1 · Pick a post you might reply to
2 · Compare the drafts — same post, three voices
The interesting part is the denominator. If everyone’s volume doubled while replies halved, buyers didn’t get colder — the inbox got 4x noisier. Which means the fix is being recognizable, not louder.
When you segment those replies, how many came from accounts that had already seen a rep’s name somewhere — a comment, an event, a mutual thread? That split changed how we staffed outbound entirely.
Honest framing: these are pre-written sample outputs, not live generation — this page is static and calls no API. They exist to show the difference a consistent voice makes. The product goes further: it learns your voice from your comment history and edits, and drafts only on posts from the 120 buyers on your watchlist.
What this demo shows — and what it deliberately doesn't
The demo makes one point: the same post deserves a different comment depending on who is replying. Read the three drafts on any sample post above. None of them is "Great insights — fully agree!". Each takes a position, references experience, or asks a question a real operator would ask. That is the difference between a comment a buyer replies to and a comment a buyer scrolls past.
What the demo deliberately doesn't do is pretend to be the product. The samples are pre-written and the page is static — there is no "generating…" spinner theater over a canned response. The production system drafts live inside the app, and it doesn't use three generic personas: it uses your voice, learned from your own comment history and your edits to its suggestions. A persona is a costume. A voice model is a fingerprint.
Why most AI comment generators sound fake
Generic AI comment generators produce the same hedged corporate sludge on every post, and buyers can spot the pattern in seconds. Your rep opens a LinkedIn post from an ICP prospect. They click the AI comment button. The output: "Great insights — fully agree! This really resonates with my work in the space." The rep pastes it, sends it, and the buyer scrolls past because they have seen that exact comment 14 times this week.
The problem is not that AI cannot help reps comment. The problem is that the AI does not know how the rep talks. It does not know the rep's industry takes, their pet peeves, the words they would never use, or the way they actually open a reply. So it produces the median LinkedIn comment, which is by definition forgettable. The feed-level version of the same failure — generic AI posts getting downranked while specific ones pass — is documented in LinkedIn content strategy in the age of AI.
Meanwhile, the founder is still spending an hour every week reviewing what reps post before they hit send — because the alternative is the team sounding like a chatbot. The motion does not scale. This is the same bottleneck an in-house ghostwriter replacement is meant to solve — and it has the same root cause: no captured voice.
How the real drafting model works
On day one we wire the drafting model to your watchlist; over the next 20–40 edited comments it learns your cadence, opinions, and vocabulary — so by the end of week one drafts read like you wrote them, not like the median of every LinkedIn post ever scraped.
The drafting model, learned from your edits
There is no upfront interview or questionnaire. The model starts with a sensible default and improves on every interaction: when you accept a suggestion as-is, edit it, or reject it, the learning service extracts patterns across seven dimensions (tone, structure, content, length, engagement, credibility, audience) and shapes future drafts accordingly. Most users see the voice "settle" after roughly one week of normal use — typically 20–40 edited comments.
The watchlist
The AI never drafts replies to noise. It only sees the 120 profiles on the watchlist — 60 buyers, 30 amplifiers, 30 deal-stage targets. So every draft is on a post that actually matters. Generic comment generators waste their drafts on the default LinkedIn feed, where 80% of posts are from creators selling courses. (If you're still building the list, the watchlist construction guide walks through the three-tier composition.)
The review loop
The first week is supervised drafting. The voice owner reviews and edits every suggestion, and the model recalibrates on each edit. By week 2 reps can ship without per-comment supervision. Reps still review every draft before sending — the AI is a starting point, never an autopilot.
What the first 90 days look like
By day 7 the voice model is live, by day 45 reps are drafting in under 3 minutes per comment, and by day 90 the founder is no longer in the per-comment review loop.
- Days 1–14: Watchlist build, drafting model wired up, reps start drafting with supervision. The model learns from every edit.
- Days 15–45: Comment-to-send time drops from ~15 minutes (rep wrote from scratch) to under 3 minutes (rep reviews and adjusts the draft). Volume of daily watchlist comments per rep climbs from 1–2 to 5.
- Days 45–90: Reply rates from buyers stabilise. The founder steps out of the review loop. The team operates on voice consistency without daily founder oversight.
"By week 3 my AEs were commenting in my voice better than I was. I stopped reviewing every reply and the engagement actually went up." — Founder, Series A devtools (anonymous)
What this is not a fit for
Skip this if you want a generic AI tool, if the voice owner refuses to do the supervised voice-model setup, or if you do not have a watchlist of real buyers to engage with. Three honest disqualifiers:
- You want a one-click AI comment button for any LinkedIn post. That is not what we build. The drafting is bound to the watchlist on purpose — it is what keeps comments landing on buyers.
- The voice owner will not edit drafts during the first week. The model learns from your edits. If the voice owner just accepts everything in week one without reviewing, the drafts stay generic — which is exactly what you are trying to avoid.
- You do not have a defined ICP. Without an ICP, there is no watchlist. Without a watchlist, the drafting model has nothing to draft on. Fix the ICP first — the SDR-side workflow shows how ICP clarity feeds the watchlist before any drafting happens.
How to know if this is the right play for you
A 30-minute walkthrough with one of our strategists is the fastest qualification path. We will look at the last 10 comments your team posted, sketch what the voice-captured drafts would have read like instead, and tell you within the meeting whether the model would actually move your reply rates — or whether your current drafting workflow is good enough.
We hit the same wall last year. What finally moved the number wasn’t new sequences — it was reps commenting on the account’s posts for two weeks before the first email. Reply rate tripled on the accounts we warmed first.