16 Tasks · 48 Prompts · April 2026
Claude vs ChatGPT vs Gemini: the same prompt, three ways.
One task, three model-tuned prompts. The same ask, written to play to each model's actual strengths. Copy whichever fits the tool you're using.
Most prompt libraries pretend the three major models are interchangeable. They aren't. Claude follows complex multi-step instructions more literally and handles nuance best. ChatGPT is the strongest conversational writer and the most forgiving of loose phrasing. Gemini is cleanest with structured output, tables, and hard constraints like character limits.
The same task, written three ways, will produce measurably different results. A prompt that works in ChatGPT often underperforms in Claude because Claude is waiting for structure you didn't give it, specifically the XML tags that Anthropic explicitly recommends for complex prompts. A prompt that sings in Claude often feels stiff in ChatGPT because ChatGPT wants to talk to you. Knowing the difference is the point.
Every task below shows the same ask in three versions. Each follows that model's official prompt engineering guidance: Anthropic's docs for Claude, OpenAI's guide for ChatGPT, and Google's strategies for Gemini. Copy the one that matches the tool you're using.
What each model is actually good at.
Claude
Best at literal instruction-following, XML structure, and nuanced tone. If your prompt has five rules, Claude will follow all five. Strongest at writing that sounds human. See Anthropic's prompt engineering guide.
ChatGPT
Best at conversational output and creative variation. Most forgiving of loose prompts. If you want three different versions of something, ChatGPT produces the widest range. See OpenAI's prompting guide.
Gemini
Best at structured output: tables, lists, hard character limits. Also strongest at factual queries needing current data (it searches natively). Use when formatting matters.
Writing the first touch — email, LinkedIn, voice note — across all three models.
TASK 01 / 16
Three-sentence cold email
The default cold opener. Short, specific, one clear ask.
Why they differ: Claude will follow your rules more literally. ChatGPT produces more natural conversational tone. Gemini is cleanest for structured variants if you need A/B tests.
CLAUDE
<role>You are a B2B sales copywriter. You do not use corporate language.</role>
<task>Write one cold email, under 75 words, exactly 3 sentences.</task>
<input>
- Prospect: [NAME, TITLE, COMPANY]
- My product: [PRODUCT]
- Specific observation to anchor on: [OBSERVATION]
</input>
<structure>
<sentence_1>A specific observation about them or their company. No flattery.</sentence_1>
<sentence_2>The pain that observation implies, in their own language.</sentence_2>
<sentence_3>One low-commitment ask. Not a meeting.</sentence_3>
</structure>
<forbidden>
- "Hope this finds you well"
- "Quick question"
- "I know you're busy"
- Em dashes
- Rule-of-three lists
</forbidden>
CHATGPT
You're a B2B sales copywriter who writes the way people actually talk.
Write a cold email for me:
- Prospect: [NAME, TITLE, COMPANY]
- My product: [PRODUCT]
- Specific observation I want to anchor on: [OBSERVATION]
Rules:
- Exactly 3 sentences, under 75 words total
- First sentence = something specific I noticed about them (no flattery)
- Second sentence = what that means for their role (in their words)
- Third sentence = one small ask, NOT a meeting request
Don't write "hope this finds you well," "quick question," or "I know you're busy." Don't use em dashes. Don't do the "not just X, but Y" thing. Write it the way you'd Slack a colleague.
GEMINI
Write a cold email. Output in this exact format:
PROSPECT: [NAME, TITLE, COMPANY]
PRODUCT: [MY PRODUCT]
OBSERVATION: [SPECIFIC DETAIL I FOUND]
CONSTRAINTS:
- 3 sentences maximum
- 75 words maximum
- Sentence 1: specific observation, no flattery
- Sentence 2: implied pain in their language
- Sentence 3: one low-commitment ask (not a meeting)
- No clichés (no "hope this finds you well", "quick question", "I know you're busy")
- No em dashes
OUTPUT: the email only, no commentary.
TASK 02 / 16
LinkedIn connection request under 200 characters
The character limit forces real discipline. Hardest to get right.
Why they differ: Gemini is the most obedient with hard character limits. Claude is best for tonal variation. ChatGPT produces the most natural-sounding single attempt.
CLAUDE
<task>Write 3 LinkedIn connection requests, each under 200 characters including spaces.</task>
<context>
- Prospect: [NAME, TITLE, COMPANY]
- Hook from their profile (real, specific): [HOOK]
</context>
<variations>
1. Curious (asking a question)
2. Complimentary-but-specific (reference a specific thing they wrote/did)
3. Useful (offering to share something)
</variations>
<rules>
- Under 200 characters, count spaces
- Do NOT mention my product
- Do NOT say "I'd love to connect"
- Each variation must give them a real reason to accept
</rules>
Count the characters of each and show the count.
CHATGPT
Write me a LinkedIn connection request under 200 characters.
Context:
- Prospect: [NAME, TITLE, COMPANY]
- Something I noticed on their profile: [HOOK]
Rules:
- HARD limit: 200 characters including spaces
- Don't mention my product
- Don't say "I'd love to connect" or "let's network"
- Give them a real reason to accept
Give me 3 variations with different vibes: (1) curious, (2) complimentary about something specific, (3) offering to share something useful. Show the character count next to each.
GEMINI
Generate 3 LinkedIn connection request messages.
INPUT:
- Name: [NAME]
- Title: [TITLE]
- Company: [COMPANY]
- Profile detail: [SPECIFIC HOOK]
CONSTRAINTS:
- Maximum 200 characters each (including spaces)
- No product pitches
- No generic phrases ("I'd love to connect", "let's network")
OUTPUT FORMAT (table):
| # | Tone | Message | Char count |
|---|------|---------|------------|
| 1 | Curious | ... | XXX |
| 2 | Specific compliment | ... | XXX |
| 3 | Offering value | ... | XXX |
TASK 03 / 16
Follow-up email that's not 'just circling back'
Second-touch email that earns its place by bringing new information.
