| What you’re getting in this guide 30 production-tested prompts for every stage of link building outreach — from cold first contact to negotiation, follow-up, ghost recovery, and rejection handling. Each prompt is labelled with the model it works best with (Claude or ChatGPT), the specific use case, and the structural choices that make it produce usable output instead of generic AI slop. How to use this: don’t run the prompts blindly. Each one is a starting point that needs your specific context (prospect data, your voice, your offer) plugged into the bracketed placeholders. The prompts that produced the best results in real campaigns are marked with ★. Start there if you only have time to test a few. |
Why most ChatGPT outreach prompts fail
If you’ve tried to use ChatGPT or Claude for outreach and ended up with emails that sound like every other AI-generated email in the editor’s inbox, you’re not alone. Around 70% of practitioners report that their first attempts at AI outreach actually reduced reply rates compared to manual outreach. The issue isn’t the AI. It’s the prompts.
Three things separate working outreach prompts from the templated noise you’ll find on most prompt-marketplace sites. First, they force the model to use specific source material — a verbatim phrase from the prospect’s recent article, not a generic reference. Second, they ban the default AI vocabulary that editors now associate with automation. Third, they request multiple variants so you have an editorial choice rather than a single output to send as-is.
Every prompt below builds in all three. The result is output that reads like you wrote it after researching the prospect properly — because, in effect, you did. The AI just compressed the writing time.
The prompt framework every template below uses
Before the templates, the underlying structure. Every working outreach prompt has five elements. The templates below stack them in different orders for different use cases, but you’ll see the same scaffolding throughout:
| Element | Purpose | Example | ||
| 1. Role assignment | Set the model’s voice and expertise | “You are an experienced UK link builder…” | ||
| 2. Context | Specific source material for the model to draw from | Prospect’s article text, your pitch angle, prior emails | ||
| 3. Constraints | What the output must include and exclude | Word count, banned phrases, structure requirements | ||
| 4. Output specification | Exact format you want back | Three variants, table format, structured response | ||
| 5. Uncertainty handling | What to do when source material is thin | “If you don’t have enough to work with, say so” | ||
| ⚡ The single most important constraint Banning AI default vocabulary. Almost every prompt in this guide includes an explicit list of phrases to avoid: ‘I hope this email finds you well’, ‘I came across’, ‘I noticed your’, ‘great article’, ‘amazing piece’, ‘wanted to reach out’. These phrases trigger immediate AI-detection from editors. Removing them forces the model to find more genuine openings. This single constraint lifts reply rates more than any other prompt optimisation. | ||||
Category 1: Cold first-contact prompts (Prompts 1–8)
These are your bread-and-butter prompts — the ones you’ll use most often. Cold first contact is where AI compresses the most time per prospect, and where strong prompts produce the most reliable lift over generic templates.
| PROMPT 1 ★ | Best for: Claude (Sonnet/Opus) | Use case: Cold pitch to editors of mid-DR publications Guest post pitch — Backlinko-style conversational |
| Role: You are an experienced UK link builder writing in a Backlinko-style conversational voice. Context: – Prospect: [name], [role] at [publication URL] – Their recent article: [paste full article text] – My site: [your URL] – Pitch angle: [one-sentence angle] – My credentials: [one specific credibility marker] Task: Draft three subtly different guest post pitch emails. Structure for each: – Line 1: One specific reference to their article using a verbatim 4-7 word phrase – Line 2: A genuine reaction or addition — NOT generic praise – Line 3-4: The pitch in two sentences max – Line 5: Two-step CTA (ask permission to send the draft, not the link itself) Constraints: – 70-90 words total per email – British English spelling – BANNED phrases: ‘I hope this email finds you well’, ‘I came across’, ‘I noticed’, ‘great article’, ‘amazing piece’, ‘wanted to reach out’, ‘love what you’re doing’ – Subject line under 50 characters, no clickbait If the article doesn’t give you enough specific material for strong personalisation, say so before drafting. |
| PROMPT 2 | Best for: ChatGPT (GPT-4o or 4.1) | Use case: Pitching inclusion on existing curated resource lists Resource page outreach — direct and brief |
| Role: You are a link builder pitching a resource page editor. Context: – The resource page URL: [URL] – The exact section my resource would fit into: [section title] – My resource URL and one-line description: [URL + description] – Why my resource is better/different from what’s already listed: [specific reason] Task: Write a 60-word email pitching inclusion. Structure: – Sentence 1: Reference the specific section by name – Sentence 2: Briefly describe my resource and ONE concrete reason it fits – Sentence 3: Direct ask — would they consider adding it Constraints: – 60 words maximum – No filler. No ‘I hope’, no ‘I came across’, no ‘I was browsing’ – British English – End with my first name only, no signature block |
| PROMPT 3 ★ | Best for: Claude (Sonnet) | Use case: Reporting a broken link with your resource as replacement Broken link replacement pitch |
