1. Write the questions a customer would actually ask
Start from the business's category and city, not its name. Customers who don't know the business yet ask open questions, so a fair check uses open questions too. A useful set of about ten covers different intents:
- "Best" and "most trusted": What is the best roofing contractor in Austin, TX?
- Price: How much does a roofing contractor cost in Austin, TX, and which companies offer fair prices?
- Urgency: I need an emergency roofing contractor in Austin, TX today. Who should I call?
- Reviews: Affordable roofing contractor in Austin, TX with good reviews?
- Comparison: Compare the top 5 roofing contractor companies in Austin, TX.
- Trust signals: Which licensed and insured roofing contractors in Austin, TX do you recommend?
- Local and independent: Top-rated locally owned roofing contractor in Austin, TX?
Avoid putting the business name in the question. "Is Example Roofing any good?" tells you how an assistant describes a business it has been handed, not whether it recommends that business to someone who has never heard of it.
2. Ask every assistant, the same way
ChatGPT, Claude, Gemini and Perplexity often name different businesses for the same question. Ask each question to each assistant, in a fresh conversation, without earlier context that could steer the answer. Note whether web search was on: assistants that search the web before answering tend to name more current, local businesses than ones that answer from training data alone.
3. Record more than "yes or no"
For every answer, write down:
- whether the business was named, by name or by its website;
- its position if the answer is a list;
- every other business named — this becomes the competitor list and share of voice;
- which websites the assistant cited or linked to.
The last two are usually more useful than the yes/no. If the same review platform or directory keeps getting cited, a complete, accurate profile there is a concrete next step. If one competitor appears in most answers, compare its reviews, service pages and listings with your client's.
4. Match names carefully
Assistants shorten, re-order and sometimes misspell business names ("Example Roofing" vs. "Example Roofing & Exteriors LLC"), and businesses in different cities can share a name. Count a mention when the name clearly refers to the same business — or the answer links to its domain — and double-check near-matches by hand.
5. Treat it as a snapshot
Answers change between runs, between days and between users. The consumer apps can also personalize results by location and history, and the models behind them change over time. One check is a snapshot; the trend over several monthly checks says more than any single result.
The catch: it doesn't scale by hand
Ten questions across four assistants is 40 answers per business — and each answer needs reading, name-matching, competitor extraction and source tallying. For one prospect that's an afternoon. For a client list, every month, it becomes a job of its own.
That's what AnswerRoster automates: it asks the questions through the providers' APIs, checks each answer for the business's name and domain, extracts competitors and cited sources, and produces a white-label report under your agency's brand. See an example report walkthrough or the sample PDF.