Most "best ChatGPT prompts" lists are a wall of text you skim once and never use again. This one is built differently: every prompt below is a real, maintained page on our free prompt library, with the full copy-paste text, when to use it, and which model handles it best. Nothing is gated — you can use all of them without an account.

They are grouped by the four jobs that eat most professional weeks: email, meetings, planning, and research. Pick the three that match what you do most, save them somewhere you can trigger by keystroke, and stop there. Three prompts you actually use beat thirty you bookmarked.

Email and communication

Email is the highest-frequency, lowest-creativity task most professionals have — which makes it the best possible place to start. The prompts that work here are the ones that force a structure: what happened, what you need, by when, and one clear next step.

Professional Email — quick, clear emails that actually get a reply. The default when you are staring at a blank compose window.
Polite Decline — say no without burning the relationship. Most people write this badly because they over-explain; the prompt keeps it short and warm.
Follow-up Nudge — re-open a stalled thread without nagging. Assumes good intent, gives one clear next step.
Difficult Feedback — deliver hard feedback with care, specific rather than vague, without softening it into meaninglessness.
Apology Email — own the mistake, state the fix, avoid the defensive paragraph that makes it worse.
Negotiation Reply — anchor, concede once, end with a question.

Time saved, realistically: two to four minutes per email on the ones you would otherwise stare at. For most people that is 30 to 60 minutes a week, concentrated in exactly the emails they were dreading.

Meetings and notes

The meeting is rarely the problem — the unwritten recap is. Notes that never become action items are the single most common leak in professional workflows, and it is entirely automatable.

Meeting Recap — turn raw notes or a transcript into decisions, owners, deadlines, and a parking lot.
1:1 Agenda — build an agenda from last meeting's notes plus three questions worth asking.
Status Update — wins, blockers, asks, in five sentences your manager will actually read.
Async Standup — replace the daily standup with a written update that takes two minutes to write and ten seconds to read.
Meeting Killer — decide whether a recurring meeting should exist at all, and draft the message that cancels it.

The habit that makes this work: run the recap prompt in the two minutes after the call ends, while you still remember the subtext. A recap written the next morning is always worse.

Planning and prioritisation

These are the prompts that pay off weekly rather than daily. Their value is not speed — it is that they force you to write down what you are not doing, which is the part everyone skips.

Weekly Review — what moved, what stalled, what to drop before next week starts.
Weekly Reset — the operational version: inbox, calendar, and task list back to a clean state.
Time Audit — find where the hours actually went, which is almost never where you assumed.
Focus Block Design — structure deep work around your real calendar rather than an idealised one.
Decision Framework — score options against criteria you name up front, before you have a favourite.
Pre-mortem — assume the plan failed, work backwards to why, then prevent the top two causes.
Delegation Map — separate what only you can do from what you have simply never handed over.

Research and learning

Research prompts are where AI is most useful and most dangerous. The useful part is structure and speed. The dangerous part is that every model will fill a gap with something plausible rather than admitting it does not know. Every prompt below includes an instruction to flag uncertainty — keep that line when you customise them.

Topic Brief — get oriented in an unfamiliar subject fast, with the open questions listed.
Competitor Scan — a structured comparison instead of twelve open browser tabs.
Literature Summary — condense long sources while keeping the caveats intact.
Customer Interview Guide — non-leading questions that produce usable answers.
First Principles — strip an assumption back to what is actually known.
5 Whys — get past the first plausible cause to the real one.

Always verify. Any number, date, quote, or citation that will appear in your work needs to be checked against a primary source. Treat the model as a fast research assistant whose work you review, not as the source.

How to customise any prompt: the CARE idea

Copying a prompt gets you 70% of the value. The last 30% comes from adapting it to your situation, and the reason most people's edits make output worse is that they add detail in the wrong places. The shorthand we teach is CARE — Context, Ask, Refine, Evaluate. It is deliberately simple, and it is the same four moves whether you use ChatGPT, Claude, or Gemini.

C — Context

Tell the model who you are, who the output is for, and what already happened. "I am a freelance UX designer writing to a client who missed two review deadlines" produces a fundamentally different email than "write a follow-up." Context is the single highest-leverage edit, and it is the one most people skip.

A — Ask

State exactly what you want back, including the shape. Length, format, tone, and what to leave out. "120 words, no bullet points, warm but direct, do not apologise" is an ask. "Make it good" is not. Being specific about the format also makes the output easier to paste somewhere without reformatting.

R — Refine

Treat the first output as a draft to react to, not a result to accept. The most productive follow-up messages are short and surgical: "second paragraph is too formal," "cut 30%," "keep the structure, change the opening." Two rounds of refinement beat one enormous opening prompt, because you learn what the model got wrong before you invest more instructions.

E — Evaluate

Before you send or publish, check the things a model cannot check for you: is every fact true, does it sound like you, and would you be comfortable if the recipient knew how it was drafted? This step takes thirty seconds and is the entire difference between AI that helps your reputation and AI that quietly damages it.

That is the concept in outline. The full CARE framework — with the worked examples, the failure modes for each step, and the variants for writing, analysis, and research — is one of the chapters inside The AI Productivity Playbook, alongside the Daily AI Loop and the 30-day adoption challenge.

Make them stick: the two-minute setup

The reason prompt lists do not change anyone's week is friction. If using a prompt requires finding a blog post, the model wins by default and you type something lazy instead. Fix that once:

Choose three. The three tasks you did most last week. Not the most interesting ones — the most frequent.
Save them where you type. A snippet manager (Raycast, Alfred, espanso), a Notion page pinned to your sidebar, or saved instructions in the assistant itself.
Give each a trigger. Something you will remember: ;email, ;recap, ;week.
Edit them after two weeks. You will have learned which line the model keeps ignoring and which context it always needs. That edited version is worth more than anything you copied.

Which model to run these on

All of the prompts above work on ChatGPT, Claude, and Gemini, but they are not equally suited to each. As a rough guide: email and meeting prompts run fine anywhere and are worth putting on whichever assistant is already open. Planning prompts benefit from a model that will argue with you rather than agree — Claude is the usual pick. Research prompts belong on whichever model can cite live sources, which in practice means Gemini or a search-grounded mode.

The one thing that genuinely changes results across models is context length. If your prompt involves pasting a long transcript, a full document, or a week of notes, use the model with the largest window rather than chopping the source into pieces — summaries of summaries lose exactly the specifics you needed. Our ChatGPT vs Claude vs Gemini comparison breaks down the published limits.

Three mistakes that waste the time you just saved

Accepting the first output. The draft is a starting point. One round of refinement takes fifteen seconds and roughly doubles the quality; skipping it is why people conclude AI writing is bland.
Pasting confidential material without checking. Client names, contracts, personal data, and internal financials belong only in a tool your organisation has approved. Redact before you paste, and check your workspace's data settings.
Sending unread. A model will state a date, a number, or a commitment you never made, in a completely confident tone. Thirty seconds of reading before you hit send is the entire risk-management strategy.

What to expect, honestly

A good prompt does not produce finished work. It produces a solid draft in ten seconds instead of a blank page in ten minutes, and it removes the activation energy that makes you postpone the task in the first place. Across email, meeting recaps, weekly planning, and research, most people who set up three saved prompts report getting back somewhere between two and five hours a week — and the bigger win is usually that the dreaded tasks stop being dreaded.

Browse the full free library — 150+ prompts across 8 categories, all with the complete text and no signup. When you want the worked examples, the Notion templates, and the 30-day plan that turns individual prompts into a system, that is The AI Productivity Playbook.

ChatGPT Prompts for Productivity: The 2026 Working Shortlist