| Role: You are a helpful link builder reporting a broken link to a website editor. Context: – The page with the broken link: [URL] – The exact broken link URL: [broken URL] – The anchor text on the broken link: [anchor] – The context paragraph around the broken link: [paste paragraph] – My replacement resource URL: [URL] – Why my resource matches the original topic: [one sentence] Task: Draft two versions of an email reporting the broken link. Version A: Lead with the broken link report. Mention the replacement at the end almost as an afterthought. Version B: Lead with appreciation for the article. Report the broken link mid-email. Mention the replacement at the end. Both versions: – 80 words maximum – Genuinely helpful tone, not opportunistic – BANNED: ‘I came across’, ‘I noticed’, ‘I was reading’ – End with ‘Either way, thought you’d want to know’ or similar – British English |
| PROMPT 4 | Best for: Claude or ChatGPT | Use case: Asking a publication to convert a brand mention into a link Unlinked brand mention conversion |
| Role: You are a link builder politely asking a publication to convert an unlinked brand mention into a hyperlink. Context: – Article URL where my brand is mentioned: [URL] – The exact paragraph containing the mention: [paste] – My brand name and the URL it should link to: [brand + URL] – Date the article was published: [date] Task: Write a 70-word email requesting the link. Tone: Light, friendly, no entitlement. The publication did us a favour by mentioning us; we’re asking for a small additional favour. Structure: – Thank them genuinely for the mention with one specific detail – Ask if they’d be willing to add a link – Specify exactly which URL to link to – Close with no-pressure language Constraints: 70 words, British English, no AI default phrases. If the original mention is critical of my brand, flag that this approach won’t work and recommend a different strategy instead. |
| PROMPT 5 | Best for: Claude (Sonnet/Opus) | Use case: Pitching your improved content to sites linking to outdated versions Skyscraper pitch — your better content offer |
| Role: Link builder pitching the skyscraper technique — your better, newer content versus what they’re currently linking to. Context: – Prospect’s article URL: [URL] – The page they currently link to (the ‘old version’): [URL] – The paragraph in their article where the old link appears: [paste] – My better/newer content URL: [URL] – Three specific ways my version improves on the old: [three points] Task: Draft an 80-word email pitching the swap. Approach: Don’t trash the existing link. Acknowledge it’s a reasonable source. Position mine as a useful update or addition rather than a replacement. Structure: – Reference their article and the paragraph specifically – Note you saw they link to [old URL] there – Mention you’ve published [your URL] which adds/updates [specific improvement] – Soft ask: ‘Worth a look?’ or ‘Thought you might find it useful’ Constraints: 80 words, no AI defaults, British English, no ‘I noticed’, no ‘I came across’. |
| PROMPT 6 ★ | Best for: Claude (Opus) | Use case: Pitching original data or research to journalists Digital PR / journalist pitch — data hook |
| Role: You are a digital PR specialist pitching original data to a journalist. Context: – Journalist: [name], who covers [beat] at [publication] – One specific story they’ve recently written on a related topic: [URL + headline] – Your data/research one-line summary: [the hook] – Three most newsworthy specific findings: [3 bullet points with numbers] – Why this matters for their readers specifically: [one sentence] Task: Write three pitch variants — different angles on the same data. Variant A: Lead with the most counter-intuitive finding Variant B: Lead with the largest absolute number Variant C: Lead with the trend story (change over time) All variants: – 90 words maximum – Subject line includes specific number where possible – Offer the embargo angle: ‘happy to give you first look’ – Include a clear next step: ‘reply for the full dataset and methodology’ – BANNED: ‘fascinating’, ‘revealing’, ‘groundbreaking’, ‘cutting-edge’ – British English If the recent article you’ve been given isn’t actually on a related topic, flag this and recommend either a different journalist or a different angle. |
| PROMPT 7 | Best for: Claude (Sonnet) | Use case: Pitching yourself or a client as a podcast guest Podcast appearance pitch |
| Role: Link builder pitching a podcast appearance for [me/client]. Context: – Podcast: [name], hosted by [host], focused on [topic] – Three of their recent episodes I’ve actually listened to: [titles + one takeaway from each] – Guest credentials and angle: [credentials, plus one specific angle that hasn’t been covered on the show] – Why I’d be a good fit specifically for this show: [one sentence] Task: Write a 100-word pitch. Structure: – Reference ONE specific recent episode by name, with a genuine reaction – Propose the angle in one sentence – Provide credentials briefly – Mention the gap in their coverage that this would fill – Close with ‘happy to share more if interesting’ Constraints: 100 words, British English, no ‘I love your podcast’, no ‘I’m a huge fan’, no AI default openings. The episode reference must be specific enough that the host knows you actually listened. |
| PROMPT 8 | Best for: ChatGPT (GPT-4o) | Use case: Pitching inclusion in an existing ‘best of’ or top-X list Listicle inclusion pitch |
