30 AI prompts that save 10 hours a week (with recommended models)

The short version.

• 30 prompts, organised into 6 use cases.
• Each prompt names the recommended model and the time it saves.
• All have been tested in production. No filler. No "prompt engineering as a career" nonsense.
• Replace the [BRACKETED PLACEHOLDERS] with your specific input.

For an expanded, searchable, copy-to-clipboard version, use our Prompt Library tool.

Why use a prompt library

• Generic prompts produce generic outputs.
• Specific prompts that name a role, output format, and constraints work much better.
• Reusing tested prompts saves the friction of inventing each one.
• The single biggest productivity gap in 2026 is between users who reuse prompts and users who write each from scratch.

Time savings at a glance

Category	# of prompts	Estimated weekly time saved
Email and communication	5	2 hours
Meetings	4	1.5 hours
Writing and editing	6	2.5 hours
Research and analysis	5	2 hours
Code and technical	5	1.5 hours
Decisions and planning	5	0.5 hour
Total	30	~10 hours/week

Assumes a knowledge worker who uses 4 to 6 of these prompts daily. Less use = less saved.

Email and communication (5 prompts)

1. Harsh email rewrite

Best model: Claude 4.6 Sonnet. Time saved: 15 to 30 min per hard email.

Below is the email I need to send. Make it land better.

Specifically:
1. Cut every word that isn't earning its place.
2. Put the most important point in the first two sentences.
3. Sound human, not corporate.
4. Flag anything that could be misread before I send.

Context: [WHO / RELATIONSHIP / GOAL]

My draft:
[PASTE]

2. Cold outreach generator

Best model: GPT-5. Time saved: 20 to 40 min per round of outreach.

Write 3 versions of a cold outreach email from me to [TARGET], offering [OFFER].

Each version should:
- Be under 90 words
- Reference one specific detail about them (I'll fill in)
- Have a clear, low-friction ask
- Avoid every cold-email cliche

Format: subject line + body, separated.

3. Reply draft from context

Best model: Claude 4.6 Sonnet. Time saved: 10 min per important reply.

I'm replying to this email:
[PASTE EMAIL]

I want to:
[YOUR INTENT]

Constraints: [TONE / DEADLINE / WHAT TO AVOID]

Draft 3 versions of my reply. After each, note in [brackets] the trade-off it makes.

4. Difficult conversation rehearsal

Best model: Claude 4.7 Opus. Time saved: 30 to 60 min of overthinking.

I have to have a hard conversation with [PERSON] about [TOPIC].

Help me prepare by:
1. Drafting 3 opening lines I could use, with their pros and cons.
2. Predicting their 2 most likely responses.
3. Drafting how I'd respond to each.
4. Flagging the trap I'm most likely to fall into.

5. Apology that doesn't over-apologise

Best model: Claude 4.6 Sonnet. Time saved: 15 min per situation.

I need to apologise for [SITUATION]. The person is [RELATIONSHIP].

Help me write an apology that:
- Owns the specific thing I did
- Doesn't grovel or over-apologise
- Says clearly what I'll do differently
- Is under 100 words

Meetings (4 prompts)

6. Meeting prep brief

Best model: Claude 4.7 Opus. Time saved: 30 to 45 min per meeting.

I have a meeting with [PERSON] tomorrow about [TOPIC].

Context:
- [POINT 1]
- [POINT 2]
- [POINT 3]

I want from the meeting: [OUTCOME]

Produce a 1-page brief covering:
1. The 3 points I most need to make
2. The 3 questions I should ask them
3. The 2 objections most likely to come up + my responses
4. The single trap I should avoid

7. Meeting summary from notes

Best model: GPT-5 or Claude 4.6 Sonnet. Time saved: 20 min per meeting.

Below are my raw meeting notes. Produce:

1. A 100-word executive summary
2. Action items as a checklist with owners and dates
3. Decisions made (vs items still open)
4. Questions to follow up on

My notes:
[PASTE]

8. Stand-up update template

Best model: GPT-5 Mini (fast and cheap). Time saved: 5 min daily = ~25 min/week.

Based on this list of what I did since last stand-up, write a 3-sentence update covering:

1. What I shipped
2. What I'm working on now
3. Anything blocking me

Be specific. No filler. Don't include things that don't matter to the team.

My work this period:
[PASTE]

9. Tough 1:1 prep

Best model: Claude 4.7 Opus. Time saved: 25 min per hard 1:1.

I have a 1:1 with [REPORT / MANAGER]. I need to bring up [TOPIC].

The history: [BRIEF CONTEXT]

Help me prepare by:
1. Drafting the opening sentence
2. Naming the 3 outcomes I want, ranked
3. The signal I should look for to know they're hearing me vs not
4. The follow-up I should commit to in the meeting

Writing and editing (6 prompts)

10. Editor's brutal pass

Best model: Claude 4.7 Opus. Time saved: 30 to 60 min per article.

You are a senior magazine editor. Edit the text below.

Specifically:
1. Cut every sentence that doesn't earn its place
2. Tighten weak verbs and remove adverbs
3. Flag any claim that needs a source
4. Suggest one stronger lead and one stronger ending

Return the edited version inline, reasoning for each cut in [brackets]. Don't be polite.

