100 Best Claude Prompts of 2026

The best Claude prompts of 2026 aren’t clever one-liners — they’re structured requests that give Claude a role, fenced context, and a reason for the task. I’ve curated the 100 below from heavy daily use into ten sections of ten: each opens with the strongest prompt in full, copy-paste ready, followed by nine proven variants.

TL;DR:

• 100 curated prompts across 10 sections: productivity, coding, writing, business, marketing, research, learning, documents, career, and personal decisions.

• Every prompt uses at least one Claude-specific technique — XML tags, role framing, thinking triggers, or the “explain why” rule.

• Adapt the bracketed placeholders; the structure is the part that matters.

What Made a Prompt “Best” in 2026

Three filters decided this list. The prompt had to survive repeated use — no one-time party tricks. It had to exploit something Claude is specifically good at: XML-tagged context, roles with real depth, or reasoning triggers. And it had to fail safely, with an escape hatch (“say ‘unknown’ rather than guessing”) so a bad day produces an honest answer instead of a confident wrong one. The theory lives in our Claude prompts library and Anthropic’s prompt engineering docs; this page is the practice.

Section	Prompts	Core technique
Everyday productivity	1–10	Caps and formats
Coding & development	11–20	Grounding in real code
Writing & editing	21–30	Voice samples, capped feedback
Business & strategy	31–40	Adversarial roles
Marketing & sales	41–50	Audience + constraint
Research & analysis	51–60	Escape hatches
Learning	61–70	Interactive formats
Data & documents	71–80	XML-fenced sources
Career	81–90	Specific evaluator roles
Personal decisions	91–100	Thinking triggers

The Skeleton Behind Most of These Prompts

One thing before the list: nearly every serious prompt below is a variation on the same five-part skeleton. Learn it once and you can repair, shrink, or extend any prompt on this page:

You are a [ROLE with specific seniority and domain].

<context>
[Why you need this, who it's for, what happens with the output]
</context>

<task>
[The specific job, stated as an action verb]
</task>

<input>
[The material to work on — paste it here, however long]
</input>

<requirements>
- [Format, length, tone constraints — with the reason for each]
- [What to do if information is missing: ask, or state assumptions]
</requirements>

Begin your response with the deliverable itself — no preamble.

Each part earns its place: the role line shifts Claude’s vocabulary and quality bar in one sentence, the context block is the “explain why” rule in action, the tags keep long pasted material from bleeding into your instructions, and the closing line kills the throat-clearing. When a prompt from this list underperforms for you, check which of the five parts your adaptation dropped — that’s the fix nine times out of ten.

Everyday Productivity (1–10)

1. Triage my task list below. For each task: do, delegate, defer,
or delete — with one line of reasoning. Context: my actual goal
this week is [goal], so judge every task against that, not
against general productivity advice.

<tasks>[paste list]</tasks>

Anchoring the triage to one stated goal is the “give Claude the why” rule in miniature.

2. Turn these meeting notes into decisions made, owners, deadlines, and open questions — nothing else.
3. Draft my out-of-office, three versions: internal, client-facing, and one for the colleague covering for me.
4. Here’s my calendar for next week. Find the conflicts, the meetings I should decline, and two blocks for deep work.
5. Rewrite this Slack message so it can’t be misread as passive-aggressive. Keep it under 50 words.
6. Summarize this 40-message email thread: what’s decided, what’s contested, and what I’m actually being asked to do.
7. Build a checklist for [recurring task] so complete that I could hand it to a new hire tomorrow.
8. I keep procrastinating on [task]. Break it into steps so small the first one takes under five minutes.
9. Write the agenda for a 30-minute meeting about [topic] — timed segments, one decision per segment, and what to send attendees beforehand.
10. Review my weekly routine below and find the two changes with the highest return for the least disruption.

Coding and Development (11–20)

11. You are a staff engineer. Review the code below for correctness
bugs and unhandled edge cases only — skip style. Quote the exact
line for each finding, propose the fix, and if nothing is seriously
wrong, say so. Do not invent problems.