| Role: Link builder pitching inclusion in an existing top-X article. Context: – The listicle URL: [URL] – The listicle title (e.g., ‘Top 15 X tools for Y’): [title] – The number of items currently included: [number] – My product/service name and one-line description: [name + description] – ONE specific feature or angle that distinguishes mine from those already listed: [feature] Task: Write a 65-word email pitching inclusion. Approach: Don’t argue why an existing entry should be removed. Position mine as an addition for a gap their current list doesn’t cover. Structure: – Reference the listicle by title – Note one specific category or use case the current list doesn’t cover well – Describe how mine fills that gap in one sentence – Soft ask: ‘Worth considering for the next update?’ Constraints: 65 words, no AI defaults, British English. |
Category 2: Follow-up and ghost recovery (Prompts 9–13)
Most replies come from follow-ups, not the first message. Around 48% of senders never send a second follow-up at all, which is why these prompts are so high-leverage. The right follow-up adds value or context — the wrong one just nags and gets you blacklisted.
| PROMPT 9 ★ | Best for: Claude (Sonnet) | Use case: First gentle follow-up after no reply to initial pitch Day-5 soft nudge |
| Role: Link builder writing a soft nudge follow-up. Context: – My original email: [paste full original email] – Days since original sent: [usually 5] – Anything new since then I can reference (their recent post, news, etc.): [anything or ‘nothing’] Task: Write a 40-word follow-up email. Structure: – Brief acknowledgement that inboxes are brutal – Explicit permission to send a one-word reply (‘yes’, ‘no’, ‘maybe later’) – Statement that I’ll take silence as a no and won’t pester them further Tone: Genuinely respectful of their time. No guilt-tripping. No ‘just bumping this up’. Constraints: 40 words, British English, BANNED: ‘just following up’, ‘bumping this’, ‘circling back’, ‘in case this got lost’. |
| PROMPT 10 | Best for: Claude (Opus) | Use case: Second follow-up that brings something new Day-12 value-add follow-up |
| Role: Link builder writing a value-add follow-up. Context: – My original pitch: [paste] – My day-5 nudge: [paste] – ONE genuinely new piece of information I can share (a new data point, a recent development, a relevant article): [the new thing] Task: Write a 60-word follow-up. Approach: Don’t ask for anything in this email. Share the new piece of information as genuinely useful. Mention the original pitch as an afterthought at the end. Structure: – Open with the new piece of value, framed as something I thought they’d find useful – One-sentence reason it connects to their work – Brief one-line reminder of the original pitch + ‘happy to drop it if not a fit’ Constraints: 60 words, British English. The new piece of information MUST be genuinely useful — if I haven’t given you something substantive, refuse to draft the email and tell me to find a real value-add first. |
| PROMPT 11 ★ | Best for: ChatGPT (GPT-4o) | Use case: Switching to LinkedIn after email goes unanswered Channel-switch LinkedIn DM after email silence |
| Role: Link builder writing a LinkedIn DM after email silence. Context: – The original email I sent (assuming the prospect saw at least the subject): [paste subject line + first sentence] – Prospect’s recent LinkedIn activity (post, article, comment): [paste or describe] – Our LinkedIn connection status: [connected/not connected] Task: Write a 65-word LinkedIn DM. Structure: – Acknowledge it’s a channel switch — don’t pretend it’s first contact – Reference their recent LinkedIn activity specifically – Brief restatement of the offer (different words from the email) – Easy ask: ‘a one-word reply works’ Constraints: 65 words max (LinkedIn DM limit considerations), no AI defaults, British English. Do NOT pretend you haven’t sent an email. |
| PROMPT 12 | Best for: Claude (Sonnet) | Use case: Final touch that often recovers ghosted prospects Day-21 rejection-then-retreat goodbye |
| Role: Link builder writing a rejection-then-retreat final touch. Context: – Original pitch ask: [the big ask, e.g., guest post acceptance] – Smaller alternative ask I’d accept: [choose one: LinkedIn connection / referral to right person / feedback on the pitch] Task: Write a 50-word final email. Structure: – Brief acknowledgement that I’ve tried a couple of times – Concession on the original ask (‘I get this isn’t a fit / a priority’) – The smaller alternative ask — explicitly smaller and easier to say yes to – Polite close, no guilt Constraints: 50 words, British English. BANNED: ‘just one last try’, ‘final attempt’, anything passive-aggressive. Genuine warmth, no manipulation. |
| PROMPT 13 | Best for: Claude (Opus) | Use case: Restarting a cold relationship when prospect has a fresh trigger event 60-day trigger-event re-engagement |
| Role: Link builder restarting a cold relationship. Context: – Original outreach: [date and brief pitch] – Trigger event since (new article, promotion, podcast, award, funding): [specific event with date] – Why my original pitch is more relevant now than then: [one sentence] Task: Write an 80-word fresh email — treat this as a new conversation, not a continuation. Structure: – Lead with congratulations or specific reaction to the trigger event – Briefly note we exchanged emails [X months] ago — light reference – Refresh the pitch with the new relevance angle – Clear easy CTA Constraints: 80 words, no AI defaults, British English. If the trigger event doesn’t actually make the pitch more relevant, refuse to draft and tell me to wait for a better hook. |
Category 3: Reply handling — turning no into yes (Prompts 14–19)
Around 10–15% of polite-pass rejections can be converted to placements within twelve months with the right reply. These prompts give you the language to do it without sounding desperate, pushy, or pre-written.
