gaash.ai

Templates & downloads

The prompt pack: ask the assistants yourself

Twenty questions written out for you to paste into ChatGPT, Claude, Gemini and Perplexity, with brackets for your trade, your town and your name. They are grouped the way a buyer's questions come: the problem before any names, the comparison, your own name, and the facts. Ask each one three times in a fresh chat, and write down every answer.

These are the same kinds of question our panel uses, written as templates. A single answer proves little, because the same question can name different businesses on different runs, so the pack comes with the way to run it. For scale: in the probe that accompanied our audit of 9,483 small business websites, one such question per business on Claude named 14.9% of 9,240 businesses.

  • Twenty questions, ready to paste
  • Grouped by kind of question
  • With the way to run them
Templates & downloadsgaash.ai

Problem questions: 1 to 6

Asked by a buyer who knows no names yet. These show whether a stranger hears yours.

  1. "who is the best [trade] in [town] for [the job]"
  2. "[the problem, in the buyer's words] in [town], who can fix it today"
  3. "[trade] near [neighborhood or landmark] open on [day]"
  4. "which [trade] in [town] does [a specific service] for [a kind of customer]"
  5. "how much does [the job] cost in [town], and who does it"
  6. "[trade] in [town] that [a condition buyers filter on, such as a language, weekend hours or a plan you take]"

Comparison questions: 7 to 10

Asked by a buyer weighing you against someone. These show how you are described beside a competitor.

  1. "[your business] or [competitor] for [the job]"
  2. "is [your business] cheaper than [competitor] for [the job]"
  3. "[your business] or [the chain or the dealer] for [the job]"
  4. "which is better for [a need]: [your business] or [competitor]"

Your own name: 11 to 15

Asked by a buyer who has heard of you. These show what the assistants say about you, and whether it is true.

  1. "what does [your business] in [town] do"
  2. "is [your business] any good for [the job]"
  3. "what are the opening hours of [your business]"
  4. "how do i book [your business], and do they take walk-ins"
  5. "is [your business] still open"

Fact checks: 16 to 20

The facts an answer most often repeats. Check each reply against what your front desk would say.

  1. "what is the phone number of [your business]"
  2. "where exactly is [your business], and which entrance do i use"
  3. "does [your business] do [a service you offer]"
  4. "does [your business] do [a service you do not offer]", which catches an answer that guesses
  5. "who runs [your business], and how long has it been open"

How to run them so the answers mean something

  1. Fill in the brackets once, then leave the wording alone. A question changed between months starts a new series.
  2. Write the place into every question. "Near me" means wherever you happen to be when you ask.
  3. Open a fresh chat for each question, signed out or with memory off where the assistant allows it, so the answer is closer to what a stranger gets.
  4. Ask each question three times in each of the four assistants, with web search on where it is offered.
  5. Paste every answer into a sheet with the date, the assistant and the run, and mark it: named, recommended, first, who else was named, and any wrong fact.
  6. Do it again next month with the same questions, and compare question by question.

Twenty questions, four assistants and three runs is 240 answers, a long morning by hand. Setting up prompt monitoring gives the sheet and a smaller panel to start with, and how to measure AI visibility explains how to tell a real change from the variation between runs.

Questions about the pack

Why ask each question three times?

Because the same question can name different businesses on different runs. One answer is a sample of one; three runs on each assistant show whether a result holds. Named in three of three is a different position from named in one of three.

Can I change the questions?

Before the first run, yes: make them sound like your buyers, in your trade and your town. After that, keep them fixed, so this month compares with the next. If a question turns out to be one nobody asks, retire it and add a new one as a new line.

What if an assistant gets a fact about us wrong?

Write down the wrong fact and, where the answer shows its sources, the page it came from. Then fix the source: your own page first, then the listing or directory that still states it. When AI gets facts about your business wrong covers the order.

Sources: our audit of 9,483 small-business sites, with its method · how the visibility score is measured · setting up prompt monitoring

Want the pack by email?

Leave your email and a person sends these twenty questions and the record sheet from hello@gaash.ai within one working day, as one message. No sequence, no newsletter.