Skip to main content

What makes a good prompt to track

The single highest-leverage decision in MentionBird — how to phrase prompts so they produce data you can act on.

Prompt choice decides how useful your entire account is. A well-phrased prompt returns brands, citations, and a clear picture of who's winning. A badly-phrased one returns paragraphs of generic prose, checked faithfully every day, telling you nothing.

Five rules cover most of it.

1. Never mention your own brand

The most important rule.

Track "What are the best running shoes for marathon training?" — not "Are Nike shoes good for marathons?"

The second question guarantees Nike appears, because you put it there. It measures nothing. The first tells you whether Nike gets recommended when nobody prompted it — which is the moment that actually influences a buyer.

The same applies to competitors. A prompt naming Hoka teaches you about Hoka's description, not about who wins the category.

2. Ask for a recommendation, not an explanation

Phrasings that ask which, best, or what should I buy tend to make an assistant look things up and name products. Those are the answers containing brands and citations — the data you came for.

Definitional and how-to phrasings usually get answered from the model's memory instead, with no sources and no brands. "What is a carbon-plated running shoe?" produces a competent paragraph and nothing you can act on.

The trap: comparing concepts instead of choices

This one catches almost everyone, because it looks like exactly the kind of comparison you're supposed to be tracking.

"Carbon plate vs traditional racing shoes" is a comparison — of technologies. A thorough answer explains how each works, who each suits, and names no brands at all. Track it for six months and you'll learn nothing about your position.

"Should I get carbon-plated shoes for my first marathon, and which ones?" is a comparison of choices. Now the answer has to name products.

The test to apply before adding any prompt:

Could a thorough answer to this be written without naming a single brand?

If yes, the prompt won't measure visibility, however natural it sounds. Adding "and which ones?" or "which should I buy?" to a conceptual question is often the whole fix.

Shapes that reliably return brands

  • Best X for Y — "Best running shoes for flat feet"

  • Which should I choose — "Which running shoe brand is best for beginners?"

  • What should I buy for a situation — "What shoes should I buy for my first marathon?"

  • Most recommended — "What are the most recommended running shoes for high arches?"

  • Worth it or not — "Are carbon-plated running shoes worth the money, and which are best?"

3. Leave the year out

Avoid "best running shoes 2026". A tracked prompt runs for months, and a year-stamped one drifts from what people are asking as soon as the calendar turns — while your historical trend line silently becomes a comparison between different questions.

Use relative phrasing if recency matters: "best running shoes right now".

4. Use the words your customers use

Track the question as a real person would type it, not as your marketing team would write it. Someone shopping for shoes asks about knee pain and wide feet and running a first marathon. They don't ask about performance footwear solutions.

If a phrase never appears in a support conversation or a sales call, it probably isn't worth a tracking slot.

5. Start narrow, then expand

A focused set of prompts you understand beats a long list you skim. Begin with the questions where being recommended would genuinely win you a customer, watch them for a week or two, then widen.

Every prompt occupies capacity on every platform you track it on, so a list assembled carelessly costs you the slots you'd rather spend on questions that matter. See How limits work.

Two questions before you add anything

  1. Could this be answered well without naming a brand? If so, it won't measure visibility.

  2. If an assistant answered this well, would a real buyer be closer to choosing? If it's idle curiosity, it isn't worth the slot.

If a prompt you've already added is returning nothing useful, Why a prompt returned no results or no citations covers how to diagnose it.

Did this answer your question?