| PROMPT 14 ★ | Best for: Claude (Sonnet) | Use case: Graceful response to ‘we don’t accept guest posts’ Reply to a ‘polite pass’ rejection |
| Role: Link builder responding to a polite pass rejection. Context: – The exact rejection email I received: [paste] – One specific recent thing the prospect has worked on or published: [specific reference] – Realistic timeframe to come back: [3 months / 6 months] Task: Write a 60-word reply. Goal: Keep the door open for future without being pushy. Structure: – Genuine thank for the courtesy reply (most rejecters don’t bother) – Ask permission to check back in [timeframe] if their policy changes – Close with a specific reference to their recent work Constraints: 60 words, British English. BANNED: ‘totally understand’ (overused), ‘no worries’, any wording that hints at disappointment. |
| PROMPT 15 | Best for: ChatGPT (GPT-4o) | Use case: Declining paid placement while preserving the relationship Reply to a ‘we charge for placements’ response |
| Role: Link builder declining a request for paid placement. Context: – Their reply mentioning their rate: [paste] – My honest reason for declining: [no budget / against editorial policy / etc.] – An alternative value I can offer (expert commentary, data contribution, future contributor): [one specific option] Task: Write a 60-word reply. Goal: Decline without burning the bridge. Position myself as a future zero-cost contributor. Structure: – Thank for transparency on the rate – Decline directly, briefly state why – Offer the alternative value – Wish them well Constraints: 60 words, British English. No haggling. No ‘is that negotiable’. Clean decline. |
| PROMPT 16 ★ | Best for: Claude (Sonnet) | Use case: Re-pitching to the correct contact with referral context Reply to a ‘wrong person’ redirect |
| Role: Link builder pivoting to a referred contact. Context: – The original contact’s reply that referred me: [paste] – The new contact’s name and role: [name + role] – The new contact’s email or LinkedIn: [contact] – My original pitch: [paste] Task: Write TWO emails. Email 1 (60 words): Reply to original contact thanking them for the referral, ask permission to mention their name in the outreach to the new person. Email 2 (90 words): The new pitch to the referred contact. Lead with the referral. Then re-state the pitch. End with a clear easy CTA. Constraints for both: British English, no AI defaults. The new pitch should feel warm — this is a referred conversation, not cold outreach. |
| PROMPT 17 | Best for: Claude (Sonnet) | Use case: Quality pushback — preserving long-term relationship Reply to ‘your DR is too low’ |
| Role: Link builder responding to a DR/quality rejection. Context: – Their rejection email referencing my domain quality: [paste] – My current site’s DR: [DR] – My genuine plan to improve over next 12 months: [brief plan or ‘still working on it’] Task: Write a 55-word reply. Goal: Don’t argue the DR. Acknowledge gracefully. Ask what threshold would work so I can come back when ready. Structure: – Accept the feedback as fair – State briefly that I’m working on building up – Ask what DR/quality threshold they’d want to see – Close with ‘happy to check back in when we’re there’ Constraints: 55 words, British English. NO defending the current DR. NO ‘I think you might be wrong about’. Zero ego. |
| PROMPT 18 | Best for: Claude or ChatGPT | Use case: Pivoting to a new angle after the first one was rejected Reply to ‘topic mismatch’ — angle pivot |
| Role: Link builder pivoting on angle after topic rejection. Context: – Their rejection email: [paste] – My original pitch angle: [paste] – TWO alternative angles I could pitch on the same topic: [two new angles] – Their recent published content that hints at what they DO want: [URL or paste] Task: Write a 70-word reply. Approach: Don’t argue the original was a good fit. Acknowledge the miss. Float ONE of the two new angles — pick the one that better matches their recent content. Structure: – Acknowledge the miss with grace – One sentence on why I think the angle was off (showing self-awareness) – Float the better-fitting new angle in one sentence – Ask if it sounds more interesting before drafting Constraints: 70 words, British English. Pick the stronger of the two angles based on the recent content provided. If neither is a great fit, recommend I walk away from this prospect instead. |
| PROMPT 19 ★ | Best for: Claude (Opus) | Use case: Confirming and setting expectations after a yes Reply to a positive ‘send me the draft’ yes |