<code>[paste code]</code>
<context>This runs in [where]; a bug costs [what].</context>

The single most reused prompt in my file — the stakes line changes review strictness to match reality. Forty-nine more live in the coding prompts list.

12. Debug this: symptom, expected behavior, code, and what I’ve ruled out are tagged below. Give three ranked hypotheses and the cheapest test for each.
13. Refactor this function for readability with byte-identical behavior. List any risky change separately as “proposed, not applied.”
14. Write unit tests for this function: boundaries, error paths, empty inputs. Skip happy-path tests that restate the implementation.
15. Explain this legacy file in three passes: purpose, flow of one realistic input, and the two most fragile parts.
16. Here’s my schema in a tag. Write the query for [question], using only columns that exist, and name the index it needs.
17. Act as a principal engineer and argue against my architecture below — failure modes, 10x-load breaking points, and the simplification you’d make.
18. Given this profiler output as ground truth, fix only the top bottleneck. Refuse to micro-optimize anything the profile shows is cheap.
19. Before writing code for this feature, restate the requirements, list assumptions, and ask up to three clarifying questions.
20. Diagnose this failing CI log: root cause vs cascade failures, and the one line to fix first.

Writing and Editing (21–30)

21. You are a developmental editor. Give me only the three
highest-impact structural revisions for the draft below — what's
weak, why it costs readers, and the fix demonstrated on my own
sentences. Then stop.

<draft>[paste]</draft>
<goal>[what the piece must do, for whom]</goal>

Capping feedback at three forces prioritization. The writing prompts list expands this into 40 prompts across drafting, fiction, and email.

22. Extract a style guide from the five writing samples below — rhythm, register, openings, closings — so I can reuse it as a system prompt.
23. Draft this article from my tagged notes only; where a fact is missing, write [NEEDS SOURCE] instead of inventing one.
24. Cut 25% of the words from this passage without losing a fact or softening a claim.
25. Generate 10 headlines, then kill 7 as a jaded editor, explaining each kill. Every survivor must be true to the piece.
26. Find every AI-sounding stock phrase in my draft and replace each with something a human would say.
27. Write this scene so the subtext I’ve tagged is never stated — only shown through action and dialogue.
28. Turn this transcript into a blog post: cut filler, keep the speaker’s phrasing for key ideas.
29. Write a 120-word email where the ask appears in the first two sentences and again as the final line.
30. Summarize this piece at three lengths — sentence, paragraph, page — preserving the author’s qualifiers exactly.

Business and Strategy (31–40)

31. Act as a skeptical CFO. I'll pitch you an internal project;
poke holes in the numbers, timeline, and assumptions. Think through
second-order effects before responding. End with the three
questions you'd need answered before approving.

<pitch>[paste plan]</pitch>

Adversarial role framing is the highest-value business technique: agreeableness disappears once Claude’s assigned job is finding problems.

32. Run a premortem: it’s one year on and this project failed. Write the honest postmortem, most likely causes first.
33. Turn this rambling doc into a one-page brief an exec reads in two minutes: verdict, three supporting points, one risk, one ask.
34. Draft the agenda and pre-read for a decision meeting on [topic] — frame the two real options and what evidence would settle it.
35. Compare these two vendor proposals against MY criteria below, then commit to a recommendation with a confidence level.
36. Write the job description for [role] that repels the wrong candidates as effectively as it attracts the right ones.
37. Model this pricing change: list every customer segment, how each reacts, and where the revenue risk concentrates.
38. Draft difficult-conversation talking points for [situation]: what to say, what not to say, and likely responses with replies.
39. Review this contract as the counterparty’s lawyer: which clauses would you exploit? (Then I’ll fix them.)
40. Build a 30/60/90 plan for my new role from the tagged context, with one measurable outcome per phase.

Marketing and Sales (41–50)

41. You are a direct-response copywriter. Write 5 subject lines
for the email below — under 45 characters each, because mobile
truncates. No clickbait: our unsubscribe rate spikes when subject
lines overpromise.

