TL;DR
You draft a Slack message, reread it, and it lands as either too soft (“just wanted to circle back if possible…”) or too sharp (“this is blocking me, please respond”). Both cost you a reply or a relationship. Paste your raw draft into an AI chat with one line of relationship context and one line of desired outcome, and ask for “direct, kind, no corporate softeners, end with one specific ask and a deadline.” For voice matching, Claude (Sonnet 4.6) is the strongest of the three big chat models as of June 2026; GPT-5.5 drifts corporate, Gemini 3.1 Pro reads functional. Slack’s own “Edit with AI” can change tone too, but only inside a canvas, not the message box — so most people rewrite in a chat tab and paste back.
Who this is for
Anyone whose work happens in Slack threads: engineers chasing code review, PMs nudging a designer, managers giving feedback, founders chasing investor intros, and ICs who freeze before sending a hard message. It is especially useful for non-native English speakers who want tone confidence without sounding like a chatbot, and for direct-by-default people whose messages keep landing harder than they meant.
When to reach for it
Before sending any message where the wrong tone has a real cost: chasing a slipped deadline, pushing back on scope, asking for help, raising a concern with a peer, declining a request. Skip it for routine acknowledgments, emoji reactions, and threads where the relationship is already high-trust enough that tone does not matter.
Pick your tool (June 2026)
You have two routes: a general AI chat (paste draft, paste back the rewrite) or Slack’s own AI features. They are not interchangeable, and the table below is the part people get wrong.
| Tool | Best for tone rewrite | Where it works | Cost (June 2026) |
|---|---|---|---|
| Claude (Sonnet 4.6) | Strongest voice matching; paste 5-10 of your real messages and it copies rhythm and vocabulary | Any browser/app tab, paste back into Slack | Free tier (limited) / Pro $20/mo |
| ChatGPT (GPT-5.5) | Fast, fine for routine; tends to drift upbeat-corporate without firm constraints | Any tab, paste back | Free (ads on US Free since Feb 2026) / Go $8 / Plus $20 |
| Gemini 3.1 Pro | Accurate but functional, reads more like a draft than a finished message | Any tab; native in Google Workspace | Google AI Pro $19.99/mo |
| Slack “Edit with AI” | Quick tone change (formal ↔ friendly), proofread, shorten, translate | Inside a Slack canvas only, not the message composer | Bundled in Business+ ($15/user/mo annual, $18 monthly) and Enterprise+ |
Two facts that change how you’ll use this:
- The standalone Slack AI add-on is gone. Slack retired the $10/user a-la-carte AI add-on after August 17, 2025. AI is now bundled: every paid plan gets conversation/thread summaries and huddle notes, while the rewrite/recap/translation/writing-assistance suite is gated to Business+ and Enterprise+ (Slack help).
- Slack’s tone rewrite lives in canvases, not the message box. You highlight text in a canvas, click “Edit with AI” (or hover and click the sparkles), and pick “change tone” among proofread / shorter / longer / translate (Slack help). That is great for a structured update, but for a one-off DM most people are faster pasting the draft into a chat model and pasting the result back.
For the actual rewrite, Claude wins on voice matching because you can hand it a sample of your writing and it adapts to your rhythm and word choice rather than flattening you into “professional.” That is the whole game here — see the Claude writing workflow guide for how to set up a reusable voice prompt, or the ChatGPT writing assistant guide if you live in ChatGPT.
Before you start
- Have your raw draft in front of you, even if it is one ugly sentence. Rewrites are better with a starting point than with a brief.
- Know the relationship in one line: “peer engineer, high trust, has missed the last two reviews” gives the model 80% of what it needs.
- Decide the outcome you want in one sentence. Not the message — the outcome. “I want a code review by Thursday,” not “I want to mention the PR is open.”
- Pick a single chat surface for the rewrite step. Switching between ChatGPT, Claude, and Slack dulls your prompt muscle.
- Save one sample of your real voice somewhere. Paste it in so the model rewrites in your tone, not a generic corporate one.
Step by step
-
Write the ugly draft. Two sentences max — what is going on, what you want. Do not polish it.
-
Paste it into your chat model with this scaffold:
Rewrite this Slack message. Relationship: [one line] Desired outcome: [one line] Tone: direct, kind, no corporate softeners ("just wanted to", "circling back", "hope this finds you well"). End with one specific ask and a deadline. Keep it under 4 sentences. Here are 3 samples of my real voice to match: [paste] -
Read the rewrite aloud. If it sounds like a person you would respect saying it to you, it passed. If it sounds like a press release or an HR memo, reply “drop one level of formality.”
