Jonathon's AI Wiki

The LinkedIn Engagement Machine — Transcript-to-Post AI Workflow (Notion AI Recipe)

10 min read

Source: ai-research/notion-linkedin-engagement-machine-2026-05-02.md (Notion AI Recipe at notion.so/jonathonc/The-LinkedIn-Engagement-Machine-fc325679607683a9bd5d014477c930d8, fetched via claude_ai_Notion MCP 2026-05-02; recipe dated 2026-01-08; 15-30 minute session). Parent vault page: AI Recipe Vault.

A four-prompt-chain LinkedIn content workflow that takes a raw video transcript or voice-note brain dump (500+ words) and produces a polished LinkedIn post with a strategic lead-magnet hook in 15-30 minutes. Each prompt assumes the Justin-Welsh persona (the $5M solopreneur LinkedIn business framing) and runs in the same conversation thread so the AI keeps your transcript context across all three steps. An optional fifth prompt builds the lead magnet itself. Source claims: 3-5x more comments than generic AI posts and 20-50+ lead-magnet requests per post.

Key Takeaways

  • Same-thread, three-prompt chain is the architectural choice. Step 1 (Ideation Engine) extracts 3-5 post concepts from your transcript. Step 2 (Writing System) drafts the full post for a chosen concept. Step 3 (Polish Filter) scores it /10, lists fixes, and rewrites brutally. Critical: all three run in the SAME chat — restarting between steps loses transcript context and produces “generic AI slop” per the source.
  • The four prompts are the artifact. Each one is ~50-200 lines, copy-paste, parameterized only by [PASTE TRANSCRIPT] / [PASTE CONCEPT] / [PASTE DRAFT]. No no-code platform, no skill, no MCP — pure prompt chain runnable in claude.ai or ChatGPT paid. The full text is captured in the ai-research source.
  • Justin Welsh persona is load-bearing. Every prompt opens by anchoring the model to his style: short punchy sentences, parenthetical proof points (saved $2,400/month), pattern interrupts (One word. Boom.), the reframe (“Most people [common behavior]. I [contrarian approach].”), confessions (“Here’s something unhinged: I tested everything”). The persona constrains output away from corporate-AI voice without you having to spell out negatives.
  • Lead-magnet hook is the engagement-driver. Every post promises a deliverable unlocked by commenting a keyword (checklist, template, framework, swipe file, tool stack). Source frames this as the comment-engine — the keyword comment ritual is what mechanically multiplies engagement (3-5x baseline), and the DM-the-magnet workflow is what converts engagement to qualified inbound.
  • The optional Step 5 builds the lead magnet itself. A fourth prompt — Lead Magnet Builder — uses the same conversation context to produce a structured outline (cover + intro + 3-5 actionable sections + quick-start + what’s-next) from the lead-magnet idea identified in Step 1. Output formats as a PDF / Notion template / Canva guide. This closes the loop so you can run the workflow end-to-end without leaving the chat.
  • Repurposing playbook is built in. A polished LinkedIn post becomes a Twitter thread (break at pattern interrupts), an email newsletter (expand the story section), and an Instagram carousel (one slide per framework bullet). Single transcript → multi-platform asset pipeline.
  • Batch-the-ideation tip is the volume play. Run Step 1 against 5-10 transcripts at once to bank 15-20 post concepts. Then write and schedule 2-3 per week. Source frames this as how the recipe scales from “one polished post” to “consistent weekly content with breathing room.”

How It Works

StepActionTypeTime
1Extract 3-5 post concepts from transcriptPrompt 1 — Ideation Engine~5-10 min
2Write full post from chosen conceptPrompt 2 — Writing System~5-10 min
3Polish with ruthless editing filterPrompt 3 — Polish Filter + review~5-10 min
4Final tweaks and publishManual deploy + comment monitoring~2-3 min
5 (optional)Build the promised lead magnetPrompt 4 — Lead Magnet Builder~5-10 min

Total: 15-30 minutes if you skip Step 5; 25-40 minutes including the lead-magnet build.

The full text of all four prompts is in the ai-research source file — the wiki’s not the place to reproduce 600 lines of prompt body verbatim. Read once, save the prompts to a Claude Project called “LinkedIn Engagement Machine” or to a Claude skill, and stop copy-pasting from Notion.

What Each Prompt Adds

Prompt 1 — Ideation Engine. Four-stage thinking (Identify the Gold → Frame Each Idea → Structure the Concept → Validate). For each concept it returns: hook angle, why-it-works (psychology), core transformation, key proof points, full post structure outline, lead-magnet name + components, engagement-driver CTA template, and a self-validation pass. The “Identify the Gold” stage looks specifically for: surprising stats with specific numbers, counterintuitive insights, before/after with timestamps or dollar amounts, common-struggles-with-uncommon-solutions, and “I wish someone told me this sooner” moments.

Prompt 2 — Writing System. Hook (3 lines, immediate tension, specific numbers, “wait what?” moment). Story (3-6 short paragraphs pulled from transcript, max 3 lines each, rhythm: short-punch → longer-explain → short-punch). Value framework with options (time-stamped workflow / emoji bullets / numbered steps / before-after) — each item must include a parenthetical proof. Reframe (1-3 lines before CTA, philosophy-shift). CTA (4-part structure: transition question → comment-word action → deliverable name → mild FOMO). Five “Signature Techniques” enforced: parenthetical proof, pattern interrupt, sensory detail (47 browser tabs open not it's overwhelming), confession, timestamp credibility.