Why they differ: Claude handles the 'refuse if nothing new' logic best. ChatGPT writes the warmest re-engagement. Gemini structures A/B test variants cleanly.
CLAUDE
<role>Sales copywriter. Ruthless about follow-up quality.</role>
<task>Write a follow-up email that introduces NEW information, not a restatement of my first email.</task>
<context>
- First email sent: [DATE, TOPIC]
- Original ask: [ASK]
- New information I now have: [NEW INFO]
</context>
<rules>
- Line 1: the new information (no "checking back in", no "following up")
- Line 2: how it changes or sharpens what I said before
- Line 3: the original ask, reframed in light of new info
- Under 75 words
</rules>
<critical>If the [NEW INFO] I provided is weak, shallow, or not actually new, refuse to write the email and tell me the follow-up shouldn't be sent. Suggest waiting for a real trigger instead.</critical>
CHATGPT
Write me a follow-up email. But here's the catch: it has to bring NEW information, not just restate my first email. No "just circling back" energy.
- First email topic: [TOPIC]
- Original ask: [ASK]
- New info I have now: [NEW INFO]
Write:
- Line 1: the new info (skip the "following up" setup)
- Line 2: how that changes the picture
- Line 3: the ask, reframed
Under 75 words. Conversational, like a peer not a rep.
If the "new info" I gave you is weak or not actually new, tell me the follow-up shouldn't be sent. I'd rather wait.
GEMINI
Create a follow-up email.
INPUTS:
- Original email date: [DATE]
- Original topic: [TOPIC]
- Original ask: [ASK]
- New information available: [NEW INFO]
REQUIREMENTS:
- Must introduce new information (not restate original)
- Structure: new info → implication → reframed ask
- Maximum 75 words
- Zero filler phrases ("just circling back", "checking in")
DECISION LOGIC:
If [NEW INFO] is generic, weak, or not genuinely new → output: "RECOMMENDATION: Do not send. Wait for a real trigger event." + explanation.
Otherwise → output the email.
TASK 04 / 16
Voice note script for LinkedIn DM
LinkedIn voice DMs get 5-10x the reply rate of text, but the script has to sound spoken.
Why they differ: ChatGPT is strongest at spoken-feel output. Claude follows the timing structure most precisely. Gemini is best if you want a polished alternative.
CLAUDE
<task>Write a 45-60 second LinkedIn voice note script.</task>
<important>This script will be SPOKEN, not read. Use short sentences. Add natural pauses marked with "...". No jargon.</important>
<context>
- Prospect: [TITLE] at [COMPANY]
- Specific detail I'll reference: [DETAIL]
- Why I'm reaching out: [REASON]
</context>
<structure>
<opening seconds="0-5">Their name + one hyper-specific detail</opening>
<middle seconds="5-40">Why I'm reaching out, conversationally</middle>
<close seconds="40-60">Light ask</close>
</structure>
<rules>
- No "hope you're doing well"
- No corporate vocabulary
- Write like I'd actually talk to a peer
- Mark pauses explicitly with "..."
</rules>
CHATGPT
Write me a 45-60 second LinkedIn voice note script.
This is going to be SPOKEN out loud. Write it the way I'd actually talk to a peer at a conference, not how I'd write an email. Short sentences. Natural pauses (mark them with "..."). No jargon.
Context:
- Prospect: [TITLE] at [COMPANY]
- Specific thing I'd reference: [DETAIL]
- My reason for reaching out: [REASON]
Timing:
- First 5 seconds: their name and one specific detail that proves I'm not spamming
- Next 35 seconds: why I'm reaching out, said like a human
- Last 20 seconds: a light ask
Read it back to yourself. If it sounds like you're "presenting" instead of "talking," rewrite it.
GEMINI
Generate a LinkedIn voice note script.
DURATION: 45-60 seconds when spoken
INPUTS:
- Prospect: [TITLE] at [COMPANY]
- Specific detail to reference: [DETAIL]
- Reason for outreach: [REASON]
TIMING STRUCTURE:
| Segment | Duration | Content |
|---------|----------|---------|
| Hook | 0-5s | Name + one specific detail |
| Body | 5-40s | Reason for reaching out |
| CTA | 40-60s | Light ask |
FORMATTING:
- Written for speech, not reading
- Short sentences (<15 words each)
- Mark natural pauses with "..."
- No jargon, no corporate phrases
OUTPUT: script only, with pause markers.
Account research, call prep, competitive intel — where each model's strengths are most visible.
TASK 05 / 16
One-page account research brief
Research a target account in 90 seconds with genuine, grounded output.
Why they differ: Gemini wins on current data (it searches natively). Claude is best for structured synthesis. ChatGPT writes the most readable summary.
CLAUDE
<role>B2B sales analyst. You cite sources or say "UNKNOWN".</role>
<task>Produce a one-page research brief on [COMPANY] in under 300 words.</task>
<sections_required>
<who>One sentence on what they do. No marketing language.</who>
<signals>Recent hires, funding, layoffs, product launches. Cite sources.</signals>
<buyer>Who likely owns the buying decision for [MY PRODUCT]. Title + reporting line.</buyer>
<hook>One specific hook I can use in a cold email, tied to a signal above.</hook>
</sections_required>
<critical>
Do NOT fabricate. If you don't know, write "UNKNOWN - investigate before outreach".
No "in today's competitive landscape" language.
Every signal must have a source or be marked as unverified.
</critical>
CHATGPT
Act as a B2B sales analyst. Give me a one-page research brief on [COMPANY], under 300 words.
