Writing Assistant · Voice DNA
It drafts in their voice.
And it can prove it.
Every profile carries a Voice DNA record plus the posts that person actually published. The assistant drafts from both, and when a fact is missing it ships a [placeholder], never an invention. This is Michael Maximoff’s real record. Below it, the real editor, in his real voice.
7-day full-access trial, $0.95 to start. Cancel anytime.
His real record, 8 of 9 fields filled. The one he left empty stays empty here too: his published feed already does that job.
- Voice
- Style
- Target audience“CMO, CEO, founders, owners, startups, B2B executives, marketing and sales leaders”
- Company
- Offering“b2b appointment setting, omnichannel marketing, full-funnel marketing, B2B sales, B2B growth”
- Always include“numbers, data, examples from the previous experience, work with clients”
- Never include“generic advices”
- Sample postsPaste 3–5 of this person's best posts. The Assistant uses these as few-shot examples.
- Extra context
His real record, 8 of 9 fields filled. The one he left empty stays empty here too: his published feed already does that job.
One house style with your name on it. The em dash every fourth line, the motivational close, a number it made up because the post needed one.
One voice per profile, learned from what they actually shipped. The em dash is banned at the system level, by name. So are invented numbers.
The engine’s own rule, word for word: “Never use: The em dash symbol (AI tell)”.
The real interface
The actual editor, mid-draft.
The editor inside buyWords, playing its real flow: a blank page, a topic handed over by Scout, a draft that previews exactly like a LinkedIn post, then a profile switch that re-keys everything, history included.
The posts in the preview are not AI output staged for a screenshot. They are posts Michael Maximoff and Jared Schieber actually published, with their real results printed beneath. Two profiles, two voices, one engine.
Voice, derived
The fingerprint is checkable.
Voice traits here are not adjectives typed into a form. They are measured from 9 posts Michael Maximoff actually published, listed below with live LinkedIn links, so you can check any row yourself.
- 6 of 9posts open in first person
- 3 of 9openers carry a question
- 220 wordsmedian post length
- 1 emojimedian per post
- 2 to 15blank-line beats per post
- 7 to 24 wordsopener length range
| Shipped | First line, his words | Words | Beats | ER* | Views |
|---|---|---|---|---|---|
| Apr 4, 2026 | My co-founder is obsessed with AI. Can anyone relate? | 220 | 12 | 0.5% | 12,187 |
| Apr 13, 2026 | I was driving, and after the first 20 minutes, I said to myself, “What an incredibly insightful conversation, was I really part of it?” | 188 | 9 | 0.5% | 8,374 |
| May 4, 2026 | shown aboveWe grew the BLKNS community to 3,600 members. Then we stopped investing. Why? | 338 | 15 | 0.9% | 7,454 |
| Mar 9, 2026 | I'm watching Apollo, Outreach, SalesLoft, Nooks, Instantly and many others all converge into the same tool. | 249 | 13 | 1.5% | 3,861 |
| Apr 2, 2026 | Just wrapped up an interview with marketing GOAT Udi Ledergor, ex-CMO and Chief Evengelist at Gong. Excited to share this one soon 💣 | 47 | 2 | 4.3% | 2,914 |
| Feb 24, 2026 | You shouldn't have to pay to learn. | 144 | 9 | 4.4% | 2,523 |
| May 6, 2026 | Upd from Belkins: we’re steadily trying to figure out Lead Gen 3.0. | 296 | 15 | 4.2% | 2,115 |
| May 3, 2026 | Abbas Somji, thank you for inviting me onto your podcast. It’s refreshing to be on the other side of the interview for a change. | 63 | 3 | 3.6% | 2,095 |
| Apr 9, 2026 | I bet 90% of tech leaders spend at least 1–2 hours a day working on AI. People reading this fall into this group. | 239 | 10 | 3.9% | 2,012 |
| 9 posts · totals | 1.6% | 43,535 | |||
Columns abridged on phones · words, beats and the per-post rate read on wider screens
*Engagement rate: (reactions + comments + shares) / views. Words and beats are counted from the published text. Every date links to the live LinkedIn post; the tagged row is the one quoted in full in the editor above.