-
Check the ask is single, specific, and time-bound. Multi-asks dilute the thread; vague asks get ignored.
-
Edit one word. Even one word of your own voice turns an AI rewrite from “polished” to “real.” The receiver senses the difference even when they cannot name it.
-
Send. If you hesitated more than 30 seconds reading the rewrite, the draft was right and the rewrite is fine — your hesitation is the cost of sending a hard message at all, not a tone problem.
First-run exercise
Pick three messages you sent last week that did not get the response you wanted. For each, write the original outcome in one line, run the rewrite workflow, and compare. You will find one of three things: the tone was actually wrong (good — the workflow paid off), the ask was vague (better — fix it next time), or the message was fine and the receiver was the problem (also good — stop blaming your tone). The first run is a free calibration of which of your drafts actually needed rewriting.
Quality check
- The message ends with one specific ask and a deadline. Two asks split attention; no deadline gets deprioritized.
- Read aloud, it sounds like you, not a customer-service script. Customer-service voice (“Thank you for your patience, I appreciate your time…”) signals you do not actually care about the outcome.
- No softeners you would not say in person: “just wanted to,” “I hate to ask,” “if it is not too much trouble.” They invite the receiver to say no by default.
- Length is under 4 sentences. Messages over 4 sentences become threads where everyone reads the first line and misses the ask.
- The first sentence carries the most important information. People scroll past second sentences in Slack.
How to reuse this workflow
- Save 3 rewrite prompts as snippets: “chasing a slipped deadline,” “pushing back on scope,” “asking for help.” Most Slack tone problems map to one of these three.
- Keep a
voice/doc with 5-10 messages you wrote that landed perfectly. Paste it as context every time; voice matching gets sharper with more samples. - Track which rewrites get the outcome you wanted. After a month you will see which scaffolds actually work for your team.
- Build a “phrases I avoid” list — corporate softeners and passive-aggressive tells — and feed it as constraints.
- For recurring threads (1:1 followups, sprint nudges), save the rewritten template and tweak per instance instead of regenerating. The same habit pays off in your weekly report and 1-on-1 prep.
Common mistakes
- Asking AI to “make this nicer.” You get corporate softeners and lose the direct ask. Ask for “direct and kind” instead.
- Skipping the relationship line. Without it, the model defaults to formal-strangers tone, which reads cold to a peer.
- Sending the rewrite verbatim. The receiver senses generic phrasing even when they cannot name it. One word of your own voice fixes this.
- Letting the rewrite balloon to 6 sentences. Slack rewards 2-3 sentence messages; models default to longer.
- Two asks in one message. The second gets dropped, then resentment when nothing happens.
- Expecting Slack’s “Edit with AI” to fix a DM in the composer. It only edits canvas text, so for a quick message you are faster in a chat tab.
FAQ
- Will the receiver notice the message is AI-rewritten? Not if you edit one word of your own voice in. The “AI sound” is generic phrasing, which you remove by being specific.
- Which model is best for tone in June 2026? For voice matching, Claude (Sonnet 4.6) is the most consistent of the three big chat models — paste a few of your real messages and it copies your rhythm. GPT-5.5 is faster but drifts upbeat-corporate without firm constraints; Gemini 3.1 Pro is accurate but reads more functional.
- Can I do this inside Slack without a separate tab? Only via a canvas. Slack’s “Edit with AI” changes tone, proofreads, shortens, and translates highlighted canvas text, but it does not act on the message composer, and it requires Business+ or Enterprise+ since the standalone AI add-on ended in August 2025.
- What about messages in another language? Tell the model the target language and ask for natural register. Non-native rewrites often over-formalize; constrain with “casual Slack register, not formal.” Slack’s canvas AI also translates if you are already in Business+.
- What if my real voice is curt by default? Use the rewriter to add one warmth signal per message (a name, a thanks, an acknowledgment of context). Not corporate softeners.
- Can I use this for emails too? Yes for short emails. Long emails (200+ words) need different scaffolding — closer to a memo than a Slack message. See the AI email triage tutorial.
Related
- AI 1-on-1 meeting prep tutorial
- AI meeting summary tutorial
- AI weekly report tutorial
- Claude writing workflow
- ChatGPT writing assistant
Tags: #slack #communication #Tutorial