Choosing your ten prompts

An allocation that balances

A workable split across ten: three category, three comparison, two recommendation, two brand. That gives you enough of each type to notice a pattern, and it deliberately weights toward the top of the journey where most brands are weakest.

Adjust for your situation. If you are in a category where buyers already know they need the thing, drop a category prompt and add a recommendation one. If you suspect an accuracy problem, take a second brand prompt.

Where to source them

The best prompts come from three places, in this order of quality. First, questions real buyers asked you — sales call notes, support tickets, the questions that come up in demos. These are unimpeachable because they actually happened.

Second, what the engines themselves surface. Ask an answer engine a broad question about your category and read what follow-up questions it suggests or what sub-topics it volunteers. That is a direct read on how the engine has structured the space.

Third, demand data — the aggregate of what people are asking AI in your industry. Useful for coverage, weaker for specificity, because a high-volume generic prompt often has less decision-weight than a low-volume specific one.

The freeze rule

Once you take your baseline, do not change the prompt set until after your verification run in Module 6. This is the rule that makes the whole experiment valid.

If you edit a prompt between baseline and follow-up, you cannot attribute a change in the number to your content work, because the question itself changed. This sounds obvious and is violated constantly, usually with the excuse that the prompt was badly worded. If a prompt is badly worded, fix it now — before the baseline — and accept that anything you change afterwards is a new experiment, not a continuation of this one.

Where this lives in SolCrys

• Add your ten prompts on the workspace prompts surface.
• Review anything the generator produced and rewrite it into an actual buyer question.
• Read them once end to end before you run anything. This is your last cheap chance to fix wording.

Lab: Build and freeze the set

Needs a free SolCrys account (no credit card). No quota consumed.

1. Write ten prompts, roughly three category, three comparison, two recommendation, two brand.
2. Source at least three of them from questions real buyers actually asked you.
3. Read the full set aloud. Anything that sounds like a keyword rather than a question, rewrite.
4. Stop. Do not run yet — the next lesson spends the first of three runs.

This is the last no-quota step. Everything after this costs part of your budget.

Start Free — free plan, no credit card.

Check yourself

• How many of your ten came from something a real buyer actually said?
• If you had to defend each prompt as representing a real decision, which one would you struggle with?
• Why does editing a prompt after the baseline invalidate the comparison?

Go deeper

This lesson is the spine. These guides are the depth.

Continue

• Previous: A prompt set is not a keyword list
• Next: Your first run: taking a baseline