By Clint Simon
Most people I know use ChatGPT, Claude, Copilot, or Gemini the same way they used Google in 2008. They type a question, skim the answer, and move on. That works, but it leaves about 80% of the value on the table.
I have been using these tools all day, every day, for a couple of years now. The difference between a casual user and someone who actually gets real value out of an AI assistant is not the tool. It is a small number of habits. None of them are technical. None of them require coding. They take five minutes to learn and they pay back forever.
This guide is for information workers, the people who spend their days in email, docs, meetings, spreadsheets, and slides. PMs, managers, analysts, marketers, HR, ops, finance, sales, admins, consultants. If your job involves writing things, deciding things, or making sense of things, this guide is for you.
I am not going to explain what AI is or how it works. I am going to show you the moves I actually use. Each technique has a short why, a short how, and a few real examples you can copy.
TL;DR: Treat your AI like a thoughtful colleague, not a search box. Teach it who you are. Give it real context. Ask it to plan. Iterate. Verify.
These are the three beliefs you need to internalize. Everything else follows.
An AI assistant is fast, well-read, eager, and occasionally wrong with great confidence. That is exactly what a smart intern looks like. You would never ship an intern's first draft straight to your VP. Same here. The intern is a force multiplier when you direct them well, review their work, and give them context. Treat your AI the same way.
The first response is rarely the best one. The best response usually shows up in turn three or four, after you have said "shorter", "less formal", "lead with the result", or "what would change your mind?". Plan to have a conversation, not a transaction.
Names, dates, numbers, quotes, policy language, legal wording, medical information. Always check. AI assistants will produce plausible-looking content that is simply wrong. The fix is not to stop using them. The fix is to treat their output like a sharp first draft from a colleague who is new to your domain.
TL;DR: Smart intern, not oracle. Iterate, do not transact. Verify the parts that matter.
This is the single best move in the entire guide. If you only do one thing, do this one.
Every time you start a new chat, your AI knows nothing about you. It does not know your role, your team, your writing style, your stakeholders, your deadlines, your preferences. So it gives you a generic answer that sounds like it was written for everyone. Which means it was written for nobody.
The fix is to write down who you are, once, and put it somewhere the AI will read every time. Every major assistant supports this:
- ChatGPT: Settings, Personalization, Custom Instructions. Also Memory.
- Claude: Projects (each Project has a system prompt). Also Styles.
- Copilot: Pages and Notebooks let you attach persistent context.
- Gemini: Saved Info, and Gems for repeatable use cases.
Write a profile in plain English. Three to ten sentences is plenty to start. Cover four things: who you are, what you do, how you write, and what you care about.
Then give it your style by example. Paste two or three things you have written that you are proud of. Say "this is how I write. Match it."
A starter profile (drop this into ChatGPT Custom Instructions or a Claude Project):
I am a senior product manager at Acme Software. I own the billing experience
for our small business customers. My peers are engineering leads and design
leads. I report to the VP of Product. I write in plain, direct English.
Short sentences. No corporate jargon. I lead with the conclusion and then
explain. When you draft something for me, default to that style. When I ask
for analysis, give me your reasoning, not just a verdict. When you are
uncertain, say so.
A style sample prompt:
Here are two emails I wrote that I think sound like me. When you draft on
my behalf, match this voice.
[Paste email 1]
[Paste email 2]
Now draft a reply to the message below in that same voice.
A stakeholder profile (paste once, reuse forever):
Key stakeholders I work with regularly:
- Priya, VP Product. Cares about revenue impact and timing. Hates surprises.
Likes one-page summaries with a clear ask.
- Marco, Engineering Lead. Cares about scope clarity and dependencies.
Skeptical of vague requirements.
- Jen, Design Lead. Cares about user outcomes and consistency.
When you draft something for one of them, tune the framing accordingly.
TL;DR: Write a short personal profile, paste a writing sample, and save it. Every answer you get after that will be sharper than the answer the person next to you is getting.
The single biggest reason AI output is bad is not the model. It is missing context. You are holding the situation in your head and forgetting that the AI is not. The fix is to front-load the relevant information before the ask.
Use what I call the context sandwich. Four layers, in order:
- Role. Who should the AI be answering as? "Answer as a careful editor."
