Best AI Prompts for Writers

AI summary

Seven AI prompts for writers, essayists, newsletter operators, and book authors that scaffold the editorial work without flattening voice: outline stress-testing, sentence-level self-edit, reader-as-stranger read, title generation, pitch letters, voice audits, and open-rate diagnosis. AI handles structure; your voice handles everything that matters.

Prompt 1

Outline Tester

Most writers test their outlines on themselves, badly. This prompt tests it the way a reader would, before you commit a week to a draft that does not work.

Here is the outline for [PROJECT: essay / chapter / article / talk]:

[PASTE OUTLINE]

The one thing I want the reader to walk away with: [THESIS IN ONE SENTENCE]
The audience: [WHO YOU ARE WRITING FOR]
The length target: [WORDS]

Stress test the outline:

1. THE THESIS TEST: read the outline back to me and state the thesis it actually delivers. If it is not the thesis I stated, the outline is broken.
2. THE PROMISE-PAYOFF MATCH: for each section, what the section promises and whether the section pays it off.
3. THE WEAKEST SECTION: the section a reader is most likely to skim or close on. Why.
4. THE BURIED LEDE: if a moment 2/3 through the outline should actually be the opening, name it.
5. THE MISSING TURN: every good piece has a turn (a complication, a contradiction, a deepening). Does this outline have one. If not, where could it go.
6. THE CUT CANDIDATE: which section, if cut, would actually strengthen the piece.
7. THE READER'S QUESTION the outline leaves unanswered. Add it or accept it.

Do not rewrite the outline. Stress-test it. I will rewrite.

When to use: Day one of any longform piece, before drafting. · Best model: Claude. The discipline about not rewriting is what makes the prompt work.

Prompt 2

Sentence-Level Self-Edit

You have a draft. You have read it 12 times and stopped seeing it. This prompt is the line-edit pass that makes you see it again.

Here is a section of my draft I want to line-edit:

[PASTE 300-1000 WORDS]

My voice notes (so the edit respects me, not flattens me):

[2-3 SENTENCES ON WHAT YOUR VOICE IS: rhythmic / dry / warm / chopped / Latinate / etc.]

Line edit:

1. THE WORDS TO CUT: every word that does not earn its place. List them.
2. THE PHRASES THAT REPEAT: where I say the same thing twice in different words.
3. THE WEAK VERBS: every "is" or "there are" that hides a stronger verb. Suggest the stronger verb but do not impose it.
4. THE CADENCE CHECK: where two or three sentences in a row share the same shape and would benefit from variety.
5. THE STRONGEST SENTENCE: which sentence is doing the most work and should be protected through any rewrite.
6. THE LINE TO RECONSIDER: the sentence that might say something I do not actually mean. Flag and explain.

Do NOT rewrite the section in your own voice. Edit my voice harder, not into a different voice. If you suggest a replacement word, keep it in my register.

When to use: Last pass before you send a draft anywhere. · Best model: Claude. Voice preservation matters more than any other model trait here.

Prompt 3

Reader-as-Stranger Read

You read your draft like the writer. Your reader will read it like someone who has never thought about this. This prompt simulates that read.

Here is a piece I am about to publish:

[PASTE FULL PIECE]

What I assume the reader already knows: [LIST]
What I assume the reader cares about: [LIST]

Read the piece as a smart but unfamiliar reader would:

1. THE FIRST 100 WORDS: did the piece earn my attention. Yes / no / where I would have stopped.
2. THE CONFUSION POINTS: 3-5 places where, reading cold, I would have lost the thread or wondered what something meant.
3. THE ASSUMED KNOWLEDGE: things the piece assumes I know that, as a stranger, I would not.
4. THE EMOTIONAL READ: what the piece made me feel, where it tried to make me feel something and missed.
5. THE QUESTION I LEAVE WITH that the piece did not answer.
6. WHAT I WOULD TELL A FRIEND about this piece if I recommended it. If I cannot summarize, the piece does not have a clear point.
7. THE ONE LINE I WOULD QUOTE if I shared this online.

Do not soften the read. The goal is to surface what readers will actually experience, not protect the writer.

When to use: Before publishing anything longer than 800 words. · Best model: Claude or Grok. Grok is sharper on “would have stopped here” reads.

Prompt 4

Title and Subtitle Generator

Titles do the work of getting the piece read at all. Most writers throw the first title that works. This prompt produces 10 so you pick the best.

Here is my piece:

THESIS in one sentence: [THESIS]
OPENING in two sentences: [PASTE]
KEY MOMENT or BEST LINE: [QUOTE]
WHERE THE PIECE WILL BE PUBLISHED: [PLATFORM]
MY VOICE: [BRIEF DESCRIPTION]

Produce 10 title-and-subtitle pairs in 5 categories:

1. CONTRARIAN: titles that take a popular belief and challenge it.
2. SPECIFIC PROMISE: titles that name what the reader will walk away with.
3. CURIOSITY GAP: titles that pose a question the reader cannot stop thinking about.
4. PERSONAL/CONCRETE: titles built from one specific moment or detail in the piece.
5. PROVOCATIVE: titles that risk alienating some readers in order to attract the right ones.

