The Core Insight
Most advice about "prompting" teaches you to speak to machines differently than you speak to people. That creates a split in how you think and communicate. The better approach: develop communication habits that are inherently clear, structured, and intent-driven. These work everywhere.
Good communication with an LLM is just good communication. The techniques below aren't tricks — they're what effective communicators already do when speaking to colleagues, writing briefs, or giving instructions. You're not learning a new language. You're sharpening the one you already use.
Six Foundational Principles
People infer your goal from context, but they do it imperfectly. LLMs do it imperfectly too. Make your intent explicit: "I need help deciding between X and Y" is better than a long description of both options with no ask.
Humans tune out when you ramble. LLMs get diluted attention on long inputs too. Provide the context that matters for the specific request. If you're writing to a colleague, you wouldn't include their entire employment history.
Headings, numbered lists, and clear paragraph breaks help humans scan. They help LLMs parse intent. A well-structured email is also a well-structured prompt. Use formatting as thinking scaffolding, not decoration.
"Make it better" is useless to a human colleague and useless to an LLM. "Shorten the intro to two sentences and add a concrete example after the third paragraph" works for both. Specificity is a muscle — train it.
You rarely get the perfect answer from a colleague on the first ask. You ask follow-ups, clarify, redirect. Do the same with LLMs. Treat the first response as a draft, not a final answer.
A quick Slack message doesn't need the same rigor as a project brief. Neither does a casual question to an LLM need the same structure as a complex analytical task. Calibrate your effort to the importance of the output.
Before You Speak or Type: The 4-Second Filter
Run these four questions through your head before making any request — to a person or an LLM. It becomes automatic within a week.
Distinguish the outcome you want from the request you're about to make. Sometimes the request you were about to make doesn't serve the outcome.
Strip background to what's relevant. If your colleague already knows the project context, don't re-explain it. If you're starting fresh with an LLM, include only the context this specific ask requires.
Scan for pronouns with unclear referents ("fix it"), ambiguous scope ("the whole thing"), and vague quality words ("good", "professional"). Replace them.
Having a mental (or explicit) success criterion helps you evaluate responses. It also forces you to make your request precise enough that a good answer is possible.
Techniques That Transfer
-
Lead with the ask
Put your request at the top, context below. In emails, this is the "BLUF" method (Bottom Line Up Front). With LLMs, front-loaded requests get more consistent results. Both audiences appreciate not having to read three paragraphs before discovering what you want.
-
Use constraints as creative tools
"Write me something good" paralyzes. "Write a 150-word product description for a water bottle, tone: wry and understated" liberates. Constraints help humans focus and give LLMs a clear target. Apply them generously.
-
Show, don't just tell
"Like this: [example]" is the most powerful communication tool that exists. It works in design reviews, it works in creative briefs, and it works in prompts. One good example replaces a paragraph of description.
-
Separate multiple asks
If you need three things, number them. Don't bury a second request in the middle of a paragraph. Humans miss it. LLMs deprioritize it. Make each ask a visible, distinct item.
-
Specify the audience for the output
"Explain this for a new hire" and "explain this for our CTO" produce very different responses from both a colleague and an LLM. Always know — and state — who the output is for.
-
Set the format
"Can you give me a bulleted summary?" or "I need a one-paragraph response" — telling people the shape of the answer you want is normal. It's equally normal with LLMs. Don't leave format to chance.
-
Name your assumptions
"I'm assuming we're targeting the US market — correct me if that's wrong." This prevents both humans and LLMs from building on incorrect premises. Explicit assumptions are a gift to any listener.
-
Use role framing for perspective shifts
"Play devil's advocate on this" or "If you were the end user, what would confuse you?" works with colleagues in a brainstorm and with LLMs in analysis. You're not tricking the LLM — you're giving any listener a useful lens.
Side-by-Side: Before and After
These examples show the same intent expressed two ways. The "after" version works identically well whether sent to a coworker in Slack or pasted into an LLM.
Example 1 — Vague vs. Directed
"Can you look at my presentation and tell me what you think?"
"Can you review slides 4–7 of this presentation? I'm worried the data story isn't clear for a non-technical audience. Flag anything confusing and suggest restructuring if needed."
Example 2 — Context Dump vs. Scoped Context
"So we've been working on this project for like three months and the client keeps changing their mind and the team is stressed and I need help figuring out what to do about the timeline."
"I need help adjusting a project timeline. Context: client has changed scope twice in 3 months; team is at capacity. What I need: a revised 6-week plan that accounts for a likely third scope change. Happy to share the current plan."
Example 3 — Buried Ask vs. Front-Loaded
"I was reading about competitor X and their new feature and it reminded me of that idea we had last quarter and I think we should probably revisit it, what do you think about maybe putting together some thoughts on it?"
"I'd like to revisit our Q3 feature idea — Competitor X just launched something similar. Can you draft a one-page comparison of their approach vs. ours so we can decide if it's worth re-prioritizing?"
Example 4 — Open-Ended vs. Constrained
"Write something for the homepage."
"Write a hero section for our homepage. Audience: small business owners who've never used automation. Tone: confident but not corporate. Max 40 words for the headline, 20 for the subhead."
Patterns to Adopt in Everyday Speech
These are small shifts in how you phrase things habitually. They make you clearer to everyone.
| Pattern | Instead of… | Try… |
|---|---|---|
| Lead with intent | "So I was thinking about the design and—" | "I need feedback on the header design. Here's the context:—" |
| Number your asks | A paragraph with two questions buried inside | "Two things: 1) Can you check the math on slide 3? 2) Can you suggest a better chart type for the data?" |
| Replace vague with specific | "Make it more engaging" | "Add a surprising stat in the opening line and shorten the paragraph to 3 sentences" |
| Give format expectations | "Can you help with this?" | "Can you give me 3 bullet points summarizing this?" |
| Name the audience | "Write an intro for this talk" | "Write an opening for a room of 200 junior engineers — keep it approachable" |
| State constraints early | "I need a plan for the rollout" | "I need a 2-week rollout plan. We have 3 people and zero budget for ads." |
| Use examples | "Something like our usual style" | "Like this email we sent last month: [link or paste]" |
| Separate exploration from decisions | "Should we use React or Vue?" | "I'm exploring React vs. Vue — can you lay out 3 pros/cons of each? I'm not deciding yet." |
The Iteration Loop
Rarely does the first answer nail it — from a person or a machine. Build a habit of iterating quickly rather than trying to get everything perfect in one shot.
Don't spend 20 minutes crafting the perfect prompt or email. State your intent, give key context, and send it. You can always refine.
What's missing? What's off? Be specific in your own mind before giving feedback. "Not quite" is not useful feedback to anyone.
"This is good, but: the tone is too formal — make it conversational. And swap the third point for something about cost savings." Works for both audiences. The more precisely you redirect, the fewer rounds you need.
Diminishing returns are real. At some point, take the 90%-there output and polish it yourself. This is true whether you're editing a colleague's draft or an LLM's output.
The One Thing to Remember
You are not switching between "human mode" and "LLM mode." You are developing a single, unified communication habit: say what you mean, give the context that matters, structure it clearly, specify the shape of the answer you want, and iterate until it's right. The only thing that changes between audiences is how much shared context you can assume — and even that varies more between individual humans than between "humans" and "LLMs" as categories.
Start with one habit: always state your intent in the first sentence. Once that's automatic, add the next. Within a month, you'll find that your emails are better, your meetings are shorter, your Slack messages get faster replies, and your LLM conversations produce better results — all from the same set of habits.