My product: [PRODUCT DESCRIPTION]
Four sections only:
1. What they actually do (one sentence, no marketing speak)
2. Signals of growth or pain in the last 90 days (recent hires, funding, layoffs, launches) — with source links
3. Who likely owns the buying decision for my product (title + who they report to)
4. One hook I could use in a cold email, tied to a real signal from section 2
Important: if you can't find real info for any section, say "UNKNOWN - investigate before outreach." Don't make things up. No filler phrases like "in today's competitive landscape." Just signal.
GEMINI
Research [COMPANY] and produce a structured sales brief.
MY PRODUCT: [PRODUCT]
OUTPUT SECTIONS (required):
## WHAT THEY DO
One sentence, plain language, no marketing terms.
## RECENT SIGNALS (last 90 days)
| Signal | Date | Source | Implication |
|--------|------|--------|-------------|
| ... | ... | ... | ... |
## LIKELY BUYER
- Title: ...
- Reports to: ...
- Team size: ...
## OUTREACH HOOK
One specific angle tied to a signal above.
CONSTRAINTS:
- Maximum 300 words total
- All claims must cite sources
- Use "UNKNOWN" where data unavailable
- No fabrication
TASK 06 / 16
Discovery call question bank
15-20 discovery questions tailored to a specific deal and persona.
Why they differ: Claude is best at persona-specific nuance. ChatGPT generates the widest creative range. Gemini structures the question categories most cleanly.
CLAUDE
<task>Generate 15-20 discovery questions tailored to this specific deal.</task>
<context>
- Prospect: [TITLE] at [COMPANY]
- My product: [PRODUCT]
- Known context: [WHAT I ALREADY KNOW]
- Key unknowns: [WHAT I NEED TO LEARN]
</context>
<categories>
1. CURRENT STATE (how they operate today)
2. PAIN & PRIORITY (what's broken, how much it hurts)
3. DECISION (who decides, how, by when, budget)
4. FUTURE STATE (what "good" looks like to them)
</categories>
<requirements>
- Tailored to THIS [TITLE] at a company with [CHARACTERISTICS]
- No "what keeps you up at night" or similar clichés
- Phrased conversationally, not interview-style
- End with ONE unexpected question that might surface something rehearsed answers wouldn't
</requirements>
CHATGPT
Help me prep for a discovery call. Generate 15-20 discovery questions tailored to this specific deal.
- Prospect: [TITLE] at [COMPANY]
- My product: [PRODUCT]
- What I already know: [CONTEXT]
- What I most need to learn: [UNKNOWNS]
Group them into four buckets:
1. Current State (how they operate today)
2. Pain & Priority (what's broken, how much it hurts)
3. Decision Process (who, how, by when, budget)
4. Future State (what "good" looks like)
Rules:
- Ban all clichés ("what keeps you up at night", "paint me a picture")
- Phrase questions the way I'd actually ask them, not like an interview script
- Tailor to someone at a company with [CHARACTERISTICS]
- End with one "unexpected" question that might surface something they haven't rehearsed an answer to
GEMINI
Generate a discovery call question bank.
DEAL CONTEXT:
- Prospect: [TITLE] at [COMPANY]
- Product: [PRODUCT]
- Known: [CONTEXT]
- Unknown: [GAPS]
OUTPUT: Markdown table with 15-20 questions.
| # | Category | Question | Purpose |
|---|----------|----------|---------|
| 1 | Current State | ... | ... |
| 2 | Pain & Priority | ... | ... |
| ... | ... | ... | ... |
CATEGORIES REQUIRED:
1. Current State (3-5 questions)
2. Pain & Priority (3-5)
3. Decision Process (3-5)
4. Future State (3-5)
5. Wildcard — 1 unexpected question
BANNED PHRASES:
- "What keeps you up at night"
- "Paint me a picture"
- "Where do you see yourself"
All questions must be conversational, not interview-style.
TASK 07 / 16
Competitor battlecard for a live deal
Who's in the deal with you, what's their angle, how do you win.
Why they differ: Claude gives the most nuanced 'don't trash competitor' framing. ChatGPT writes the most readable narrative. Gemini structures win/loss dimensions cleanest.
CLAUDE
<role>Competitive strategy analyst. You never recommend trashing a competitor.</role>
<task>Based on the context below, predict which competitor is most likely in this deal and how I should position.</task>
<context>
- Company: [COMPANY]
- Product I sell: [PRODUCT]
- Signals I've seen: [SIGNALS]
- Known tech stack: [STACK]
</context>
<output>
<most_likely_competitor>
<name>...</name>
<confidence>HIGH/MEDIUM/LOW</confidence>
<reasoning>...</reasoning>
</most_likely_competitor>
<their_likely_pitch>What they'd tell [PROSPECT]</their_likely_pitch>
<their_real_strengths>Not strawman versions. Honest.</their_real_strengths>
<their_real_weaknesses>Specific, not vague</their_real_weaknesses>
<diagnostic_question>ONE question to ask [PROSPECT] that reveals if [COMPETITOR] is actively in this deal</diagnostic_question>
<positioning>How to position without trashing them</positioning>
<do_nothing_check>Is the REAL competitor "do nothing" or "build in-house"? Status quo wins more deals than competitors.</do_nothing_check>
</output>
CHATGPT
Help me build a battlecard for an active deal.
- Company: [COMPANY]
- My product: [PRODUCT]
- Signals I've seen: [SIGNALS]
- Known stack: [STACK if any]
Figure out which competitor is most likely in this deal based on the context. Then tell me:
1. The competitor's likely pitch to [PROSPECT]
2. Their genuine strengths (not weak-man versions)
3. Their genuine weaknesses
4. ONE specific question I can ask [PROSPECT] that reveals whether this competitor is actually in the deal
5. How to position WITHOUT trashing them (senior buyers notice trash-talking and it loses deals)
Also — reality check: is the most likely "competitor" actually "do nothing" or "build in-house"? Status quo wins more deals than vendors. If that's the case, tell me.