For each, give the title (max 12 words) and a subtitle (one sentence).

Do NOT use: "How I", "What I learned", "The truth about", "Why X is actually Y" unless the piece really does make that claim. No clickbait that does not pay off in the content.

At the end, name your top 3 with one-sentence reasoning, and the one that is most likely to underperform.

When to use: After the draft is locked, before publish. · Best model: Claude (most disciplined about no-clickbait rule).

Prompt 5

Pitch Letter Drafter

You have an idea. You need to pitch it to an editor or producer. Most pitches die in the opening paragraph. This prompt drafts one that gets read.

I want to pitch this piece:

IDEA in one sentence: [IDEA]
WHY NOW: [TIMELINESS]
WHY ME: [YOUR EXPERTISE OR ACCESS, BRIEFLY]
TARGET OUTLET: [PUBLICATION]
TARGET EDITOR: [NAME / SECTION]
WHAT I KNOW ABOUT THE OUTLET'S RECENT WORK: [WHAT THEY HAVE PUBLISHED LATELY]

Draft a 250-word pitch letter:

1. SUBJECT LINE: 6-9 words, specific to the idea, not "Pitch" or "Story idea."
2. OPENING: 2 sentences that frame the idea as a story the editor's readers would actually want.
3. THE STAKES: 2-3 sentences on why this matters and why now.
4. THE STRUCTURE: 1 sentence on how the piece would work (reported feature / essay / Q&A / etc.).
5. WHY ME: 1 sentence, no resume dump.
6. THE CLOSE: clear ask and a deliverable timeline.

Do NOT use: "As a longtime reader...", "I have always wanted to write for you", "In our increasingly...", "In today's world." Show you have read the outlet by referencing a specific recent piece (without flattery).

Do not invent expertise. Do not promise sources I do not have.

When to use: After 30 minutes of research on the editor and outlet. · Best model: Claude. The no-flattery and no-cliche discipline is what differentiates a successful pitch.

Prompt 6

Voice Audit

Your voice has drifted and you do not know where. This prompt finds the drift by comparing recent work to earlier work.

Here are 2-3 paragraphs from a piece I wrote in the past month:

[PASTE]

Here are 2-3 paragraphs from a piece I wrote 1-2 years ago that I consider strong:

[PASTE]

Identify the drift:

1. WORDS I USE NOW that I did not use then.
2. CADENCE CHANGES: am I writing longer or shorter sentences. Latinate or Anglo-Saxon. Concrete or abstract.
3. TONE CHANGES: more authoritative? More hedged? More performative?
4. THE QUALITY THAT MADE THE OLD PIECE WORK that the new piece is missing.
5. THE QUALITY THE NEW PIECE HAS that the old piece did not, that I should keep.
6. THE ONE HABIT to break in my next piece, with a specific exercise.
7. THE QUESTION I should ask myself before publishing for the next month.

Be direct. Voice drift is usually an unconscious response to audience pressure or AI rewriting; name it if you see it.

When to use: Once a quarter, or whenever you stop liking your own writing. · Best model: Claude. The discipline about voice analysis matters and Claude is the most disciplined about avoiding flattery.

Prompt 7

Newsletter Open-Rate Diagnosis

If you write a newsletter, opens are the single largest signal you have. Most writers panic-tweak everything when opens dip. This prompt finds the real cause.

Here is the open-rate trend for my newsletter over the past 12 issues:

[FOR EACH: subject line, send date, open rate, comparison vs prior 4-issue average]

My audience: [BRIEF]
My publication cadence: [WEEKLY / TWICE-WEEKLY / DAILY]

Diagnose:

1. THE TREND: is the trend down, flat, or noise. Be statistical, not emotional.
2. THE SUBJECT-LINE PATTERN: across the 12, which subject-line types open best.
3. THE TIMING PATTERN: are open rates correlating to day or hour of send.
4. THE CONTENT PATTERN: based on subject lines, which topics open best.
5. THE LIST-HEALTH FACTOR: any signs of list fatigue or deliverability issues vs content fit issues.
6. ONE EXPERIMENT to run this week to test the most likely cause.
7. ONE THING NOT TO DO: a common panic move that would make things worse.

Do not let me panic-tweak. Most open-rate dips are 2-3 issues of noise, not a trend.

When to use: When opens have dropped 15 percent across 3+ issues. · Best model: Claude or Grok. Both push back on emotional decision-making with data.

Connectors are now standard

Claude, ChatGPT, and Grok all support connectors that let your AI read live data from your work tools (Gmail, Notion, GitHub, Asana, HubSpot, Stripe, and many more) instead of relying on you to paste context. For writers this means the AI can read your Notion or Google Drive draft folder, your Beehiiv or Substack newsletter archive, your Descript transcript file (if you draft from audio), or your Gmail thread with editors and readers.

For writers, the connectors worth pairing with these prompts:

Get Smarter About AI Every Morning

Free daily newsletter. Built for people who want to use AI well, not chase every model.

Free forever. Unsubscribe anytime.

Two ways to go further