How to choose the buyer prompts you track

When you track AI visibility, the prompts are the measurement. A score of 62 means nothing on its own; it means "across these particular questions, on these engines, this is how often and how prominently we were recommended". Choose questions nobody asks and you will optimise for an audience that doesn't exist. Choose only questions you already win and you will feel great while losing deals.

This guide is a practical method for building a prompt set that reflects how your buyers actually ask — and for keeping it useful over time.

In short

• Start from buyer situations, not keywords. Real prompts are long, specific and conversational.
• Cover the four buying intents: best X for Y, X vs Y, alternatives to X and how to….
• Keep most prompts unbranded — they measure whether you're recommended when nobody asked about you.
• Change the set deliberately and rarely, so your trend stays comparable over time.

Why prompt selection matters so much

AI assistants answer the question in front of them. "Best CRM" and "best CRM for a ten-person agency that bills hourly" can produce completely different lists, because the second one triggers different searches and rewards different pages. Your prompt set decides which of those realities you see.

A good set has three properties:

• Representative — it mirrors the questions real buyers ask, in their words.
• Balanced — it covers different intents and segments, so one type of question doesn't dominate the score.
• Stable — it changes slowly, so a movement in the score reflects a change in the answers rather than a change in the questions.

Start from buying situations

Before writing a single prompt, list the situations in which someone would go looking for a product like yours. For each, note who they are, what triggered the search and what constraints they mention.

A useful way to structure this is the four intents most buying questions fall into:

The brand names above are fictional; use your own market's. "How to" prompts are easy to forget, but they catch buyers earlier — before they know which category of product they need.

Where to find real wording

The best prompts come from the language your buyers already use. Good sources:

• Sales calls and demo notes. What did prospects say they were looking for, and what else were they considering?
• Support tickets and onboarding questions. These reveal the problems people expected your product to solve.
• Reviews of you and your competitors. Review sites are full of phrases like "we needed something for a team of five that…".
• Communities. Forums, subreddits and Slack groups where your buyers ask each other for recommendations.
• Your search data. Question-style queries in your search console and site search are a strong signal of real phrasing.
• Your own team. Ask sales, success and support for the five questions they hear most.

How to write a good prompt

Once you have raw material, turn it into prompts that behave like a real buyer's question.

Write the way people talk to assistants

People type full sentences to AI assistants, often with context. "Best CRM" is a search keyword; "what CRM should a ten-person B2B sales team use if we live in Gmail?" is a prompt.

Include the constraints that matter

Team size, budget, industry, region, integrations and must-have features all change the answer. If a constraint is common among your buyers, put it in some of your prompts.

Keep most prompts unbranded

If your brand is in the prompt, the assistant will almost always mention you — that tells you little. Unbranded prompts measure whether you are recommended unprompted, which is the point. Keep branded prompts for specific questions such as "X vs Y" comparisons and "alternatives to [competitor]".

Avoid leading questions

"Why is Northwind the best CRM for agencies?" presupposes the answer. Ask neutrally, the way a buyer who hasn't decided would.

One question per prompt

Compound prompts ("what's the best CRM and how much does it cost?") produce answers that are harder to compare over time. Split them.

Use your market's language

If you sell in Germany, track German prompts. Assistants answer in the language they're asked, and the sources they retrieve differ by language and country.

How many prompts, and in what mix

More prompts give a more stable, more representative score, but each one costs a check on every engine, every week. A practical starting point:

• 25–50 prompts per brand for a single product and market.
• A rough mix to start from (not a rule): around four in ten "best X for Y", and the rest split across comparisons, alternatives and how-to questions.
• More prompts for the segments that matter most commercially — if half your revenue comes from agencies, half your "best X for Y" prompts can mention agencies.

When you add a brand in Mentioned, it reads your homepage and proposes 25 prompts across the four intents for you to edit. Paid plans track 50 prompts per brand, and you can add more with prompt packs.

Organise prompts so you can read the results

A flat list of 50 prompts is hard to reason about. Group them so the dashboard answers specific questions:

• Tag by intent — so you can see that you win "best for" questions but lose comparisons.
• Tag by segment or persona — agencies vs in-house teams, SMB vs enterprise.
• Tag by product line — if you have more than one.
• Mark a few priorities — the handful of questions closest to revenue. In Mentioned these are pinned prompts; on Enterprise plans pinned prompts can be checked daily.

Mentioned supports tags, filters and CSV import and export, so you can draft the set in a spreadsheet with your team and import it in one go.

Keep the set stable — and change it deliberately

Your visibility score is calculated across your active prompts. Add ten prompts you happen to win, and the score rises without a single answer changing. That is why changes to the set should be:

• Rare — review quarterly rather than weekly.
• Batched — make changes together, so there is one clear "set changed" moment in the trend.
• Recorded — note what changed and why, so nobody mistakes a methodology change for a marketing win.

Retire prompts that no longer reflect how buyers ask, or that cover products you no longer sell. Add prompts when you launch into a new segment, region or category.

A worked example

Here is an illustrative starting set for a fictional CRM, "Northwind", that sells to small B2B sales teams:

Best X for Y

• What's the best CRM for a ten-person B2B sales team?
• Which CRM works best for a sales team that lives in Gmail?
• What's a simple CRM for a startup's first sales hire?
• Best CRM with built-in email sequences for small teams?

X vs Y

• Northwind vs Fieldline — which is better for a small sales team?
• Is Kestrel or Northwind easier to set up?

Alternatives to X

• What are cheaper alternatives to Kestrel CRM?
• Alternatives to Fieldline for a team that needs better reporting?

How to…

• How do I track deals without a spreadsheet?
• How can a small team automate follow-ups after a demo?
• How do I forecast pipeline for a five-person sales team?

Notice the pattern: specific situations, natural wording, and the brand appearing only where the buyer would name it.

Checklist

• Situations listed before prompts written
• All four intents covered
• Mostly unbranded prompts; branded ones limited to comparisons and alternatives
• Constraints that matter to your buyers included
• Prompts in each market's language
• Tags for intent, segment and priority
• A quarterly review on the calendar

Once your set is in place, the free audit gives you a first read on two engines, and Mentioned tracks the full set weekly across all four.