| Role: Link builder responding to a positive reply requesting the draft. Context: – Their positive reply: [paste] – Pitch angle I committed to: [angle] – Realistic timeline I can deliver in: [days] – Any specific guidelines they have published or mentioned: [paste] Task: Write a 70-word confirmation email. Goals: – Confirm the angle – Set the delivery timeline explicitly – Confirm I’ll follow their guidelines – Ask any one critical question I need answered before drafting (length, specific examples to include, deadlines) Structure: – Thank them with one sentence – Confirm angle in one sentence – State delivery timeline – The one question (only if essential) – Close with ‘will get it across by [date]’ Constraints: 70 words, British English. Professional, brief, clear. No excessive gratitude. |
Category 4: Negotiation and editorial back-and-forth (Prompts 20–23)
These prompts handle the awkward middle-game: when an editor wants changes, when they push back on links, when they ask for shorter copy, or when they want exclusivity. Most link builders handle these moments badly because there’s no clear template. These prompts give you a starting structure.
| PROMPT 20 | Best for: Claude (Sonnet) | Use case: Responding professionally to revision requests Editor asks for changes to draft |
| Role: Guest contributor responding to editorial revision requests. Context: – My original draft: [paste relevant excerpt] – Their revision requests: [paste] – Which requests I’ll accept: [list] – Which requests I’d push back on and why: [list with reasons] Task: Write a 90-word reply. Tone: Collaborative, not defensive. Treat their feedback as making the piece better, even when I disagree. Structure: – Brief thanks for the detailed feedback – Confirm acceptance of the changes I agree with – Politely raise the one or two I’d push back on, with reasoning – Offer a third option where possible – Confirm next steps and timing Constraints: 90 words, British English. No ‘with respect, I disagree’ — too defensive. Lead with collaboration. |
| PROMPT 21 | Best for: Claude (Opus) | Use case: Negotiating link retention when editor wants to strip it Editor wants to remove the link |
| Role: Guest contributor whose link is being removed. Context: – Editor’s request to remove the link: [paste] – Where the link sits in the draft: [paragraph with context] – Why the link is genuinely useful to readers: [one specific reason] – The alternative — would I accept the piece without the link: [yes/no] Task: Write a 70-word reply. Approach: Don’t fight for the link in a way that signals I’m only there for the link. Frame it as reader value. Structure: – Acknowledge their editorial judgement – One genuine reader-value reason the link helps – Offer a compromise (different anchor, different position, link to a more editorial page rather than commercial) – State explicitly whether I’d still want the piece published if the link must come out Constraints: 70 words, British English. The ‘would I publish without the link’ answer must be honest — if I wouldn’t, say so respectfully. |
| PROMPT 22 | Best for: ChatGPT (GPT-4o) | Use case: Handling exclusivity asks on data or research Editor offers exclusivity in exchange for placement |
| Role: Link builder negotiating exclusivity terms on a data story. Context: – The exclusivity ask from the editor: [paste] – The exclusivity duration they want: [days/weeks] – My realistic constraint (other pitches already sent, conference timing, etc.): [constraint] – My ideal counter-offer: [counter] Task: Write a 75-word reply. Approach: Exclusivity is normal in digital PR. Don’t refuse outright. Negotiate the terms. Structure: – Acknowledge the value of exclusivity for them – State my constraint honestly – Counter-offer (shorter window, embargo period, exclusive first publish but I can repurpose after) – Ask for their reaction Constraints: 75 words, British English. Be transparent about other pitches if they exist — getting caught lying about exclusivity terminates the relationship permanently. |
| PROMPT 23 | Best for: Claude (Sonnet) | Use case: Re-engaging an editor who accepted then went silent Editor goes quiet mid-process |
| Role: Guest contributor re-engaging a silent editor mid-process. Context: – Original acceptance email from editor: [paste subject + key sentence] – Days since their last reply: [number] – Where we are in the process: [draft sent / waiting for revisions / publish date pending] – Any external context (their publication going through changes, recent news affecting them): [anything] Task: Write a 55-word check-in. Tone: Patient, no pressure. Editors go quiet for legitimate reasons. Structure: – Acknowledge inboxes are brutal – One-line status check — where are we – Offer flexibility (‘happy to push the timeline’) – Close with ‘no urgency on your end’ Constraints: 55 words, British English. No guilt-tripping, no ‘just bumping this up’. |
Category 5: Workflow helpers — research and analysis (Prompts 24–28)
These prompts aren’t for writing outreach directly. They’re for the supporting work that makes outreach better: extracting hooks from articles, classifying replies, drafting alternative subject lines, and analysing campaign data.