GEMINI
Create a competitive battlecard.
INPUTS:
- Target account: [COMPANY]
- My product: [PRODUCT]
- Signals observed: [SIGNALS]
- Known stack: [STACK]
OUTPUT:
## MOST LIKELY COMPETITOR
- Name:
- Confidence: [HIGH/MEDIUM/LOW]
- Why:
## COMPETITIVE COMPARISON TABLE
| Dimension | Competitor | My Product |
|-----------|-----------|------------|
| Core strength | ... | ... |
| Pricing model | ... | ... |
| Implementation | ... | ... |
| Known weakness | ... | ... |
## DIAGNOSTIC QUESTION
One question to verify competitor presence: ...
## POSITIONING (no trash-talk)
- Lead with: ...
- Acknowledge: ...
- Differentiate on: ...
## STATUS QUO CHECK
Is the real competitor "do nothing" or "build in-house"? [Yes/No + reasoning]
TASK 08 / 16
Stakeholder map for a complex deal
Who matters, what they care about, how to win them.
Why they differ: Claude handles the psychological/political nuance best. ChatGPT writes the most readable narrative. Gemini produces the cleanest tabular map.
CLAUDE
<task>Build a stakeholder map for a deal with [COMPANY] on [PRODUCT].</task>
<context>
- Deal size: [$X]
- Known contact: [NAME, TITLE]
- Company size: [HEADCOUNT]
- Industry: [INDUSTRY]
</context>
<for_each_stakeholder>
<title>Title and seniority</title>
<role>CHAMPION / INFLUENCER / BUYER / USER / BLOCKER / UNKNOWN</role>
<personal_incentive>What they individually care about (not the company's interest — theirs)</personal_incentive>
<kill_risk>What they might do to slow or kill this deal</kill_risk>
<win_play>How to win them over or neutralize them</win_play>
</for_each_stakeholder>
<include_from>
- Champion's team
- Finance / procurement
- IT / security
- Legal
- End-users
- Exec sponsor
</include_from>
<critical>Flag any stakeholder role that's suspiciously absent from my current view. Missing a stakeholder is how deals die.</critical>
CHATGPT
Map the stakeholders for my deal with [COMPANY].
- Deal size: [$X]
- My known contact: [NAME, TITLE]
- Company size: [HEADCOUNT]
- Industry: [INDUSTRY]
- Product: [PRODUCT]
For each likely stakeholder, tell me:
- Their probable title and seniority
- Their role: Champion / Influencer / Buyer / User / Blocker / Unknown
- What THEY personally care about (not what the company cares about — their career incentive)
- How they might kill or slow this deal
- The best way to win them (or neutralize them)
Make sure to cover: the champion's team, finance/procurement, IT/security, legal, end-users, and an exec sponsor.
Most important: flag any role that's suspiciously MISSING from my current view. The stakeholder you don't see is the one that kills the deal.
GEMINI
Generate a stakeholder map for this deal.
DEAL:
- Company: [COMPANY]
- Headcount: [N]
- Industry: [INDUSTRY]
- Product: [PRODUCT]
- Deal size: [$X]
- Known contact: [NAME, TITLE]
OUTPUT TABLE:
| Title | Role | Personal Incentive | Risk to Deal | Win Play |
|-------|------|-------------------|--------------|----------|
| ... | Champion | ... | ... | ... |
| ... | Economic Buyer | ... | ... | ... |
| ... | IT/Security | ... | ... | ... |
| ... | Procurement | ... | ... | ... |
| ... | End User | ... | ... | ... |
| ... | Exec Sponsor | ... | ... | ... |
AFTER THE TABLE:
## MISSING FROM MY VIEW
List any stakeholder role that's suspiciously absent. These are deal-killers when ignored.
Halfway through
These work in any chat window. The Vault works on its own.
Prompts are what you paste. Agents are what you deploy. If the side-by-side above is useful, the Vault is 50 pre-built agents that run end-to-end.
See the Vault $99.99 →LinkedIn posts, blog briefs, case studies, and anything you'll publish.
TASK 09 / 16
LinkedIn post that attracts your ICP
Post content that makes your buyer come to you.
Why they differ: ChatGPT is strongest at punchy, scroll-stopping rhythm. Claude is best at nuanced POV. Gemini is best if you want structured A/B variants.
CLAUDE
<task>Write a LinkedIn post designed to attract engagement from [ICP].</task>
<goal>Get people from [ICP] to comment so I can follow up in DM.</goal>
<context>
- Topic: [TOPIC — pain-adjacent to what I sell]
- My POV (unpopular or not-widely-said): [POV]
- ICP: [TITLE + INDUSTRY]
</context>
<structure>
<line_1>A specific claim or observation. Not a question.</line_1>
<lines_2_through_5>Why I think this, told through a story or concrete example.</lines_2_through_5>
<line_6>One sentence that invites disagreement or addition.</line_6>
<line_7>No question. No "what do you think?" Let it land.</line_7>
</structure>
<rules>
- Under 150 words
- Punchy line breaks
- No hashtags
- First line must stop the scroll without being clickbait
</rules>
CHATGPT
Write a LinkedIn post that attracts my ICP to comment.
My goal: get people who match [ICP - title + industry] to engage so I can DM them.
