AI Visibility Test Kit

Act as an AI search analyst for a small business. Build a test I can run by hand each month to see whether AI assistants mention us, and how.

Business [NAME, WHAT WE DO] · where we sell [CITY, REGION OR ONLINE] · customers [WHO BUYS, AND WHAT THEY ASK BEFORE BUYING] · competitors [3 TO 5 NAMES] · what we want to be known for [TOPICS OR SERVICES] · assistants to test [CHATGPT / CLAUDE / GEMINI / PERPLEXITY / COPILOT / GOOGLE AI OVERVIEWS] · our website [URL]

Build:
1. Question set: 12 questions a customer would type, in their words: four asking for a recommendation, four comparing options, two asking how to do something we help with, and two that name us directly. Only those last two contain our name
2. How to run it: the same steps for every assistant, so the months compare: a new chat, the question pasted exactly, the date noted, and what to do about memory or account history where the assistant has it
3. Scoring key for each answer: MENTIONED FIRST, MENTIONED, NOT MENTIONED or WRONG FACTS ABOUT US, plus which competitors appear and which sources the answer cites
4. Log sheet as a table I can paste into a spreadsheet: date, assistant, question, score, competitors named, sources cited, what was wrong
5. Monthly summary: the counts to take from the log and the three comparisons worth making with the month before
6. What the test cannot tell me, in three lines

Write the questions from what I said customers ask, and do not tune them so that we come out well, because a flattering test hides the problem it is meant to find. Predict no results and state no fact about how any assistant chooses what to mention. Answers vary from run to run, so the log records what was seen on one date, and the summary counts and compares without ranking.

Success means the 12 questions could be pasted as they are, the same steps work for every assistant I listed, and every column in the log can be filled from one answer.

Apply every rule, limit, and format above to every item in the deliverable, not just the first one. Deliver the complete result in one response, every section filled in, nothing deferred to a follow-up.

If a required input is missing, ask for it once before starting; otherwise proceed and label any assumption you made.

Notes: R54 build; tuned to vendor prompting guidance