Six small prompts that make a big difference.

"Prompt engineering" is a phrase that makes a small skill sound large. Yes, there are people who write hundred-line prompts with role definitions, formatting rules, and chain-of-thought instructions. They get great results. They also got those results because they iterated on their prompt for weeks.

You don't have weeks. You have a Tuesday afternoon and an email to send. Here are six phrases that, in our testing, consistently change AI output for the better, in ways you can feel after a single run. Use them at the end of your normal request, or sprinkle them in where they fit. Combine two or three at once for compound effect.

1. "Give me three options."

This is the most powerful single phrase you can add to almost any AI request. A single output reads like the AI's first guess. Three options reads like a thinking partner. The trick is that the model, asked for variety, will deliberately push two of them in different directions — so you get the safe version, the bold version, and something in between.

You'll throw away most of what comes back. That's fine. You'll find an angle you wouldn't have thought of alone.

Try with: caption generation, email drafts, name generation, headlines, taglines, "ways to phrase this awkward sentence."

2. "Explain it to a curious twelve-year-old."

The instruction "explain X simply" is the most-given and least-useful direction in AI writing. Models interpret "simply" as "shorter," which is not the same as "clearer." A short bad explanation is still a bad explanation.

"Explain it to a curious twelve-year-old" works better because it triggers a different mode entirely. The model picks concrete examples instead of abstractions. It uses analogies. It explains the why before the what. The output is often the explanation a competent adult actually wanted in the first place.

Try with: summaries of technical content, explanations of news, jargon translation, instructions you need to relay to a non-expert.

3. "Cut every sentence by half."

AI writing tends toward bloat — long sentences, every idea wrapped in qualifiers. Asking for "shorter" makes the response somewhat shorter. Asking to "cut every sentence by half" makes the response dramatically tighter, because it gives the model a concrete instruction to apply per sentence, not a vague mood.

Sometimes the result is too brief and you'll need to selectively restore. That's fine. It's easier to add back than to cut from scratch.

Try with: drafts that feel padded, blog post intros, anything you're trying to fit into a character limit.

4. "What am I missing?"

Most people use AI writing tools to expand on what they already wrote. The opposite use is more valuable: paste your draft and ask "what am I missing?" or "what's the strongest counter-argument here?" or "what would a skeptical reader push back on?"

The output won't always be right, but it's almost always thought-provoking. The point is not to follow every suggestion — it's to make sure you've at least considered them before sending the email or hitting publish.

Try with: proposals before sending, persuasive writing, any decision being communicated for the first time.

5. "Write in the style of [person or publication]."

"Write in a friendly tone" is a vague instruction. "Write in the style of The Economist" or "in the style of an Anthony Bourdain essay" or "in the style of a Pixar narrator" is concrete enough that the model has something specific to imitate.

This works because models have read enormous amounts of writing by well-known authors and outlets, so the stylistic pattern they associate with each name is dense and specific. The output won't be identical to that author, but it will pick up rhythm, vocabulary, and structural habits that "friendly" never would.

Try with: creative writing, brand copy, anything where the voice matters as much as the substance.

6. "Don't apologize. Don't summarize what you just said. Don't add a closing line."

This is the most boring tip and one of the most useful. AI writing has three structural tics that mark it instantly: it apologizes for being AI, it wraps with a sentence that restates the main point, and it ends with something like "let me know if you'd like me to adjust this!" Even the best models do this when not told otherwise.

Add the line above and the output drops directly into your document, ready to send. It's a small thing, but it saves the awkward editing pass where you strip out the AI-ness afterward.

Try with: any email or message you're going to send directly.

Combining them

The real wins come from stacking. "Give me three options, written in the style of a New Yorker columnist, each one cut to half the length you'd normally use, and don't add a closing line" is a single sentence appended to your normal request. It's not prompt engineering. It's just being specific in the same way you'd be specific to a freelancer you hired for the afternoon.

The phrases above aren't magic. They work because they replace mood-words ("friendly," "professional," "short") with concrete instructions. The model can do mood-words, but it does concrete instructions much, much better.

---

More reads: How AI writing tools actually work · The truth about AI detectors.

Want to test these? Try the Email Rewriter with the phrase "cut every sentence by half" added to your draft, or the Caption Generator with a specific style description.