← Blog

Prompting an AI Well: A Guide That Works

August 2, 2026

I used to get frustrated with “generic” AI answers until I noticed the pattern: almost every disappointing one traced back to my own vague prompt, not a weak model. A handful of concrete habits consistently produce better output.

Give It a Role, Not Just a Topic

"Write about marketing" invites generic filler. "You're a marketing consultant advising a 5-person bakery with a $200/month ad budget — suggest three concrete actions" gives the model constraints to reason within, which produces sharper, more specific answers.

Show, Don't Just Tell

If you want a particular tone or format, include a short example rather than describing it abstractly. Models are far better at matching a pattern you show them than interpreting an adjective like "professional but friendly."

Split the Big Ask Into Steps

A single sprawling prompt ("write my whole business plan") tends to produce shallow, generic coverage of everything. Asking for one section at a time, reviewing, then continuing produces noticeably deeper results per section.

Say What You Don’t Want, Too

Positive instructions alone leave room for the model's default habits (excessive hedging, listicle formatting, corporate tone) to creep back in. Explicitly naming what to avoid — "no bullet points, no summary at the end" — is often more effective than only saying what you want.

Make It Grade Its Own Answer

Following up with “review that answer for factual errors or unsupported claims” catches a surprising number of issues — a second, focused pass tends to surface things I missed on the first read too.

Refine, Don’t Restart

If an answer is close but not quite right, refining it in the same conversation (“keep this structure but make it more concise”) usually beats writing an entirely new prompt from scratch, since the model retains the useful context of what you already liked. I default to this now rather than starting over.