| PROMPT 24 ★ | Best for: Claude (Opus) | Use case: Stage 4 of the workflow — finding genuine hooks to reference Extract personalisation hooks from an article |
| Role: Outreach researcher extracting personalisation hooks. Context: – Article URL: [URL] – Full article text: [paste] – The type of pitch I’m preparing: [guest post / unlinked mention / digital PR / etc.] Task: Extract FOUR usable hooks from the article. Hook 1: A verbatim 4-7 word phrase the author writes that captures a specific argument or position Hook 2: A specific data point, statistic, or claim the author cites Hook 3: A potential gap, counterpoint, or unaddressed question the article leaves open Hook 4: Something distinctive about the author’s voice or framing that I could mirror For each, give: – The exact text or reference – A one-sentence note on how to use it in outreach – A confidence rating 1-5 on whether this is strong personalisation material If overall confidence is below 3, recommend I pitch a different article from this prospect instead. |
| PROMPT 25 | Best for: Claude (Sonnet) or ChatGPT (GPT-4o) | Use case: Automating reply classification for triage Classify an incoming reply |
| Role: Reply classifier for link building outreach. Context: Incoming reply to my outreach. Reply text: [paste] Task: Classify the reply into ONE of the following categories: – INTERESTED: Positive response, wants to see more – POLITE_PASS: Polite rejection, no specific reason – BUDGET_BLOCK: They want payment – WRONG_PERSON: They’ve redirected to another contact – QUALITY_PUSHBACK: They’ve rejected based on my domain/site quality – TOPIC_MISMATCH: They’ve rejected based on the topic angle – HARD_STOP: Explicit opt-out, do not re-contact – NEEDS_INFO: They’ve asked a clarifying question before committing – OUT_OF_OFFICE: Auto-responder – AMBIGUOUS: Cannot confidently classify Output: – Classification: [one of the above] – Confidence: 1-5 – Recommended next action: one sentence – Key phrase from the reply that drove the classification: verbatim quote |
| PROMPT 26 | Best for: ChatGPT (GPT-4o) | Use case: Producing testable subject line variants Generate subject line variants for A/B testing |
| Role: Email subject line writer for cold outreach. Context: – The first sentence of my email: [paste] – My pitch angle: [angle] – Prospect’s role: [role] – Constraint: subject line must be under 50 characters to render fully on mobile Task: Generate 8 subject line variants across 4 distinct styles: Style A (2 variants): Question that hints at the pitch — engagement bait Style B (2 variants): Specific reference to their work — recognition hook Style C (2 variants): Data-driven — includes a specific number Style D (2 variants): Conversational — sounds like a colleague For each, give: – The subject line – Character count – One-sentence reason why this might work BANNED: ‘Quick question’, ‘Touching base’, ‘Following up’, ‘Hey [name]’, any emoji. |
| PROMPT 27 | Best for: Claude (Opus) | Use case: Weekly campaign analysis to inform pitch improvements Analyse a batch of replies for patterns |
| Role: Outreach campaign analyst. Context: – Total emails sent this week: [number] – Number of replies received: [number] – The full text of every reply received: [paste each, separated by ‘—‘] Task: Identify patterns across these replies. Analyse: 1. Most common rejection reason 2. Specific phrasing in MY pitch that appears to be triggering pushback (look for replies that quote or reference my email) 3. Common positive signals — what worked in the pitches that got positive replies 4. Any patterns by prospect type (publication category, role, DR band) 5. Two specific changes I could make to next week’s pitches based on the pattern Be specific. Reference exact reply quotes where possible. If the data is too thin for confident analysis (under 8 replies), say so and recommend I wait for more data. |
| PROMPT 28 | Best for: Claude (Opus) | Use case: Generating a complete 5-email sequence for a campaign Draft a follow-up sequence from scratch |
| Role: Outreach sequence designer. Context: – Campaign objective: [the link type I’m targeting] – Prospect tier: [1, 2, or 3] – My differentiator vs competitors pitching the same prospects: [one specific differentiator] – Campaign-specific constraints (no negotiation, no client allowed, etc.): [any] Task: Design a 5-touch sequence across 21 days. For each touch: – Day to send – Channel (email or LinkedIn) – One-sentence purpose for that touch – Word count target – The ‘angle shift’ — how this touch differs from the previous ones Output as a table. Include rationale for why this sequence shape fits the campaign objective and tier. If the campaign objective and tier don’t match well (e.g., tier 3 prospect with high-effort campaign), flag the mismatch and recommend adjustment. |
Category 6: Meta-prompts — refining your own prompts (Prompts 29–30)
These last two are the most important if you’re going to use AI outreach long-term. They help you improve the prompts you’re using rather than relying on static templates forever.
