Thread prompts · Hook + structure

10 AI thread prompts for high-completion threads

Most AI-generated threads die after tweet 3. These 10 prompts produce threads with strong hooks, structural arcs, and the cliff-hanger pattern that earns full reads — copy-paste ready for ChatGPT, Claude, Gemini.

A great thread does three things: hooks in tweet 1, sustains attention tweet-by-tweet, and ends with a payoff. AI struggles with all three by default. The prompts below constrain the AI to produce thread structures that consistently earn completion — variables in {curly braces}.

The prompts

Prompt #1

7-tweet framework thread

Write a 7-tweet X (Twitter) thread teaching the {N}-step framework for {topic}. Structure: Tweet 1 (HOOK): a sharp opening with the framework promise + curiosity gap. Don't reveal the steps. Tweets 2-{N+1}: one step per tweet, with a label + explanation + concrete example. Final tweet: a one-sentence summary + invitation to bookmark. Each tweet max 230 chars. Number them with '1/', '2/', etc. Voice: confident, plain English, no jargon.

Why it works

7-tweet threads are the highest-engagement length. The hook + 5 steps + summary structure is proven. Forcing concrete examples per step prevents generic AI 'be intentional' output.

Best for

Claude (best at framework structure + examples).

Variables to fill in

{N (number of steps)}{topic}

Prompt #2

Personal-cost lesson thread

Write a 6-tweet personal-cost-lesson thread about {experience}. Structure: T1: HOOK opening with a specific cost (dollar amount, time elapsed, opportunity missed) you paid. Don't reveal the lesson yet. T2: Context — what was happening, what you believed at the time. T3: The mistake / cost in detail. T4: The realization that came after. T5: The general lesson, framed for the reader. T6: A short reflection + permission to learn from your mistake. Max 230 chars per tweet. Voice: honest, lightly self-deprecating, no self-pity.

Why it works

Personal-cost threads are the highest-engagement narrative format. The 6-tweet structure threads the needle between 'too short' and 'too long'. Tone constraints prevent the typical AI 'and now I'm grateful' ending.

Best for

Claude 3.7+ (best at narrative tone).

Variables to fill in

{experience}

Prompt #3

Numbered list thread

Write a 9-tweet thread of {N} things about {topic}, where {N} = 7. Structure: T1: HOOK promising the list with a curiosity gap (e.g., 'Number 4 changed everything'). T2-T8: one item per tweet, numbered, with a 2-sentence explanation. T9: closing reflection. Max 220 chars per tweet. Order the items strategically — strongest at #1 and #7, most surprising at #4. Voice: direct, no fluff.

Why it works

Numbered lists earn high completion rates (predictable structure). The strategic ordering — strong opener + 'mid-list surprise' + strong closer — maximizes completion. Curiosity gap in T1 drives full-thread reads.

Best for

GPT-4 (best at concise list formatting).

Variables to fill in

{topic}{N (number of items)}

Prompt #4

Case-study thread

Write an 8-tweet case-study thread about {someone/something} achieving {outcome}. Structure: T1: HOOK with the outcome (specific number / result). T2-T3: the starting state — what they/it looked like before. T4-T6: the key 3 decisions/actions that mattered. T7: the result + what's transferable. T8: the takeaway for the reader. Max 240 chars per tweet. Use real specifics; don't fabricate numbers.

Why it works

Case studies earn high authority + saves. The structure (start → 3 decisions → result → takeaway) is proven. The 'transferable' constraint forces the AI to think about reader applicability.

Best for

Claude (best at narrative + analysis blend).

Variables to fill in

{someone/something}{outcome}

Prompt #5

Contrarian-argument thread

Write a 6-tweet thread making a contrarian argument about {topic}. Structure: T1: HOOK stating the common view + your dissent. T2: Why the common view is widely held (steelman it). T3-T4: 2-3 pieces of evidence supporting your contrarian position. T5: When the common view IS right (showing nuance). T6: The actionable takeaway. Max 240 chars per tweet. Voice: confident but not aggressive. Steel-man the opposing view first — this earns trust.

Why it works

Steelmanning before disagreeing is high-trust contrarian content. Most AI contrarian output skips the steelman, which reads as straw-manning. The 'when the common view IS right' nuance prevents flame-war comments.

Best for

Claude 3.7+ (best at nuanced argument).

Variables to fill in

{topic}

Prompt #6

Behind-the-scenes thread

Write a 7-tweet behind-the-scenes thread about {project/decision/process}. Structure: T1: HOOK with what's visible to outsiders. T2: What outsiders don't see. T3-T5: 3 specific details from inside that surprised even you. T6: The pattern these reveal. T7: The reader takeaway. Max 240 chars per tweet. Use specific numbers / observations. Voice: matter-of-fact insider, not bragging.

