The 15 Prompts to Test Your AI Visibility Every Month

Use the same fifteen prompts every month, in fresh sessions, across ChatGPT, Perplexity, and Gemini. A fixed set is the entire point — changing the prompts between checks makes the results incomparable, which is the most common mistake in AI visibility tracking.

Fill in the bracketed parts once, save the list, and run it on the same day each month. Twenty minutes.

Why Does the Set Have to Be Fixed?

Because these systems are non-deterministic. The same prompt can produce different answers on consecutive days with nothing having changed.

The only way to see signal through that noise is a stable set run repeatedly over months. Fifteen prompts across three assistants is forty-five data points per month — enough that week-to-week randomness averages out and a real trend becomes visible.

Changing prompts between months is how a flat result gets mistaken for progress, and it's the same failure as a shifting keyword set in an SEO report.

Group 1: Do They Know You? (Prompts 1–3)

1. "What can you tell me about [Business Name] in [City]?"

The baseline. Do they know you exist, and are the basics right?

2. "Is [Business Name] any good?"

Reveals what it's drawing from reviews and whether the sentiment picture is accurate.

3. "What are [Business Name]'s hours and where are they located?"

The accuracy check. Wrong answers here point at a citation cleanup job.

Group 2: Are You Recommended? (Prompts 4–7)

4. "Who's the best [your category] in [your city]?"

The broad version. Hardest to win.

5. "Who's the best [your category] in [your neighborhood]?"

The neighborhood version. Much more winnable and much higher intent.

6. "Can you recommend a [your category] near [local landmark]?"

Tests a differently-shaped geographic constraint.

7. "I need a [your category] in [city]. Give me three options with reasons."

Forces a comparative answer, which is where you find out what it thinks distinguishes each business.

Group 3: The Situation Questions (Prompts 8–11)

These matter most, because they're the shape people actually use with assistants. Longer, constrained, with a real problem attached.

8. "I'm in [neighborhood] and [specific realistic problem]. Who should I call and roughly what should I expect to pay?"

The single most representative query on this list.

9. "I need [your service] but [common constraint — a tight timeline, a budget limit, an access issue]. Any suggestions in [city]?"

Tests whether you match constrained needs.

10. "What should I look for when hiring a [your category] in [city]?"

An advice question rather than a recommendation question. If your content gets cited here, your educational content is working.

11. "[Common customer question in your industry]?"

Your own most-asked question, verbatim. Tests whether your FAQ content is being retrieved.

Group 4: Commercial and Comparative (Prompts 12–15)

12. "What does [your service] cost in [city]?"

One of the most common questions asked about any local service, and one where publishing real local figures makes you citable.

13. "How does [Business Name] compare to other [category] in [city]?"

Reveals your perceived differentiation, if any exists.

14. "Is [Business Name] or [generic competitor description] better for [use case]?"

Note: describe a competitor type, not a named local business.

15. "What are the top [category] companies in [city]?"

The list-style question, often answered from roundup articles — which tells you which roundups to pursue.

The Follow-Up Question

After any prompt where the answer matters, ask: "Where did you get that information?"

Log the sources. Over three months you'll have a ranked picture of which pages actually influence what assistants say in your category, and that list is worth more than the visibility scores themselves.

How Should You Run It?

Fresh sessions every time. Temporary chat in ChatGPT, new threads elsewhere. Logged out if possible.

Same day each month. Consistency matters more than the specific date.

All three assistants. They draw on overlapping but different sources and the differences are informative.

Log in a spreadsheet. Columns: prompt number, assistant, month, named (yes/no), accurate (yes/no), sources cited, notes.

How Do You Read the Results?

Score it simply. Count how many of the forty-five checks named you. That's your monthly number.

Track it over months, not weeks. Three in forty-five in January and eighteen in July is real progress even if June looked worse than May.

Watch the groups separately. Group 1 problems are accuracy problems — fix citations. Group 2 and 3 problems are corroboration problems — fix reviews, content, and third-party mentions. Group 4 problems mean you're absent from the comparative and pricing content people are asking about.

Note who is named instead of you. Their sources are your targets.

What Should You Not Expect?

Consistency. You will be named in one check and absent the next. That is the nature of these systems and not a sign the work failed.

A guarantee. Nobody controls what an AI assistant says. Any provider promising you'll be recommended is describing something they can't deliver.

Fast movement. This tracks the same underlying signals as local search — profile completeness, review substance, content specificity, third-party corroboration — and those move over months.

Key Takeaways

• The prompt set must stay fixed; changing it between months makes results incomparable.

• Fifteen prompts across three assistants gives forty-five monthly data points, enough for randomness to average out.

• Group the prompts so you can tell an accuracy problem from a corroboration problem.

• Always log the cited sources — over three months that list is more valuable than the score.

Frequently Asked Questions

How do I track my business's visibility in AI search?

Run a fixed set of about fifteen customer-shaped prompts monthly across ChatGPT, Perplexity, and Gemini in fresh sessions, logging whether you're named, whether the information is accurate, and which sources are cited. Keeping the prompt set identical between months is what makes the results comparable.

How many prompts should I test for AI visibility?

Around fifteen, grouped so you can distinguish problems: a few checking whether assistants know you and have accurate details, several checking whether you're recommended in your category and area, several situation-based questions, and a few commercial or comparative ones.

Why do AI results change every time I check?

These systems are non-deterministic — the same prompt can produce different answers on consecutive days with nothing having changed. This is why a fixed prompt set run monthly over several months is necessary, and why any single check should not be treated as a result.

Can an agency guarantee my business appears in AI search?

No. No provider controls what AI assistants say, and a guarantee is a claim nobody can deliver. What can be improved is the underlying source material — profile completeness, information consistency, review substance, specific site content, and third-party mentions.

Want This Tracked Monthly?

Watt Consulting runs a fixed prompt set for Seattle businesses across three assistants and reports the trend with the sources behind it. Start your project.