- Background. What does it need to know? The thread, the people, the constraints.
- Task. What do you want it to do, specifically?
- Format. How should the answer be shaped? Length, tone, structure.
The second rule: attach the actual thing. Do not describe an email, paste the email. Do not summarize the report, attach the report. Do not retype the contract clause, drop the PDF.
Replying to a long email thread:
You are helping me draft a reply. Below is a forwarded thread. The original
sender is a customer asking for a refund after using the product for four
months. Our policy allows refunds within 30 days. I want to decline the
refund but offer two months of free service as goodwill. Keep it warm,
under 150 words, and avoid sounding defensive.
[Paste full thread]
Summarizing a long report:
Attached is a 30-page market research report. I have ten minutes before a
meeting with my VP. Give me:
1. The three findings most relevant to our SMB billing roadmap
2. One slide-worthy chart description per finding
3. Two questions a skeptical exec might ask
Understanding a vendor contract clause:
Below is one clause from a vendor SaaS agreement. Explain in plain English
what it actually obligates us to do, what risks it creates if our usage
spikes, and what a reasonable redline would look like.
[Paste clause]
TL;DR: Role, background, task, format. And paste the real artifact. Do not make the AI guess at what you already have on your screen.
When you ask for a finished product up front, you get the AI's first guess at what you wanted. When you ask for a plan first, you get to course-correct before it has invested in a wrong direction. The output is almost always better and the iteration loop is shorter.
This works for anything longer than an email. Documents, decks, talking points, project briefs, decision memos, town hall scripts.
Add one line to your prompt:
Before you write anything, list your assumptions and outline your approach. Wait for my OK.
Then react to the outline. Add what is missing. Remove what is wrong. Then say go.
Planning a quarterly review doc:
I need to write my Q3 self-review. Before you draft anything, list the
sections you think it should have, what evidence each section needs from
me, and three angles I might be underweighting. Wait for my OK before
writing.
Planning a difficult conversation:
I need to give one of my direct reports critical feedback about missing
deadlines. Before you draft talking points, outline the structure of the
conversation, the order I should cover things in, and the two or three
reactions I should be ready for. Wait for my OK.
Planning an offsite agenda:
We have a two-day team offsite in six weeks. Twelve people, mixed PMs and
designers. Goal is alignment on next year's roadmap. Before you propose an
agenda, list the assumptions you are making, ask me three clarifying
questions, and sketch the shape of day one and day two at a high level.
Wait for my OK.
TL;DR: Ask for a plan, react to the plan, then ask for the output. Three turns. Better result every time.
A finished document is the wrong unit of work to ask for. Outlines, drafts, and revisions are three different jobs. When you ask for all three at once, the AI hedges and you get something mediocre. When you ask for them in sequence, each step is better than what you would write alone.
Three habits to combine:
- Outline first, draft second, revise third. Treat them as separate conversations.
- Match voice with examples. "Here are two things I wrote. Rewrite in that style."
- The three versions trick. When in doubt, ask for three versions with different angles, then keep the best parts of each.
Exec summary, three steps:
Step 1: Here are my raw notes on the Q3 billing migration. Give me an
outline for a one-page exec summary aimed at my VP. No prose yet.
[Paste notes]
Step 2 (after reviewing the outline): Good. Drop the "lessons learned"
section, expand "customer impact". Now write the draft.
Step 3: Tighten the opening. The first sentence should land the result. Cut
30 words anywhere you can.
Project status update, voice-matched:
Here are three weekly updates I have sent in the past. This is my voice.
[Paste 1] [Paste 2] [Paste 3]
Below are this week's bullets. Write this week's update in the same voice.
Keep it under 200 words.
Three-versions trick on a tricky paragraph:
Here is the opening paragraph of my proposal. Give me three rewrites:
- Version A: warmer, more collaborative
- Version B: more confident, leads with the recommendation
- Version C: shorter, halve the length
[Paste paragraph]
TL;DR: Outline, draft, revise as three separate moves. Feed it your voice with examples. When stuck, ask for three versions.
Most workplace friction is tone, not content. A message that says the right thing in the wrong tone creates more work, not less. AI is great at tone calibration if you tell it what you want.