The writing engine
Four layers between a prompt and a draft.
The system prompt is assembled per profile, per request. These are its real rules, word for word. Most vendors hide this layer. It is the product.
buyWords writing rules
The team brain. House rules that outrank everything, including the author's own bad habits.
- “Compress quickly. Get to the point with very little runway.”
- “Use contrast well. The market does something. The client does the opposite.”
- “Humor with a straight face. Dry, not wacky. Commit to the bit calmly.”
- “Posts should end 20-35% earlier than feels comfortable.”
Voice DNA
Every draft is a post by him, in his voice, for his audience. The audience already knows him, so the engine is forbidden to introduce or describe him in third person.
- “Just write what they would post.”
Recent published posts
His last five published posts ride into the prompt at request time. Cadence comes from what actually shipped, not from a form filled out once.
- “Do NOT reuse the same hook, opener, or topic from these posts. We're writing the NEXT post, not a sequel.”
Authoring honesty, non-negotiable
The engine may never fabricate numbers, events, outcomes, or client stories. Voice DNA, the team brain, recent posts, and the chat itself are the only sources of factual specifics. Its own words:
- “Fabricated specifics are the single biggest failure mode of AI-written posts.”
When a specific would strengthen the postI pulled data from [N] prospect calls last month. [X]% had quality issues, [Y]% couldn’t scale.Square-bracketed placeholders the author fills before publishing. Or a directional claim with no fake number. Or a re-anchor to something the author actually said.
And the draft gets flagged in chat⚠ Placeholders to fill in: [N] calls, [X]%, [Y]%. Don’t ship until those are real.A draft with placeholders must be flagged outside the post fence. Paste-and-publish with invented numbers is the failure mode this engine exists to kill.
The first three layers are cached between requests, so a working session never pays full price to re-read the brain.
Clarifications
When it needs a fact, it asks. Once.
No interrogations, no open-ended forms. Every clarification the engine emits is a numbered menu with an escape hatch, and voice onboarding is capped at five questions, each skippable forever.
- VoiceWhat’s the voice/personality of this profile?
- AudienceWho are you writing to?
- ToneWhat tone should the posts hit?
- CompanyQuick sentence on what the org/role does (background context, never gets pasted into posts):
- SamplesGot a sample post or two from this profile? Paste one in. (You can paste more later.)
Every onboarding question ends with the same exact line: “Or reply skip and I won’t ask again.” And any draft verb bypasses the questions entirely; the draft comes first, every time.
Production, metered
May 2026, counted from the ledger.
Every assistant call is metered, and this is a real month off the meter: every conversation, every token, every micro-dollar, May 7, 2026 through May 28, 2026.
- conversations
- 66
- messages
- 372
- teams
- 8
- metered calls
- 193
- tokens read
- 396,499
- tokens written
- 99,392
Real model spend that month, divided across the month’s drafts, lands well under what your plan budgets per post. The receipt shows the working: spend in, drafts out, headroom left. The plans are priced on that cushion.
The ladder falls back a rung on rate limits and transient errors. Every May conversation ran on the top rung. There is no model picker in the app, on purpose: you pick a voice, not a model.
The AI budget, shown as posts: ~40 a month on Starter, ~64 on Team, ~80 on Corporate, ~160 on Agency. See pricing
The loop
Ideas come in. Drafts ship out.
The editor is a station, not an island. These are the real strings that carry work in and out of it.
All of it keys off the active profile: switch the voice and the editor, the pickers, and the history re-key.
Scout ranks fresh angles against his topics. One click hands the winner to the composer.
The same ask you watched land in the editor above. The draft comes back, previews exactly like the feed, and the chat files itself behind History.
The real toast and the picker’s real header: the draft lands in the Draft column, and next session the picker hands it back.
Topic Scout ranks the angles, the calendar greets the blank page with what is due next (beat 1 above holds the quiet state and the up-next row that replaces it), and when the post ships, analytics brings the numbers home to the board. Then Karma puts the team behind it.
Your roster already has a voice. Put it in writing.
Add a profile and the assistant reads what they actually published. The trial is the whole platform, not a demo.
7-day full-access trial, $0.95 to start. Cancel anytime.