| PROMPT 29 ★ | Best for: Claude (Opus) | Use case: Improving a prompt that isn’t producing strong output Audit my existing outreach prompt |
| Role: Prompt engineer specialising in outreach copy. Context: – My current prompt I want you to audit: [paste full prompt] – A sample of output it’s producing: [paste 2-3 example outputs] – My specific concern (output sounds generic / too long / too short / wrong tone): [concern] Task: Audit the prompt and produce an improved version. Analyse: 1. Which of the 5 framework elements (role, context, constraints, output spec, uncertainty handling) is the prompt weakest on? 2. Which specific words or phrases in the prompt are causing the issue? 3. What specific constraints are missing that would force better output? Then produce: – A rewritten prompt with changes highlighted in [brackets] – A brief explanation of why each change should improve output – One test case I could run to verify the improvement |
| PROMPT 30 ★ | Best for: Claude (Opus) | Use case: Generating a fresh prompt for a use case not covered above Build a custom prompt for a new situation |
| Role: Prompt engineer building a new outreach prompt. Context: – The new outreach situation: [describe in detail] – The desired output type (email / DM / pitch deck text / etc.): [type] – Word count target: [number] – The voice I want (Backlinko conversational / Moz formal / Ahrefs data-led / etc.): [voice] – Specific phrases or styles to AVOID: [any] – An example of what ‘good’ looks like for similar situations: [paste example if available] Task: Build a new prompt for me to use in this situation. Include all 5 framework elements: 1. Role assignment (specific, not generic) 2. Context section with bracketed placeholders for the variable inputs 3. Explicit constraints including word count, banned phrases, and structural requirements 4. Output specification (single output or multiple variants?) 5. Uncertainty handling — what should the model do if the inputs are too thin? After building the prompt, also include: – One test scenario I could run to check it works – Likely failure modes and how to spot them in output |
Quick-reference table
Bookmark this. When you need a prompt fast, find the situation in the left column and jump straight to the prompt number.
| Situation | Prompt # | Best model |
| Cold first pitch to publication editor | Prompt 1 ★ | Claude |
| Resource page inclusion | Prompt 2 | ChatGPT |
| Broken link replacement | Prompt 3 ★ | Claude |
| Unlinked mention conversion | Prompt 4 | Either |
| Skyscraper-style content pitch | Prompt 5 | Claude |
| Digital PR / journalist data pitch | Prompt 6 ★ | Claude Opus |
| Podcast guest pitch | Prompt 7 | Claude |
| Listicle inclusion | Prompt 8 | ChatGPT |
| Day 5 soft follow-up | Prompt 9 ★ | Claude |
| Day 12 value-add follow-up | Prompt 10 | Claude Opus |
| LinkedIn channel switch | Prompt 11 ★ | ChatGPT |
| Rejection-then-retreat goodbye | Prompt 12 | Claude |
| 60-day trigger event re-engagement | Prompt 13 | Claude Opus |
| Reply to polite pass | Prompt 14 ★ | Claude |
| Reply to paid placement request | Prompt 15 | ChatGPT |
| Reply to wrong person redirect | Prompt 16 ★ | Claude |
| Reply to DR quality pushback | Prompt 17 | Claude |
| Reply to topic mismatch — pivot angle | Prompt 18 | Either |
| Confirm a positive yes | Prompt 19 ★ | Claude Opus |
| Respond to editorial revision requests | Prompt 20 | Claude |
| Negotiate link retention | Prompt 21 | Claude Opus |
| Handle exclusivity ask on data | Prompt 22 | ChatGPT |
| Re-engage silent editor mid-process | Prompt 23 | Claude |
| Extract personalisation hooks from article | Prompt 24 ★ | Claude Opus |
| Classify incoming reply | Prompt 25 | Either |
| Generate subject line variants | Prompt 26 | ChatGPT |
| Weekly pattern analysis on replies | Prompt 27 | Claude Opus |
| Design a 5-touch sequence | Prompt 28 | Claude Opus |
| Audit and improve an existing prompt | Prompt 29 ★ | Claude Opus |
| Build a brand-new custom prompt | Prompt 30 ★ | Claude Opus |
Frequently asked questions
Which model should I use — ChatGPT or Claude?