- Topic: [TOPIC, pain-adjacent to my product]
- My take (should be unpopular or at least not-widely-said): [POV]
Structure it like:
- Line 1: a specific claim or observation (not a question) that stops the scroll
- Lines 2-5: why I think this, told through a story or a concrete example
- Line 6: one sentence that invites disagreement
- Line 7: no question, no "what do you think?" Let it sit
Hard rules:
- Under 150 words
- Punchy line breaks between thoughts
- No hashtags
- First line has to stop a scroll WITHOUT being clickbait ("you won't believe...")
- If the POV isn't actually unpopular, tell me and ask for a sharper one
GEMINI
Generate a LinkedIn post to attract ICP engagement.
INPUTS:
- ICP: [TITLE] at [INDUSTRY]
- Topic: [TOPIC]
- Point of view: [POV]
- Goal: drive comments from ICP
STRUCTURE:
```
[Line 1: scroll-stopping claim, NOT a question]
[Lines 2-5: evidence via story or example]
[Line 6: invites disagreement]
[Line 7: final statement, no question, no CTA]
```
OUTPUT 3 VARIANTS:
- Variant A: provocative tone
- Variant B: curious/open tone
- Variant C: contrarian tone
CONSTRAINTS:
- ≤150 words per variant
- No hashtags
- No "what do you think?"
- No clickbait openers
TASK 10 / 16
Blog post outline with SEO intent
Turn a keyword into a briefed outline ready for writing.
Why they differ: Gemini handles SEO structure and keyword coverage most systematically. Claude builds the most defensible argument flow. ChatGPT produces the most engaging angles.
CLAUDE
<task>Build a blog post outline around the target keyword: [KEYWORD].</task>
<context>
- Audience: [AUDIENCE]
- Search intent: [INFORMATIONAL / COMMERCIAL / TRANSACTIONAL]
- My product angle: [PRODUCT]
- Competitors ranking now: [CURRENT TOP 3 IF KNOWN]
</context>
<output>
<title>SEO-optimized title, under 60 characters, with target keyword near the front</title>
<hook>First 100 words. Must answer the query directly in sentence 1. No throat-clearing.</hook>
<outline>
5-8 sections. For each:
- H2 heading
- Key point (one sentence)
- Evidence I need to gather
- Specific example to include
</outline>
<search_intent_match>Explain how this outline matches the search intent better than current top 3 results</search_intent_match>
<word_count_target>1,500-2,500 words (justify)</word_count_target>
</output>
<critical>If the keyword is saturated by high-authority sites, tell me and suggest a long-tail variant that's rankable.</critical>
CHATGPT
Build me a blog post outline for the keyword: [KEYWORD]
Context:
- Audience: [AUDIENCE]
- Search intent: [INFORMATIONAL / COMMERCIAL / TRANSACTIONAL]
- My product angle: [PRODUCT]
- Current top 3 ranking pages (if I know): [URLS]
Give me:
1. A title, under 60 characters, keyword near the front, designed to beat the current top 3
2. A hook (first 100 words) that answers the search query in sentence 1 — no throat-clearing
3. A 5-8 section outline with H2, key point, evidence I need, and a specific example for each section
4. Word count target with reasoning
5. Your honest take: is this keyword actually rankable, or am I up against too much authority? If it's too saturated, suggest a long-tail variant I could win instead.
GEMINI
Build a blog post outline optimized for search.
TARGET KEYWORD: [KEYWORD]
AUDIENCE: [AUDIENCE]
SEARCH INTENT: [INFORMATIONAL/COMMERCIAL/TRANSACTIONAL]
PRODUCT ANGLE: [PRODUCT]
OUTPUT:
## META
- Title (≤60 chars):
- Meta description (≤155 chars):
- Word count target:
- Secondary keywords to include:
## ARTICLE HOOK
First 100 words that directly answer the search query.
## OUTLINE
| Section | H2 | Key Point | Evidence Needed | Example |
|---------|----|-----------|-----------------|---------|
| 1 | ... | ... | ... | ... |
| 2 | ... | ... | ... | ... |
## SERP ANALYSIS
- Current top 3 ranking strategies:
- Gap I can exploit:
- Ranking difficulty: [LOW/MEDIUM/HIGH]
## RANKING VIABILITY
If this keyword is too competitive for my domain authority, suggest 3 long-tail alternatives with:
- Estimated monthly volume
- Difficulty
- Commercial intent
TASK 11 / 16
Case study from a customer win
Turn a happy customer into a marketable asset.
Why they differ: Claude structures the problem/solution arc most rigorously. ChatGPT writes the most engaging prose. Gemini is best for the structured quote extraction.
CLAUDE
<task>Write a B2B case study from raw customer interview notes.</task>
<input>
[PASTE INTERVIEW NOTES OR CALL TRANSCRIPT]
</input>
<structure>
<summary>3 sentences max. Problem, solution, result.</summary>
<customer>Who they are, one sentence. Why they matter to the reader.</customer>
<situation>The problem in their words. Use direct quotes where possible.</situation>
<complication>What made it worse. What they tried that didn't work.</complication>
<resolution>What we did together. Specific actions, not "partnered with".</resolution>
<results>Quantified outcomes. If numbers weren't given, say so explicitly.</results>
<quote>Pull the best direct quote for hero placement.</quote>
</structure>
<rules>
- No "transformation journey" or "trusted partner" language
- Every claim needs evidence from the notes
- If a section lacks supporting detail, write "INTERVIEW GAP: [what's missing]"
- Under 600 words
</rules>
CHATGPT
Write a case study based on these customer interview notes.
[PASTE NOTES]
Structure:
1. TL;DR — 3 sentences (problem, solution, result)
2. Customer intro — who they are, why readers should care
3. Situation — the problem, in their own words, with direct quotes
4. Complication — what made it worse, what they tried that didn't work
5. Resolution — what we did specifically (not "partnered with")
6. Results — quantified outcomes (and flag it if numbers weren't given)
7. Hero quote — the best direct quote to feature at the top
Rules:
- No marketing language like "transformation journey" or "trusted partner"
- Every claim has to trace back to something in the notes
- If a section is thin because I didn't ask enough in the interview, call it out: "INTERVIEW GAP: [what's missing]"
- Under 600 words
GEMINI
Convert customer interview into structured case study.