Text:
[PASTE]

11. 15 headline options

Best model: GPT-5 or Claude 4.6 Sonnet. Time saved: 30 min per piece.

Generate 15 headlines for the article below. Each should:
- Be under 70 characters
- Include a number, name, or counter-intuitive claim where possible
- Avoid "Top 10" / "Best of" / clickbait patterns

Rank them from most likely to be clicked (#1) to least (#15). Explain the top 3.

Article:
[PASTE]

12. Tone match rewrite

Best model: Claude 4.7 Opus. Time saved: 1 to 2 hours per voice-match task.

I'll give you 3 samples of writing in a specific voice. Analyse: sentence rhythm, vocabulary, characteristic moves. Then rewrite the target text in that voice without changing meaning.

Sample 1: [PASTE]
Sample 2: [PASTE]
Sample 3: [PASTE]

Target text:
[PASTE]

Return only the rewrite. Don't explain.

13. Long-form outline

Best model: Claude 4.6 Sonnet. Time saved: 45 min per piece.

I want to write a 2,000-word piece on [TOPIC].

Produce an outline that includes:
1. A working title and dek (~25 words)
2. The single argument the piece makes
3. 4-6 section headers with one-sentence summaries
4. 2-3 specific sources, examples, or data points
5. The intended ending

Audience: [BUILDERS / EXECUTIVES / GENERAL]
Voice: [ANALYTICAL / OPINIONATED / NEUTRAL]

14. Social post from article

Best model: GPT-5. Time saved: 20 min per share round.

Below is my article. Produce:
1. A LinkedIn long-form post (~300 words) that quotes the article's strongest claim
2. A Twitter/X thread (5-7 tweets) summarising the key insight
3. A short LinkedIn caption (~50 words) for sharing the link

Voice: confident but not salesy. No "Excited to share..." openers.

Article:
[PASTE]

15. Spec doc to email

Best model: Claude 4.6 Sonnet. Time saved: 30 min per spec-to-comms task.

Below is a technical spec. Distil it into a 200-word email to non-technical stakeholders.

Cover:
- What changes
- Why it matters to them
- What they need to do (if anything)
- When it happens

Avoid jargon; explain trade-offs in plain English.

Spec:
[PASTE]

Research and analysis (5 prompts)

16. Multi-source synthesis

Best model: Claude 4.7 Opus or Gemini 3 Pro. Time saved: 1 to 2 hours per research synthesis.

I'm researching [TOPIC]. Below are 3 sources. Synthesise by:

1. Identifying where all 3 agree
2. Identifying where they disagree, on what
3. Noting any source that makes claims without evidence
4. Producing a 5-bullet summary of the actual state of knowledge

Source 1: [PASTE]
Source 2: [PASTE]
Source 3: [PASTE]

17. Steelman the opposing view

Best model: Claude 4.7 Opus. Time saved: 45 min of internal debate.

I hold this position:
[YOUR POSITION]

Argue against it as steel-manned as possible:

1. Construct the strongest counter-argument
2. Identify the assumptions in my position that are most vulnerable
3. List the 3 best pieces of evidence that someone arguing against me would cite
4. Note where my position is on weakest ground

Don't hedge. Find the weakest link before someone else does.

18. Paper summariser

Best model: Claude 4.7 Opus. Time saved: 30 to 60 min per paper.

Summarise the paper below for someone technically literate but not a specialist.

Structure:
1. The core claim, in one sentence
2. The method, in 2-3 sentences
3. The result, in 2-3 sentences
4. Why it matters, in 2 sentences
5. The biggest limitation
6. Three open questions this paper raises

Paper:
[PASTE OR LINK]

19. Compare two products honestly

Best model: Claude 4.7 Opus with web search. Time saved: 1 to 2 hours per buying decision.

Compare [PRODUCT A] and [PRODUCT B] honestly. For each:

1. The use case it's actually best for
2. The use case it's actually worst for
3. What it costs (real numbers)
4. The most common reason users complain
5. The kind of buyer who picks it for the wrong reason

Then recommend one based on: [MY CONSTRAINTS]

Don't summarise marketing pages. Be specific about trade-offs.

20. Market sizing first principles

Best model: GPT-5. Time saved: 1 hour per estimate.

Estimate the market size for [PRODUCT/CATEGORY] in [REGION] using first principles.

Show your reasoning step by step:
1. Define what counts as the market
2. Identify the population / unit base
3. Estimate adoption %
4. Estimate revenue per unit
5. Show three scenarios: conservative, base, optimistic

Cite assumptions explicitly. Flag where uncertainty is highest.

Code and technical (5 prompts)

21. Refactor with constraints

Best model: Claude 4.7 Opus. Time saved: 1 hour per refactor.

Refactor the code below. Constraints:
1. Don't introduce new dependencies
2. Match the existing style
3. Preserve the public API exactly
4. Explain what changed in 2-3 bullets at the top
5. If there's a bug you noticed, fix it but flag it separately

Code:
[PASTE]

22. Three-perspective code review

Best model: Claude 4.7 Opus. Time saved: 45 min per PR.