Three moves to learn:
- Name the tone. Not "professional", which is meaningless. Try "warm but firm", "apologetic but not groveling", "confident, not defensive", "direct without being curt".
- Ask for three versions. Warm, neutral, firm. Or short, medium, long. Then mix and match.
- The trusted senior colleague reframe. "What would a trusted senior colleague who is good at this say here?" gets you out of your own head.
Declining a meeting without burning a bridge:
A peer from another team invited me to a recurring weekly meeting that I
genuinely cannot prioritize this quarter. Draft three versions of a reply:
- Warm: declines this quarter but leaves the door open
- Neutral: declines and proposes async updates instead
- Firm: declines and explains my Q3 commitments
All three should be under 100 words, no apologizing more than once.
Asking a stakeholder for a delayed input:
I am waiting on legal review of a contract that is blocking my launch.
Draft a Teams message to the legal lead. Tone: respectful of their load,
clear about my deadline, specific about what I need. Under 60 words.
Giving constructive feedback to a peer:
A peer PM presented a roadmap proposal in a leadership meeting. Two of the
priorities are based on assumptions I think are wrong. We are friendly. I
want to share my concerns privately first, before challenging in a meeting.
What would a trusted senior colleague say? Draft a message that opens a
conversation, does not ambush, and makes my reasoning clear.
TL;DR: Name the tone precisely. Ask for three versions. Borrow the voice of someone you trust.
The AI is more useful as a thinking partner than as an answer machine. It will not make your decision for you, and you should not want it to. What it will do is structure your thinking, surface what you are missing, and argue the other side.
Four prompts to keep in your back pocket:
- "Pros and cons. Then tell me what I am missing."
- "Argue the other side as well as you can. What is the strongest case against my preferred option?"
- "Steel-man both options. Where do they actually disagree?"
- State the problem, the candidate fix, and what good looks like. Specific asks beat vague ones.
Choosing between two vendors:
I am choosing between two vendors for our CRM migration. Here are my notes
on both.
[Paste notes]
Step 1: Pros and cons for each. Then tell me what I am underweighting.
Step 2: Argue the strongest case for picking Vendor B even though my gut
says Vendor A.
Step 3: What three questions should I get answered before deciding?
Structuring a re-org proposal:
I am drafting a proposal to merge two teams under one manager. Here is the
context.
[Paste context]
Walk me through this as my thinking partner. What are the three or four
real options? For each, what are the second-order effects I might be
missing? Do not recommend yet. Just help me see the shape of the decision.
Deciding whether to escalate:
A cross-team dependency is at risk of slipping our launch by three weeks.
I am debating whether to escalate to my VP now or give the partner team
one more week. Help me think this through. What are the costs of
escalating early? Of waiting? What would a calm, experienced PM do here?
TL;DR: Use it to structure thinking, not to decide. Force it to argue the other side. State the problem, fix, and success criteria when you want sharp output.
AI is a remarkable on-ramp into any topic you do not know yet. New business domain, unfamiliar legal concept, framework you keep hearing in meetings. The trick is to use it as a tutor, not as a reference.
Three moves:
- Ladder it. "Explain like I am five. Now like I am a smart generalist. Now like I am a domain expert." Three layers of depth on the same topic builds real understanding fast.
- Demand sources. "Show me where you got that. What are the strongest counter-positions?" Healthy skepticism, not paranoia.
- Pressure test what you learned. "Quiz me on what you just told me." Reverses the loop and exposes what you actually retained.
Getting up to speed on a new business domain:
I have a meeting tomorrow with a healthcare provider customer. I know
nothing about how hospital revenue cycle management works. Teach me in
three layers:
1. ELI5 version, three sentences
2. Smart generalist version, the key concepts and how they connect
3. What a CFO at a hospital actually worries about
End with five questions I should be ready to answer in the meeting.
Decoding a legal or compliance clause:
This is a clause from our updated data processing agreement. In plain
English: what does it require us to do, what does it forbid, what is the
worst case if we get it wrong? Then show me the two or three sentences in
the clause that carry the actual obligation, and where you might be
uncertain about interpretation.
[Paste clause]
Learning a new framework (OKRs as an example):
I keep hearing leaders talk about OKRs. Teach me in three layers (ELI5,
generalist, practitioner). Then walk me through one realistic example for
a product team. Then critique the framework: where do thoughtful people
say it falls down?