<email>[paste body]</email>

42. Rewrite this product page from features to outcomes: for each feature, what the customer can now do that they couldn’t.
43. Draft a cold email to [persona] that earns a reply in under 90 words: their problem in their words, one proof point, one small ask.
44. Write the landing page for [product] answering, in order: what is it, is it for me, does it work, what’s the catch, what now.
45. Generate 10 ad variants for [offer], grouped by the emotion each targets — relief, ambition, fear of missing out — two per emotion.
46. Critique my funnel copy stage by stage: where does the promise shift, and where would a skeptical buyer bail?
47. Turn these three customer interview transcripts into the exact phrases buyers use for the problem — verbatim, with counts.
48. Write objection-handling notes for [product]: the five real objections, and answers that concede what’s true before countering.
49. Draft a launch announcement in three sizes — tweet, email, blog post — same core claim, native to each format.
50. Position [product] against [competitor] without naming them: what we’re for, who we’re for, what we refuse to do.

Research and Analysis (51–60)

51. Build a synthesis matrix from the papers tagged below: research
question, method, sample, key finding, stated limitations. Where a
field isn't stated, write "not reported" — never infer it. Close
with one paragraph on where the papers disagree.

<paper_1>[...]</paper_1>
<paper_2>[...]</paper_2>

“Not reported, never infer” is the strongest anti-hallucination clause I know, and research is where it matters most.

52. Steelman both sides of [question] with the best evidence for each, then tell me what evidence would change your mind.
53. Read this report and separate its findings from its interpretations. Which interpretations outrun the data?
54. Here’s my survey data as CSV. Describe the three strongest patterns, each with the effect size and an alternative explanation.
55. Generate the five research questions this literature review below leaves open, ranked by feasibility for a solo researcher.
56. Critique my methodology section as peer reviewer 2 — the harsh one. Every objection must be specific and answerable.
57. Translate this academic paper for a practitioner: what it found, how confident to be, and what to do differently Monday.
58. Compare how these three sources define [term]. Where the definitions conflict, show what each conflict changes downstream.
59. Audit my argument below for logical gaps: unsupported premises, causal leaps, and the strongest counterexample to each.
60. Build an annotated bibliography from the tagged sources — two sentences of contribution and one caveat per source.

Learning and Skill-Building (61–70)

61. Teach me [topic] Socratically. Ask me one question at a time,
wait for my answer, correct my misconceptions with a concrete
example, and only move on when I've demonstrated the idea back
to you. Start by finding out what I already know.

62. Explain [concept] three times: to a novice, to a peer, and to an expert who’ll challenge the edge cases.
63. Quiz me on [material] with escalating difficulty. Track what I miss and re-test my weak spots at the end.
64. Build a 30-day learning plan for [skill] at 45 minutes a day, with a self-test at the end of each week.
65. I’ll explain [topic] back to you in my own words. Interrupt only where I’m wrong, and tell me the misconception behind each error.
66. Give me deliberate-practice drills for [skill] — each isolating one sub-skill, with a way to score myself.
67. What are the five most common beginner misconceptions about [topic], and what’s true instead?
68. Turn my messy course notes below into spaced-repetition flashcards — question on one side, minimal answer on the other.
69. Roleplay a [scenario, e.g. salary negotiation] as the other side. Stay in character; debrief only when I say “debrief.”
70. Brief me in 20 minutes on [topic]: the three core concepts, the two live debates, the one smart question to ask.

Data, Documents, and Spreadsheets (71–80)

71. The contract below is tagged. Answer only from its text:
list every obligation on MY side, with the clause number for each.
If something is ambiguous, quote the ambiguous sentence rather
than resolving it yourself.

<contract>[paste document]</contract>

Requiring clause numbers makes every claim checkable, and with 1M-token context on current models, full contracts fit easily.

