br.
- A failed or cut-off provider response is not evidence that a business was absent. Each run tracks source health and marks the result inconclusive when a source is unavailable or an answer is incomplete.
- The report preserves the answer, source references, provider status and usage information for each question, making a result reviewable without treating a failed request as a negative mention.
- Fixes are ranked by impact times confidence divided by effort, and confidence comes from the run itself: the same fix ranks lower when the measurement behind it was worse. An impact figure taken from a published study carries its source into the report; the others go out without one.
- The structured-data fix never guesses. A field the site and the Google profile do not publish goes out as a visible placeholder to fill in, not as a plausible value: an invented address passes validation and misinforms.
Limits that remain
- The free audit runs one engine and five questions. The full audit, with the other engines and every question of the category, is not self-serve yet.
- Online checkout for the paid plans is not open yet.