TL;DR: Three layers of depth. Ask for sources. Have it quiz you back. Treat it like a patient tutor.
A few traps everyone hits. Knowing them in advance saves embarrassment.
Hallucinations. AI can invent names, citations, statistics, even quotes from real people. The more confident the wording, the more you should check. If the AI gives you a number, a name, a date, or a legal-sounding sentence, verify it before it leaves your hands.
Stale knowledge. Models have a training cutoff. Recent events, recent prices, recent policies, recent product releases may be wrong or missing. When recency matters, say "as of today" and use a tool with web access, or verify independently.
Privacy. Do not paste anything into a public AI tool that you would not paste into a public document. Customer data, salary information, unannounced financials, anything covered by NDA. If your company provides an enterprise AI tool, use that for sensitive work.
Over-reliance on a single answer. The first answer is one possible answer. Ask for alternatives. Ask what would change the recommendation. Ask where the reasoning is weakest.
Sounding like AI in your own writing. This one matters more than people realize. If you publish AI-toned prose under your name, people stop reading you the same way. Specific tells to scrub: em dashes, "delve", "tapestry", "it's worth noting", "furthermore", overly balanced sentences, every paragraph the same length. Read your draft aloud. If it sounds like a press release, rewrite.
TL;DR: Verify what matters. Mind the cutoff. Protect private data. Ask for alternatives. Edit out the AI tone.
- "Before you answer, list your assumptions and outline your approach. Wait for my OK."
- "Give me three versions: warm, neutral, firm."
- "Argue the other side as well as you can."
- "What am I missing?"
- "Rewrite in this voice." [paste sample]
- "Explain like I am five. Then like a generalist. Then like an expert."
- "Show me where you got that."
- "Tighten this. Cut 30% without losing the point."
- "What would a trusted senior [role] say here?"
- "Quiz me on what you just told me."
- Too generic? You did not give enough context. Paste the actual thing.
- Too long? Ask for a word count or "tighten by 40%".
- Wrong tone? Name the tone precisely. Do not say "professional".
- Misses the point? Re-state the goal in one sentence and try again.
- Sounds like AI? Paste a sample of your writing and say "match this".
- Confidently wrong? Ask "what are you uncertain about?" and verify.
- Stuck in a loop? Start a new chat. Bring only what is needed.
Each major assistant has its own way of doing the same things. Quick map:
ChatGPT
- Custom Instructions (Settings, Personalization): your standing profile. Two boxes. First box describes you. Second box describes how you want responses formatted.
- Memory: lets ChatGPT remember facts across chats. Review it periodically. Clean out anything stale or wrong.
- Projects: group related chats with shared instructions and files. Great for recurring work like "Weekly Status" or "QBR Prep".
Claude
- Projects: each Project has its own system prompt and knowledge files. This is where you build a real personal workspace.
- Styles: save reusable voice instructions and pick the right one per task. Useful if you write in multiple registers (e.g., exec summaries vs customer emails).
- Claude is unusually good at long documents and careful reasoning. Lean on it when the input is dense.
Microsoft Copilot
- Pages: persistent canvases you can co-edit with the AI. Good for drafts you keep coming back to.
- Notebooks: pin context (files, links, notes) the AI uses for every query in that notebook.
- Inside Microsoft 365, Copilot already sees your email, calendar, and files. Lean into that. "Summarize the threads I have with Priya this week."
Gemini
- Saved Info: the equivalent of memory. Tell it your role, preferences, and writing style once.
- Gems: custom personas with their own instructions. Build a "Meeting Prep Gem" or "Edit My Email Gem" and stop re-explaining yourself.
- Gemini ties tightly into Google Workspace. "Pull the last three emails from this person and draft a reply."
None of this is hard. The whole guide fits on a napkin: teach it who you are, give it real context, ask it to plan, iterate, verify. The reason most people do not get value out of AI assistants is not that they lack the skill. It is that nobody told them these five habits exist.
Pick one technique this week. Use it three times. Then pick another. Within a month, you will look back at how you used to use these tools and wonder why you ever typed a question and accepted the first answer.
If this was useful, share it with someone who would benefit. That is the only ask.
Clint