M3 · Free playbook
Five social posts, one blog, one newsletter — every week, in your voice, from your knowledge base, without you opening a doc. Seven days from sign-up to scheduled content. Same playbook we use on paid installs. Free.
TL;DR
A content engine is five things wired together: a writing brain, a voice profile, a scheduler, a newsletter tool, and an orchestrator that runs them on cron. The stack costs about $60 to $90/mo for most SMBs and takes roughly five hours of focused engineer work spread across seven days.
The hard part is not the automation. The hard part is the voice profile — what the owner sounds like in writing, what they would never say, what phrases gave their last draft away as AI. Most failed content engines publish AI-tasting slop because they skipped Day 4. Don't skip Day 4.
What follows is the seven-day plan, the exact stack, the honest tradeoffs between DIY and paid, the QA scenarios that prove the engine is working, and the risks worth respecting before you give an AI permission to publish in your name.
The 7-day plan
Each day fits in a single focused block. The Day 4 voice tuning is the work — every other day is wiring.
Day 1
~90 min
The voice profile is the most important field in the entire build. Pull 10–30 things the owner has actually written — old emails, LinkedIn posts, blog drafts, customer replies — and read them aloud. Capture cadence, idioms, sign-offs, and the five phrases the owner would never say. This becomes the anchor every draft is graded against on Day 4.
Day 2
~60 min
Five content pillars. Not eight, not three — five. They are the recurring themes the engine will rotate through. Then sign up: Buffer Standard for scheduling, Make.com Core for the automation backbone, Beehiiv for the newsletter (free up to 2,500 subs), and an Anthropic API key. Set a $200 hard cap on the API key in the Anthropic console.
Day 3
~90 min
Daily Content Generator (6am cron, picks a pillar + drafts a post). Weekly Newsletter Builder (Sunday 4pm, pulls the week's posts and stitches a newsletter). Performance Pulse (11pm daily, pulls engagement data into a Google Sheet). Topic Refresh (Monday 8am, scrapes industry RSS feeds for fresh angles). Wire each, test each manually, but do NOT turn the cron triggers on yet.
Day 4
~60 min
Generate five sample posts in the owner's voice. Send to the owner with one question: does this sound like you? Rate each: sounds-like-me / close / sounds-like-AI. If any post is rated “AI”, identify the phrase that gave it away and add it to the prompt's banned list. Re-generate. Iterate until 4 of 5 are rated sounds-like-me. Budget two rounds even if the first looks promising.
Day 5
~60 min
Run all eight base QA scenarios. Test approval end-to-end: a draft fires, owner gets a notification, owner approves, Buffer schedules. Reject path: owner kills the draft, “Killed” logs in the calendar sheet. Anti-pattern audit: generate 10 posts, grep the output against the banned-phrase list, require zero hits. Only after every QA passes do you turn the cron schedules on.
Day 6
~30 min
Flip Scenario 1 to ON. Confirm the next-run timestamp is what you expect. The first morning post fires the next day. Watch it land in Buffer's drafts queue. Approve manually for the first week — autonomous approve comes later, once you trust the voice.
Day 7
~15 min/day
Day 7 check what published. Day 14 review what landed and what didn't — identify any pillar that's bottom-quintile and consider rebalancing. Day 30 pull the report: total posts published, engagement deltas, KB gaps that surfaced, any voice drift. That report tells you whether to promote to autonomous approval or to keep the human in the loop.
The stack [Verified 2026-05-23]
This is the default stack we use on paid M3 installs. Pricing checked against vendor pages on 2026-05-23; re-verify within seven days of any paid install start — SaaS pricing has been moving in 2026.
Writing engine
$30–$60/mo at SMB volume
The model decision matters more than the platform decision. Sonnet 4.6 with a strong voice profile and the anti-patterns blocklist produces the most human-feeling drafts we've tested. Haiku is fine for classifiers; Sonnet writes the actual posts.
Social scheduling
$15–$29/mo
Integrates with X, LinkedIn, IG, FB, TikTok, and Pinterest from one queue. Hypefury is better if the owner is X-only and lives on Twitter; Buffer is the better default for businesses that publish in multiple places. Don't run both.