Review the code below from 3 perspectives:

1. A staff engineer focused on correctness and edge cases
2. A senior engineer focused on readability
3. A pragmatic principal who only mentions things worth fighting for

Separate blocks. End with one ranked list of fixes by priority.

Code:
[PASTE]

23. Bug isolation

Best model: Claude 4.7 Opus or GPT-5 with reasoning. Time saved: 1 to 2 hours of debugging.

I'm seeing this error: [PASTE]

In this code: [PASTE]

Expected: [WHAT YOU EXPECTED]
Actual: [WHAT HAPPENS]

Tell me:
1. The most likely root cause
2. The second most likely cause
3. The minimal fix
4. Any other places in the code with the same issue

24. Test suite from spec

Best model: GPT-5. Time saved: 45 min per function under test.

Generate a complete test suite for the function below using [JEST / VITEST / PYTEST].

Cover:
1. Happy path
2. Edge cases (empty, null, undefined, max, min)
3. Error paths
4. Concurrency / ordering if applicable

Each test: descriptive name + one-line comment.

Function:
[PASTE]
Spec:
[PASTE]

25. SQL from natural language

Best model: Claude 4.6 Sonnet. Time saved: 20 min per analytical query.

Write a SQL query for [DATABASE: postgres/mysql/bigquery] that answers:

[YOUR QUESTION IN ENGLISH]

Schema:
[PASTE TABLE DEFINITIONS]

Optimise for readability. Add a comment explaining any non-obvious join. Flag any assumption you made about the data.

Decisions and planning (5 prompts)

26. Strategic decision framework

Best model: Claude 4.7 Opus. Time saved: 1 to 2 hours of agonising.

I'm trying to decide [THE DECISION]. The options:

A: [OPTION A]
B: [OPTION B]
C: [OPTION C]

Structure this decision:

1. The actual question being asked (re-frame if mine is wrong)
2. The 4-5 criteria that should drive it, weighted
3. How each option scores on each criterion (1-5)
4. The biggest unknown that, if resolved, would change the answer
5. Your recommendation and confidence level

27. Pre-mortem on a project

Best model: Claude 4.7 Opus. Time saved: 30 to 45 min of risk thinking.

We're about to start [PROJECT]. Imagine it's 6 months from now and the project failed.

Walk me through:
1. The 3 most likely reasons it failed
2. The signal we should have noticed in the first month
3. The single decision we made early that doomed it
4. What we should do differently starting now

Be specific. Don't give generic risk-management advice.

28. 90-day plan

Best model: Claude 4.6 Sonnet. Time saved: 45 min of planning.

I just took on [NEW ROLE / NEW PROJECT].

Build me a 90-day plan with:
1. The single most important outcome by day 90
2. 3 metrics to track weekly
3. The 4-5 specific projects, with priorities and target completion dates
4. The 2 stakeholders I need to actively manage
5. The thing I should NOT do in the first 30 days

29. Hiring decision

Best model: Claude 4.7 Opus. Time saved: 1 hour per candidate decision.

I'm deciding whether to hire [CANDIDATE] for [ROLE].

What I observed:
[YOUR NOTES]

The role's most important requirement:
[ONE THING]

Help me:
1. Identify what's strongest and what's weakest
2. Predict where they'd struggle in the first 90 days
3. Name the question I should ask in a follow-up to resolve my doubt
4. Recommend hire / no-hire and confidence level

30. Roadmap from chaos

Best model: Claude 4.7 Opus. Time saved: 2 to 3 hours of arguing.

Below is a list of things people want me to build. Help me turn it into a roadmap.

Constraints:
- 1 person on it (me)
- 8 weeks
- Highest-impact-first ordering

For each item:
1. Impact (1-5)
2. Effort (days)
3. Whether to do, defer, or kill

End with the actual 8-week sequence.

The list:
[PASTE]

How to actually use these

• Pick 3 to start. Don't try to memorise 30. Pick the 3 that match your top 3 weekly time-wasters.
• Save them in your AI chat's starred conversations or in a Notion page. Friction kills usage.
• Modify the brackets, not the prompt structure. The structure is what makes them work.
• Pick the recommended model. Or test all four — Claude vs GPT vs Gemini will produce noticeably different outputs.

For a more polished, searchable, copy-to-clipboard version, use the Prompt Library tool on this site.

Frequently asked questions

Will these still work in 6 months?

Yes. The structure (role, format, constraints, negative instructions) is model-agnostic. Even when GPT-6 ships, these prompts will still work.

Why does the recommended model matter?

Models have different defaults. Claude follows negative instructions ("don't soften this") more reliably than GPT. GPT structures output more cleanly when given templates. Picking the right one for each task lifts results meaningfully.

Can I use these for free?

Yes. Every prompt above works on the free tier of at least one of the recommended models. You may hit rate limits faster on free tiers.

What if I want more?

Our Prompt Library tool has the full set with copy-to-clipboard, filters, and category browsing.

Related on Meridian48

• Prompt Library tool
• Which AI should I use?
• AI Pricing Tracker
• Every AI acronym you will see in 2026, explained