72. Compare these two versions of the document and list every substantive change — ignore formatting and renumbering.
73. Extract every date, amount, and named party from this document into a table, with the source sentence for each.
74. Here’s a messy CSV export. Describe its structure, find the data-quality problems, and write the cleaning steps in order.
75. Write the spreadsheet formula for [calculation], then explain it cell by cell and list the inputs that would break it.
76. Summarize this 60-page PDF for a decision-maker: what’s decided, what’s open, what they’re being asked to approve.
77. Turn this unstructured feedback dump into themes, each with a count, two verbatim quotes, and a severity rating.
78. Check this report’s numbers for internal consistency — totals, percentages, and claims that contradict the tables.
79. Draft the data dictionary for the dataset below: each column’s meaning, type, valid range, and how nulls should be read.
80. From this tagged policy document, build a decision tree for the three most common cases our support team handles.

Career and Job Search (81–90)

81. You are the hiring manager for the job posting tagged below.
Review my resume against it ruthlessly: what gets me screened out,
what's buried that should lead, and the three lines you'd rewrite
first — with the rewrites.

<job_posting>[paste]</job_posting>
<resume>[paste]</resume>

82. Interview me for [role]: one question at a time, follow-ups included, then score each answer with what a stronger answer contained.
83. Rewrite these resume bullets from duties to outcomes, keeping every number I gave you and inventing none.
84. Draft my self-review from the tagged accomplishments — confident, specific, and free of both bragging and apology.
85. Build my negotiation prep for this offer: my leverage, their constraints, the number to open with, and the script for the first two minutes.
86. Write a LinkedIn summary in my voice sample below that says what I do without a single buzzword.
87. Compare the two tagged job offers against MY stated priorities, then pressure-test my ranking of those priorities.
88. Prepare me for the “why are you leaving” question: honest, forward-looking, and three phrasings from safe to candid.
89. Turn this project I shipped into a STAR story I can tell in 90 seconds, with the metric as the punchline.
90. Audit my portfolio/GitHub as a skeptical hiring manager with six minutes: what impresses, what raises flags, what to fix this weekend.

Personal Decisions and Life Admin (91–100)

91. Help me decide: [decision]. Think through this before answering.
First list what I seem to be optimizing for based on how I framed
it, and challenge that framing if it looks off. Then lay out the
real options — including ones I didn't mention — with the strongest
case for and against each. End with the question I should answer
before anyone can advise me.

Most bad decisions are answers to the wrong question, and Claude spots that reliably when explicitly licensed to challenge the framing.

92. Plan a 7-day trip to [place] for [travelers and constraints]: realistic pacing, one contingency per day, and what to book now vs there.
93. Build a weekly meal plan around [constraints], with one shopping list organized by store aisle.
94. Review my monthly budget below and find the three cuts I’d feel least and the one expense that’s actually an investment.
95. Draft the difficult message to [person] about [situation]: honest, kind, and impossible to read as an attack. Two versions — softer and firmer.
96. Turn my vague goal of [goal] into a 12-week plan with weekly checkpoints I can verify in under a minute.
97. I’m comparing [product A] vs [product B] for [use]. Interrogate my needs first — three questions — then recommend against my answers.
98. Write my morning routine for a week where [constraint], designed around when my energy actually peaks: [pattern].
99. Help me say no to [request] gracefully: three scripts — warm, neutral, and final.
100. Run a year-in-review interview with me: ten questions, one at a time, then synthesize my answers into three themes and one resolution worth keeping.

What People Get Wrong About “Best” Prompts

• Collecting instead of adapting. A prompt library is a starting grid, not a finish line — the placeholders are where your value goes.
• Judging on one run. A good prompt is one that works the tenth time, on a mediocre input, on a Tuesday. Test before you trust.
• Ignoring model choice. The same prompt yields different depth on Haiku 4.5 versus Opus 4.8 or Fable 5. Match the model to the stakes — our model comparison guide shows how.
• Re-pasting what should be permanent. Any prompt you use weekly belongs in a Project instruction or system prompt, not your clipboard.

FAQ

Next Steps

Copy the five prompts closest to your daily work, adapt the placeholders once, and retire the ones that don’t earn their keep. When a single category becomes your bottleneck, go deep: the coding list and writing list expand their sections tenfold, and the prompts section indexes everything else.