Newsletter
$0–$42/mo
Free up to 2,500 subscribers. Native referral program, clean editor, good deliverability out of the box. Substack is fine if the owner already has a list there; ConvertKit is overkill for SMB volume.
Orchestration
$9–$16/mo
Make's scenario builder handles multi-step content workflows better than Zapier for this specific shape. Cheaper at the operation volumes the engine needs. If the owner already has Zapier from an M2 install, use Zapier — don't add a second automation tool.
Brand voice anchor
Free
The voice profile + pillars + topic backlog + banned phrases all live in a single Google Doc the owner can edit from their phone. Make.com pulls from the doc every run. Notion works too; Google Doc is the boring, reliable choice.
Images (optional)
$20–$30/mo if used
If posts need accompanying images, Nano Banana Pro at ~$0.04 per image beats Midjourney for SMB volume and beats DALL-E on consistency. Skip on Pilot tier; add on Solo+ once the voice is dialled.
All-in monthly
For a small business publishing 5 social posts/week + a weekly newsletter, expect roughly $60–$90 all-in. The Anthropic API cost is the floor; image generation is the only meaningful variable.
The non-negotiable
“In today's fast-paced world…” “Let's dive in.” “Game-changer.” “Unlock the power of.” The whole internet has learned to scroll past those phrases in under a second. If they show up in your scheduled posts, your engagement collapses — not because the algorithm punishes AI content, but because readers do.
The blocklist applies to long-form too. Blog posts that open with “In an era of rapid technological advancement” tank the dwell-time signal that Google reads. Paste the full list into the prompt under a NEVER USE section, then grep your first 10 generated posts for any hits. Zero hits is the bar.
Read the full anti-patterns blocklist →DIY or paid — honestly
Most playbook PDFs end with “or just buy our thing.” This one is honest about when DIY is the better answer.
You already write your own content and just want to scale the cadence. You know your voice cold, you can articulate the five pillars in one sentence each, and you can spare six hours over a week. Most owners under $200K revenue who are already on social belong here — the voice training is faster when you trust your own ear.
You'd rather have a voice on the line walking through the Make.com scenarios than read JSON docs alone. You don't want strangers in your Beehiiv but you want help wiring the brain. Two 90-min calls work better than one for M3 — one for setup, one for voice tuning a week later.
Your time is worth more than $150/hr and you'd rather have us write the prompts than figure them out yourself. You want one channel live in seven days. You're comfortable handing us limited-scope access to Buffer + Beehiiv + your CMS. Most owners over $200K revenue who hate writing and love results land here.
QA — the ten checks
Run each scenario manually. If any fails, fix the prompt or the data source and re-run. Do not enable cron until every check passes.
Risks + safety rails
Brand-voice drift. Generated content sounds like the owner on Day 1 and like generic AI by Day 30 if nobody is auditing it. Schedule a quarterly voice-drift check — owner reviews 20 recent posts, flags anything that doesn't sound like them, those phrases get added to the banned list.
Cron failures that go silent. If a Make scenario errors at 6am and nobody sees it, you skip a publishing day. Wire a single-line error notification to Slack or SMS — any scenario failure pings the owner within five minutes.
Regulated industries. Healthcare can't make therapeutic claims. Legal can't imply attorney-client relationships. Financial can't imply returns. Paste the industry safety rails into the system prompt and run three adversarial sample posts trying to break them.
Vacation pause. The single highest-stakes failure mode is a scheduled post going live during a crisis the owner is too distracted to notice. Build a vacation toggle into the Make scenarios — owner flips one switch, every cron pauses, nothing publishes.
DIY the whole thing free. Or pay $500 and Iris is on a call walking you through the voice tuning. Or pay $1,797 and we install M3 on three channels in two weeks.
Free
Download the PDF
This playbook as a PDF you can print and check off.
Get the PDF →
$500
Iris-Assist
You do the clicks; Iris is on the line. One 90-min call. You keep every credential.
Book a session →
$1,797 + $197/mo
Iris Build Solo
We install M3 across three channels in your accounts. 30-day money-back guarantee.
See Solo →