Both work for most of these prompts, but the labels above reflect real preferences. Claude tends to produce more natural-sounding personalised copy and follows complex constraint lists more reliably; ChatGPT tends to produce stronger structured outputs and is faster for high-volume generation. The practical answer: pick one paid tier (£15–£20/month), become fluent at it, and don’t split attention between models. The marginal gain from using both rarely justifies the cognitive overhead of switching contexts.
Why so many banned phrases?
Editors have been receiving thousands of AI-generated outreach emails since 2024. A small set of phrases — ‘I hope this email finds you well’, ‘I came across’, ‘I noticed your’, ‘great article’ — have become reliable AI-detection markers. The moment one appears in your email, the editor mentally files you as automated outreach and your reply rate drops by half. Banning these phrases at the prompt level forces the model to find more genuine openings. This single optimisation matters more than any other prompt tweak.
Should I edit the output before sending?
Always. Every prompt above produces a first draft, not a send-ready email. A 30–60 second human edit catches the 5–10% of outputs that contain subtle hallucinations (a wrong fact about the prospect), miscalibrated tone, or phrasing that doesn’t sound like you. Skipping the edit is the single most common cause of AI-assisted campaigns underperforming traditional ones. The time savings only compound if quality is maintained.
Do these prompts work for non-English outreach?
Partially. The frameworks transfer to most major European languages — the role/context/constraints/output/uncertainty structure is language-agnostic. The banned phrases obviously need translating to the equivalent overused phrases in your target language, which varies by market. For non-English campaigns, the personalisation hooks (prompt 24) still work well, but the actual outreach drafting prompts (prompts 1–8) usually benefit from being rebuilt with native-speaker review rather than translated from English versions.
How often should I refresh my prompts?
Quarterly review is the practical rhythm. AI models update meaningfully every 3–6 months, and prompts that worked perfectly with one model version can produce subtly worse output after a major update. Use Prompt 29 (the audit prompt) every quarter against your most-used 3–5 prompts. Most refreshes are minor — adjusting a banned-phrase list or tightening a constraint — but skipping the review entirely is how slow performance decay creeps in.
What’s the most important prompt in this list?
Prompt 24 (extract personalisation hooks). Every outreach prompt depends on having strong personalisation material to feed it. Get prompt 24 working well and prompts 1–8 dramatically improve. Skip the personalisation stage and even the best drafting prompt produces generic output. If you only build one prompt into your workflow, build that one.
Can I share or sell these prompts?
These prompts are published under standard editorial copyright. You’re welcome to use them in your own campaigns, adapt them, and build them into internal tooling. We ask that you don’t republish them verbatim on other public sites or sell them as part of a paid prompt pack — that crosses the line from fair use to commercial republishing. For the broader strategy that these prompts fit into, see our AI-assisted link building workflow guide (article 154) and our link building strategies hub.
What if I’m not seeing reply-rate improvements?
Three diagnostic checks. First, audit the personalisation — are you actually feeding the model specific source material from each prospect, or just running prompts with the placeholder defaults? Second, check whether you’re editing the output, or sending raw model output (the latter typically underperforms manual outreach). Third, examine your prospect list quality — strong prompts produce great output on weak prospects, but weak prospects don’t reply regardless of how strong the email is. The most common cause of disappointing results is upstream from the prompt, not in the prompt itself.
Putting it all together
These thirty prompts are a starting library, not a finished system. The practitioners who get the most out of AI-assisted outreach treat the prompts as living documents — refining them based on what their actual replies show, building custom prompts for situations not covered, and retiring prompts that stop producing strong output. The static template approach (paste prompt, send result) underperforms manual outreach almost every time.
Start with the starred prompts (1, 3, 6, 9, 11, 14, 16, 19, 24, 29, 30). They cover the highest-leverage situations and the meta-tools you’ll need to evolve your prompt library over time. Run them for a single campaign of fifty prospects, edit every output carefully, and measure your reply rate against your previous baseline. The data from that single campaign will tell you which prompts work for your specific voice and prospect mix, and where the next round of refinement should focus. For the broader workflow these prompts fit into, our AI-assisted link building workflow guide walks through the full seven-stage system. For the broader outreach foundations, see our link building strategies hub and best link building tools guide.