Prompt 3 — Polish Filter. Six-stage editing pass (Hook Analysis → Story & Flow → Proof Points → Structure → “So What?” Test → CTA Polish). Outputs: a /10 score with strongest-element callout, a brutal list of specific fixes, the fully edited version, and an explanation of the three biggest changes. Red-flag list to remove: I think / I believe, corporate jargon, explained jokes, more than 2 CTAs, lists without proof points, vague time references (recently → last Tuesday), weak verbs (utilize → use).

Prompt 4 — Lead Magnet Builder (optional). Five-section output structure (Cover + Introduction + 3-5 main sections + Quick-Start + What’s Next). Style anchor: “Be me. Use my words, my tone, my rhythm.” References the original transcript again so the magnet voice matches the post voice.

Why This Pattern Generalizes

  • The same-thread context-retention trick is the implementation pattern behind every recipe in the AI Recipe Vault — see the sister Cowork + Apify scraping recipe for an even more elaborate version (six steps of compounding context).
  • The persona-anchor (you-are-X-with-Y-philosophy) is the cheap version of domain-specific intelligence from Claude skills — an in-prompt mini-skill instead of a packaged one. Promotes well to a real Claude skill once the prompts stabilize (see Intermediate Course Module 1 on prompts-as-reusable-artifacts and Module 2 on shop-vet-build for skills).
  • The CTA-keyword-to-DM-the-magnet pattern is the same engagement engine used in Brandon Storey’s lead-magnet course (via ManyChat for Instagram, manual for LinkedIn). Both rely on a deliverable that’s actually useful — bad lead magnets kill trust faster than no lead magnet.
  • The Justin Welsh framing is one personality choice, not the template. Plenty of LinkedIn voice templates — Charlie Hills’ voice-builder generates a personalized one from scratch and uses it across 16 skills. If you want this recipe but in your voice, swap the Welsh persona block for your own voice.md.

Caveats

  • Justin-Welsh-style ≠ every operator’s voice. The recipe enforces a specific tone (parenthetical proofs, pattern interrupts, confessions). If your professional voice is more measured / technical / formal, you’ll fight the prompts. Either swap the persona block (and lose some engagement-pattern guarantees) or build a personalized version with voice-builder from Charlie Hills’ bundle.
  • Stat claims are not citation-grade. “$5M solopreneur business” / “3-5x more comments” / “20-50+ lead magnet requests” — Welsh’s number is publicly attributed; the engagement multipliers are the recipe author’s claim, not a controlled study. Treat as a credible-sounding target, not a benchmark.
  • The recipe is in jonathonc’s personal Notion, “AI Recipe” type. Like the Cowork+Apify recipe, origin is most likely a curated/duplicated AI-Recipe template rather than an Anthropic-published doc. Treat as a starting framework, not a vendor-sanctioned best practice.
  • LinkedIn algorithm sensitivity. LinkedIn periodically penalizes “engagement bait” patterns — Comment KEYWORD to get the magnet is exactly the comment-bait pattern they have flagged in the past. Watch for reach degradation; if it lands, switch to “DM me KEYWORD” or “Link in comments” variants.

Try It

  1. Run it once on real transcript. Pick a 5-10 minute video you’ve already filmed (loom, YouTube, Fathom call). Pull the transcript. Open a fresh claude.ai or ChatGPT chat. Paste Prompt 1 + transcript. Pick a concept. Paste Prompt 2 with the chosen concept. Paste Prompt 3 with the draft. Publish. Notice what feels true vs what feels Welshified.
  2. Promote to a Claude Project or skill. Once you’ve run it 2-3 times and know what to tweak, save the four prompts to a Claude Project named LinkedIn Engagement Machine (per the source’s pro-tip) — or wrap them as a single Claude skill with progress cues per the sequential-workflow pattern. Either way, stop copying from Notion every run.
  3. Voice-swap experiment. Replace the Justin Welsh persona block in all three prompts with a personalized voice.md generated by Charlie Hills’ voice-builder skill. Run the chain. Compare engagement on the Welsh version vs the personalized version over 4 posts each.
  4. Batch-bank concepts. Run Step 1 (only) against 5-10 transcripts in one session. Bank 15-20 post concepts in a Notion / Google Doc. Then use Steps 2-3 once a week to draft and ship 2-3 polished posts. This is how the recipe converts from “one post” to “weekly content engine.”
  5. Build the lead magnet first. Counter-intuitive: run Prompt 4 first to define the lead-magnet outline, then Prompts 1-3 with the magnet promise already specific. Tightens the CTA and prevents “I’ll figure out the magnet later” vapor that kills deliverability when the comments come in.
  6. Pair with a content-distribution rail. Postiz (29.2k-star OSS scheduler) handles repurposing the polished LinkedIn post into Twitter thread, IG carousel, etc. — closes the “single transcript → multi-platform” loop the recipe gestures at.

Open Questions

  • Provenance of the persona claims. Justin Welsh is a real LinkedIn creator and his $5M solopreneur business is publicly attributed. The 3-5x comments / 20-50+ lead-magnet requests numbers are the recipe author’s claim — uncited. Worth searching Welsh’s own writing for whether he sanctions this prompt-chain framing or whether the recipe is a third-party “use his style” template.
  • LinkedIn algorithm risk on the keyword-comment ritual. LinkedIn has historically de-ranked engagement-bait patterns. Track post-by-post reach to detect drag. If it appears, swap to “Link in comments” or “DM me KEYWORD” variants.
  • Voice-portability test. Does swapping the Welsh persona block for a voice.md from Charlie Hills’ voice-builder preserve engagement, or does the Welsh-specific rhythm carry meaningful uplift on its own? Worth a 4-vs-4 A/B.
  • Skill promotion ceiling. What’s the right Claude-skill abstraction here — one big sequential-workflow skill that runs all four prompts, or four discrete skills that compose? See the sequential-workflow pattern for the tradeoffs.

Graph View

Backlinks