Why it works

Behind-the-scenes content earns saves and follows. The 'surprised even you' filter forces genuinely novel observations (not just things you knew already). Steelmans against humble-brag tone.

Best for

Claude (best at insider tone).

Variables to fill in

{project/decision/process}

Prompt #7

Before/after journey thread

Write a 7-tweet before/after journey thread for {transformation}. Structure: T1: HOOK with the after state (specific). T2: The before state (specific). T3: The duration + what most people assumed it would take. T4-T6: the 3 actions that compounded. T7: the lesson + actionable next step for someone in the 'before' state. Max 240 chars per tweet. Specific numbers throughout — no vague 'I worked hard'.

Why it works

Before/after journeys earn the highest follower-conversion of any thread format (people identify with the 'before' and follow for ongoing progress). Specific-numbers constraint prevents the typical AI motivational vagueness.

Best for

Claude or GPT-4 with strict variable filling.

Variables to fill in

{transformation}

Prompt #8

Anti-advice thread

Write a 6-tweet thread on '{N} things you should STOP doing about {topic}'. Structure: T1: HOOK with the 'stop doing' promise. T2-{N+1}: one thing per tweet to stop doing, with the reason it's harmful, and what to do instead. Final tweet: the underlying principle these all share. Max 220 chars per tweet. Each 'stop doing' should be specific enough to be controversial. Voice: direct, no hedging.

Why it works

Anti-advice ('stop doing') outperforms 'do this' threads in 2026 — readers are saturated with positive advice. The 'specific enough to be controversial' filter prevents generic 'stop being negative' output.

Best for

GPT-4 (sharper at the anti-advice angle).

Variables to fill in

{N (3-5 items)}{topic}

Prompt #9

Industry-state thread

Write an 8-tweet thread on the current state of {industry/space} in {year}. Structure: T1: HOOK with the headline-state assessment. T2-T3: what's working in the space right now. T4-T5: what's broken. T6: where it's heading. T7: who's well-positioned. T8: actionable for the reader. Max 250 chars per tweet. Avoid generic 'innovation is happening fast' — be specific about companies, technologies, trends.

Why it works

Industry-state threads from credible operators earn massive shares within that industry. The structure (what's working / broken / heading / well-positioned) is comprehensive without being exhausting.

Best for

Claude 3.7+ (best at industry analysis).

Variables to fill in

{industry/space}{year}

Prompt #10

Question-led discovery thread

Write a 7-tweet thread that walks through how you arrived at an answer to: '{question}'. Structure: T1: HOOK with the question. T2: Why this question matters. T3-T5: 3 wrong (or partial) answers you tried first + why each failed. T6: the answer you arrived at. T7: how to apply it. Max 240 chars per tweet. The 3 wrong answers should be ones a thoughtful person might also try.

Why it works

Showing wrong answers BEFORE the right one earns trust (signals genuine exploration). Most AI threads jump straight to the answer; the discovery framing differentiates and earns higher completion.

Best for

Claude (best at the multi-attempt narrative).

Variables to fill in

{question}

Related prompt galleries

Tweet prompts

10 AI tweet prompts that don't sound like AI

Hook prompts

10 AI prompts for tweet hooks that stop the scroll

Viral tweet prompts

10 AI prompts for viral-engineered tweets

Common questions

How long should an AI-generated thread be?+

7 ±2 tweets is the sweet spot — long enough to deliver complete value, short enough that 70%+ of readers complete. Threads shorter than 5 tweets don't earn the format bonus from X's algorithm; threads longer than 12 tweets see completion drop to under 40%, and trailing low-engagement tweets drag the algorithmic score down.

Should I number the tweets in my thread?+

Yes for threads >5 tweets. Numbering ('1/', '2/') raises completion rates by 10-15% by signaling structure and letting readers track progress. Skip numbering on 3-4 tweet threads — the overhead isn't worth it at that length. Most of the prompts above include numbering instructions.

Can AI write threads that go viral?+

Sometimes. AI can produce structurally-strong threads consistently. The 'viral' moment usually requires the human-supplied SPECIFICS — your real numbers, your specific story, your real industry observation. AI gives you the scaffolding; your specifics make it resonate. Expect AI to produce 'consistent quality' threads; expect human-edited specifics to produce occasional viral moments.

Should I edit AI-generated threads before posting?+

Yes, always. The first pass from any AI is ~80% there. The last 20% (your voice, specific details, sharp edges, removing AI verbal tics like 'leveraged' or 'in today's landscape') is human work. Plan to spend 5-10 minutes editing every AI-generated thread before publishing. The edited version dramatically outperforms the raw output.

Skip the prompt engineering

AutoTweet's AI generates tweets in your voice using these structural patterns automatically. No prompt-tweaking required. 14 voice-matched tweets queued the moment you connect X.

Cancel anytime