SOURCE MATERIAL:
[PASTE INTERVIEW NOTES OR TRANSCRIPT]
OUTPUT SECTIONS:
## 1. TL;DR
3 sentences: problem → solution → quantified result
## 2. CUSTOMER PROFILE
| Field | Value |
|-------|-------|
| Company | ... |
| Industry | ... |
| Size | ... |
| Role interviewed | ... |
## 3. NARRATIVE
### Situation
### Complication
### Resolution
### Results
## 4. QUOTE EXTRACTION
| Quote | Use For |
|-------|---------|
| "..." | Hero header |
| "..." | Mid-body credibility |
| "..." | Close / CTA support |
## 5. EVIDENCE GAPS
List any section with insufficient source material. Mark as: "INTERVIEW GAP: [description]"
CONSTRAINTS:
- ≤600 words narrative
- Zero marketing clichés
- Every claim sourced to notes
TASK 12 / 16
Twitter/X thread from a blog post
Repurpose long content into a thread that drives traffic back.
Why they differ: ChatGPT has the best feel for Twitter voice. Claude is strongest at compression without losing the argument. Gemini structures the thread arc most cleanly.
CLAUDE
<task>Convert the blog post below into a Twitter/X thread.</task>
<input>
[PASTE BLOG POST OR URL]
</input>
<output_format>
<tweet_1>The hook. Specific claim that stops the scroll. Under 240 characters. Must work alone (most viewers only see this one).</tweet_1>
<tweets_2_to_N>The argument, compressed. One idea per tweet. Under 240 chars each.</tweets_2_to_N>
<final_tweet>Call to read the full post. Not "link in bio" — actual link + reason to click.</final_tweet>
</output_format>
<rules>
- Maximum 8 tweets
- No emojis unless the original post uses them
- No "🧵" thread emoji — it signals low effort
- No "let me tell you about"
- Preserve the original argument. Don't dilute it.
- Each tweet should be readable standalone if it goes viral mid-thread
</rules>
CHATGPT
Turn this blog post into a Twitter/X thread:
[PASTE BLOG POST]
Rules:
- Max 8 tweets
- First tweet is the hook — specific, under 240 chars, stops the scroll (it's the only one most people see)
- Middle tweets compress the actual argument, one idea per tweet, 240 chars each
- Last tweet = CTA to read the full post with the actual link + a reason to click
- No 🧵 emoji (signals low-effort)
- No "let me tell you about"
- No emojis unless the blog post has them
- Each tweet needs to work as a standalone if someone lands mid-thread
Don't dilute the argument. Compress it.
GEMINI
Convert blog post into Twitter/X thread.
SOURCE:
[PASTE BLOG POST URL OR TEXT]
OUTPUT STRUCTURE:
| # | Purpose | Tweet (≤240 chars) | Character Count |
|---|---------|-------------------|-----------------|
| 1 | HOOK (standalone) | ... | XXX |
| 2 | Setup | ... | XXX |
| 3 | Claim 1 | ... | XXX |
| 4 | Evidence 1 | ... | XXX |
| 5 | Claim 2 | ... | XXX |
| 6 | Evidence 2 | ... | XXX |
| 7 | Implication | ... | XXX |
| 8 | CTA + link | ... | XXX |
CONSTRAINTS:
- Maximum 8 tweets
- Each tweet under 240 characters
- First tweet must be fully standalone
- No 🧵 emoji
- No "let me tell you about..."
- Preserve core argument, do not dilute
- Final tweet includes actual link + click reason
Meetings, emails, docs, planning — the daily stuff where model choice has the biggest impact.
TASK 13 / 16
Meeting transcript → action items
Turn a Gong/Zoom/Fireflies transcript into a clean action brief.
Why they differ: Claude handles long transcripts most reliably and extracts ownership accurately. Gemini is best for structured output. ChatGPT writes the most readable summary.
CLAUDE
<task>Extract action items from the meeting transcript below.</task>
<input>
[PASTE TRANSCRIPT]
</input>
<output>
<summary>3 sentences max. Purpose of meeting, decision reached, biggest open question.</summary>
<decisions>
Decisions made (not "discussed" — decided). Each with who made it.
</decisions>
<actions>
| Owner | Action | Deadline | Success measure |
|-------|--------|----------|-----------------|
| ... | ... | ... | ... |
</actions>
<open_questions>
Things that came up but weren't resolved. Flag who should own resolving.
</open_questions>
<risks>
Anything I heard that concerns me about this project.
</risks>
</output>
<rules>
- If ownership was ambiguous, write "OWNER UNCLEAR - confirm with [name]"
- If a deadline wasn't stated, write "NO DEADLINE SET - recommend [suggestion]"
- Do not invent action items that weren't discussed
- Quote directly where exact wording matters (e.g., "Sarah said 'we need to ship by Friday'")
</rules>
CHATGPT
Extract action items from this meeting transcript:
[PASTE TRANSCRIPT]
I need:
1. TL;DR — 3 sentences: what was the meeting, what was decided, biggest open question
2. Decisions MADE (not discussed — actually decided) with who made each
3. Action items as a table: owner / action / deadline / success measure
4. Open questions that didn't get resolved — and who should own resolving each
5. Risks or red flags I heard
Critical:
- If an owner is unclear, write "OWNER UNCLEAR - confirm with [name]" (don't guess)
- If no deadline was stated, write "NO DEADLINE SET" and suggest one
- Don't invent action items that weren't discussed
- Use direct quotes for anything where exact wording matters
GEMINI
Parse meeting transcript into structured action brief.
TRANSCRIPT:
[PASTE]
OUTPUT:
## TL;DR
3 sentences: purpose | decision | open question
## DECISIONS MADE
| Decision | Decided By | Impact |
|----------|-----------|--------|
| ... | ... | ... |
## ACTION ITEMS
| # | Owner | Action | Deadline | Success Measure |
|---|-------|--------|----------|-----------------|
| 1 | ... | ... | ... | ... |
## OPEN QUESTIONS
| Question | Suggested Owner | Urgency |
|----------|----------------|---------|
## RISKS / RED FLAGS
- ...
RULES:
- Mark ambiguous owners as "UNCLEAR - needs confirmation"
- Mark missing deadlines as "NOT SET - suggest [date]"
- Do not fabricate items not present in transcript
- Quote verbatim when specific wording matters
TASK 14 / 16
Draft an internal announcement
Announcing a change, launch, or decision to a team or company.
Why they differ: Claude is best at managing sensitive framing. ChatGPT writes the warmest tone. Gemini structures the information architecture cleanest.
CLAUDE
<role>Internal communications expert. You write with precision but warmth.</role>
<task>Draft an internal announcement.</task>
<context>
- What's being announced: [ANNOUNCEMENT]
- Audience: [AUDIENCE]
- Sensitivity level: [ROUTINE / SIGNIFICANT / DELICATE]
- Tone: [CELEBRATORY / NEUTRAL / DIFFICULT]
- What I want them to do after reading: [DESIRED ACTION]
</context>
<structure>
<subject>Clear, specific, no buzzwords</subject>
<opener>One sentence with the news. No setup.</opener>
<context>Why this is happening. Honest, not spin.</context>
<impact>What changes for them specifically. Be concrete.</impact>
<next_steps>What they should do and when.</next_steps>
<questions>How to ask questions. Named person or channel.</questions>
</structure>
<rules>
- Under 300 words
- No "I'm excited to share" or "I'm thrilled"
- No corporate euphemisms ("restructuring" when you mean "layoffs")
- Name a specific person for questions, not "management"
- If sensitivity is DELICATE, flag any line that might read as tone-deaf and suggest alternatives
</rules>
CHATGPT
Help me draft an internal announcement.
- What I'm announcing: [ANNOUNCEMENT]
- Audience: [AUDIENCE]
- Sensitivity: [ROUTINE / SIGNIFICANT / DELICATE]
- Tone goal: [CELEBRATORY / NEUTRAL / DIFFICULT]
- What I want people to do after reading: [ACTION]
Structure:
- Subject line (specific, no buzzwords)
- One-sentence opener with the news — no windup
- Context: why this is happening (honest, not spin)
- Impact: what changes for them, concretely
- Next steps: what they should do, when
- Questions: named person or channel, not "contact management"
Rules:
- Under 300 words
- No "I'm excited to share" or "thrilled to announce"
- No euphemisms for hard news ("restructuring" = call it layoffs if that's what it is)
- If sensitivity is DELICATE, flag any line that could read as tone-deaf and offer alternatives
GEMINI
Generate an internal announcement.
INPUTS:
- Announcement: [WHAT]
- Audience: [WHO]
- Sensitivity: [ROUTINE | SIGNIFICANT | DELICATE]
- Desired tone: [CELEBRATORY | NEUTRAL | DIFFICULT]
- Desired action: [WHAT READERS SHOULD DO]
OUTPUT STRUCTURE:
### Subject
[Specific, no buzzwords]
### Body
**The news (1 sentence):**
...
**Why:**
...
**What changes for you:**
...
**Next steps:**
| Who | Action | By when |
|-----|--------|---------|
| ... | ... | ... |
**Questions:**
Contact: [NAMED PERSON], not "management"
### CONSTRAINTS
- ≤300 words
- No "excited to share", "thrilled to announce"
- No corporate euphemisms
- If sensitivity=DELICATE, output a TONE CHECK section flagging potentially tone-deaf lines with alternatives
TASK 15 / 16
Weekly planning session
Translate a messy to-do list into a prioritized week.
Why they differ: Claude is best at hard prioritization calls. ChatGPT writes the most motivating framing. Gemini structures the time blocks most cleanly.
CLAUDE
<role>Executive productivity coach. You are ruthless about prioritization.</role>
<task>Plan my week based on the inputs below.</task>
<inputs>
- My tasks: [DUMP TASK LIST]
- My calendar constraints: [MEETINGS / BLOCKS I CAN'T MOVE]
- My most important goal this week: [SINGLE PRIORITY]
- Energy pattern: when I do my best work [MORNINGS / AFTERNOONS / MIXED]
</inputs>
<output>
<one_priority>State the ONE thing that if done, would make this week a win. No hedging.</one_priority>
<schedule>
| Day | Morning | Afternoon | Notes |
|-----|---------|-----------|-------|
| Mon | ... | ... | ... |
...
</schedule>
<not_doing>
List 3-5 things from my input that I should NOT do this week. With reasoning.
</not_doing>
<risks>
What could derail the week. What to do when it happens.
</risks>
</output>
<rules>
- Schedule deep work in my peak energy window
- Batch meetings and shallow work
- No more than 2 priority tasks per day
- If my input list is too long to fit, say so and cut ruthlessly
</rules>
CHATGPT
Plan my week for me. I want a real plan, not a to-do list.
- My tasks: [DUMP TASK LIST]
- Calendar constraints: [MEETINGS / BLOCKS]
- My #1 goal this week: [SINGLE PRIORITY]
- My peak energy window: [MORNINGS / AFTERNOONS / MIXED]
Give me:
1. The ONE thing that if I do it, the week is a win. No hedging, pick one.
2. A day-by-day schedule with morning/afternoon blocks
3. A "NOT DOING" list — 3-5 things from my dump I should skip this week, with reasons
4. Risks: what could derail the plan, and what I'll do when it happens
Rules:
- Deep work goes in my peak energy window
- Batch meetings and shallow tasks
- Max 2 priority tasks per day
- If I gave you too much to fit in a week, cut it and tell me what I cut
GEMINI
Build weekly plan.
INPUTS:
- Tasks: [LIST]
- Fixed calendar: [MEETINGS/BLOCKS]
- Week's #1 goal: [GOAL]
- Peak energy: [AM/PM/MIXED]
OUTPUT:
## THE ONE THING
Single most important task this week: [...]
## WEEKLY SCHEDULE
| Day | AM Block | PM Block | Evening |
|-----|----------|----------|---------|
| Mon | Deep work: ... | Meetings/shallow: ... | ... |
| Tue | ... | ... | ... |
| Wed | ... | ... | ... |
| Thu | ... | ... | ... |
| Fri | ... | ... | ... |
## NOT DOING THIS WEEK
| Task | Reason |
|------|--------|
| ... | Deprioritized because... |
## CONTINGENCY
| Risk | Response |
|------|----------|
| ... | ... |
RULES:
- Deep work in peak-energy blocks
- Max 2 priority tasks/day
- If task list exceeds capacity, cut explicitly
TASK 16 / 16
Convert rambling notes into a doc
Brain dump → clean, shareable document.
Why they differ: Claude is best at preserving your original voice. ChatGPT produces the most engaging edit. Gemini structures the information hierarchy cleanest.
CLAUDE
<task>Transform the messy notes below into a clean, shareable document.</task>
<input>
[PASTE MESSY NOTES / BRAIN DUMP]
</input>
<preserve>
- My original voice and word choices
- The specific examples I used (even if informal)
- The order of my logic where it makes sense
</preserve>
<improve>
- Add structure (H2s, paragraph breaks, lists only where genuinely useful)
- Fix grammar and clarity
- Remove redundancy
- Tighten sentences that ramble
</improve>
<do_not>
- Add new ideas I didn't include
- Replace my plain language with jargon
- Add corporate phrases ("moving forward", "circle back")
- Start sentences with "In today's" or "In conclusion"
</do_not>
<format>
- Title (if my notes suggest one, otherwise ask)
- Summary in 2-3 sentences at the top
- Main body organized logically
- No more than 2 lists — use prose where possible
</format>
CHATGPT
I'm pasting some messy notes. Turn them into a clean, shareable doc.
[PASTE NOTES]
Your job:
- Preserve my voice and specific examples (don't corporate-ify it)
- Add structure: H2 headings, paragraph breaks, lists only when genuinely useful
- Fix grammar, tighten rambling sentences, cut redundancy
- Keep my logic in roughly the order I had it, unless reordering makes it much clearer
DON'T:
- Add ideas I didn't include
- Replace my plain words with jargon
- Add phrases like "moving forward," "circle back," "in conclusion"
- Overstructure — if it reads better as prose, keep it as prose (max 2 lists in the whole doc)
Structure:
- Title (if my notes suggest one; if not, ask me)
- 2-3 sentence summary at top
- Main body
- Any open questions I had, called out clearly
GEMINI
Transform unstructured notes into clean document.
SOURCE:
[PASTE NOTES]
PROCESSING RULES:
## PRESERVE
- Original voice and word choices
- Specific examples used
- Logical ordering when coherent
## IMPROVE
- Grammar, clarity, concision
- Structure via H2 headings
- Remove redundancy
## FORBIDDEN
- New ideas not in source
- Jargon replacing plain language
- Corporate filler ("moving forward", "circle back")
- "In today's..." or "In conclusion..."
OUTPUT STRUCTURE:
# [Title — derived from notes, or ASK USER if unclear]
## Summary
2-3 sentences capturing the core.
## [Section 1]
Prose-first. List only if genuinely parallel items.
## [Section 2]
...
## Open Questions
Any unresolved points from source marked as "UNRESOLVED: ..."
## Source Coverage
List any source content that didn't make the final doc and explain why.
Questions people ask.
Do model differences really matter that much?
For one-off tasks, less so. For repeated work, a lot. A well-tuned Claude prompt will follow your five rules every time; a ChatGPT-style prompt in Claude might follow three of them. Compound that over 100 uses and you see why the difference matters.
Why XML tags for Claude?
Claude's training leans heavily on structured prompts using XML-style tags. Anthropic's own documentation recommends this approach because the tags help Claude reliably distinguish between instructions, context, and examples. ChatGPT and Gemini don't need this, and using XML with them adds noise without benefit.
Why tables for Gemini?
Gemini outputs structured data more reliably than prose, as Google's structured output documentation explains. If you need rows and columns, comparison matrices, or strict formatting, Gemini hits the format first try where Claude and ChatGPT sometimes need reformatting.
Can I mix and match?
Yes. Some of the best workflows use ChatGPT for creative brainstorm, Claude for the final draft, and Gemini for structured output. The prompts here are starting points, not exclusive to each model.
What about Mistral, Perplexity, or local models?
Mistral behaves similarly to Gemini for structured output. Perplexity is a search wrapper, best for factual prompts with citations. Local models like Llama and Qwen work best with the ChatGPT-style prompts because they're tuned on similar data.
Official model documentation referenced
The next step
When prompts aren't enough, agents are.
50 pre-built AI agents for B2B sales. Each runs end-to-end, works across models, remembers context, and connects to your stack.
Get the Vault $99.99All Access $99.99
Legg igjen en kommentar: