Slack-First Marketing Agent: 5 Prompts That Replace 3 Tools

It's 9:47 on a Tuesday. I type into #marketing-room: @Claude new brief: Q1 launch for our analytics product, positioning against Mixpanel and Amplitude, target is PMs at Series B SaaS, budget $40K, ship date Feb 14. Thirty seconds later, a campaign brief lands in the thread — audience, message pillars, channel mix, KPIs, risks. No Notion tab. No Figma doc. No waiting for the strategist to clear their morning.

That's what a Slack-first marketing agent looks like in practice. Anthropic shipped native Claude in Slack last year and Claude Code for Slack went into beta in December 2025; the @Claude mention is now a real surface for getting work done, not just answering questions. After two months of running one inside my own team, I can hand five daily asks to it and skip three SaaS subscriptions.

Here are the five prompts — and what each one replaces.

1. The campaign brief — replaces the strategist's first day

The single biggest time leak in a marketing team is the strategist's "Day 1" on a new campaign. Pull positioning, audience, channel mix, KPIs, risks, dependencies. Most strategists spend 3-4 hours on this before they write a word of copy.

The prompt I run in #campaigns:

@Claude build a campaign brief. Product: [one-line]. Audience: [ICP, one line]. 
Top 3 competitors: [list]. Budget: [$]. Ship date: [date]. Existing assets to reuse: 
[list or "none"]. Return: positioning in 2 sentences, primary + secondary audiences, 
3 message pillars with proof points, channel mix recommendation, KPI tree from 
spend → CTR → signups → activation, top 3 risks. Keep it under 600 words.

What I like about pinning the word count is it forces the agent to compress. A brief that fits in 600 words is one a copywriter can actually use. The KPI tree piece is the part strategists skip most often — it forces the team to argue about what counts as success before they argue about creative.

This single prompt replaces the strategist's first day on the brief. It does not replace the strategist. A human still has to make judgment calls on positioning. But the empty page is no longer the bottleneck.

2. The draft copy — replaces the copywriter's first hour

Once the brief is in the channel, I ask for first-pass copy in the same thread, so the context carries forward.

@Claude draft 5 ad copy variants (3 short, 2 long) for Meta, using the brief above. 
Each variant needs: a hook in the first 5 words, one benefit, one proof point, 
one CTA. Match the brand voice doc pinned in this channel. Flag any variant 
you're unsure about.

The "flag any variant you're unsure about" line matters. Without it, the agent confidently hands you five mediocre variants. With it, you get honest signal on which ones need a human rewrite. I usually kill 2-3 of the 5 on the first read — the remaining 2 go to the copywriter for sharpening, not for blank-page work.

This prompt replaces the copywriter's blank-page hour. It does not replace the copywriter's judgment, which is what makes the difference between copy that performs and copy that's merely grammatical.

3. The competitor check — replaces the analyst's afternoon

The third prompt is the one I underestimated. Every Friday at 10am, the agent runs a sweep and posts to #intel:

@Claude check these 4 competitors' recent moves: [domain1, domain2, domain3, domain4]. 
Look at: their homepage, last 3 blog posts, last 5 ads in Meta Ad Library, any 
pricing changes. Return: what's new this week, what they're emphasizing in 
messaging, any positioning shifts, and 2-3 angles we should test or defend against. 
Cite the URL for every claim.

The "cite the URL" line is non-negotiable. The agent will hallucinate competitor moves if you let it — confident, plausible, and totally made up. I learned this the hard way in 2024 when I lost 20 minutes on a competitor pricing change that never happened. The URL requirement is the difference between useful intel and confident nonsense.

This replaces the analyst's half-day of browsing. It does not replace the analyst's synthesis — but the synthesis part is what humans are actually good at, and it goes ten times faster when the raw scan is already done.

4. The weekly recap — replaces the report builder

Mondays at 8:55am, a message appears in #weekly-recap that I did not write:

@Claude pull last week's data from [Google Ads / LinkedIn / HubSpot — list the 
connected sources]. Compare to the previous 4 weeks. Flag: spend, pipeline, 
CAC, top 3 campaigns by ROAS, bottom 3, any creative that fatigued. Output 
as a Slack post with the numbers inline, no spreadsheet.

This is the prompt I demo to other marketers because the time saved is so obvious. The weekly recap used to take someone a full morning. Now it's a 30-second read on Monday morning before standup, and any anomaly goes to the human for a real investigation.

A note on data sources: this only works if you've connected the tools properly. Anthropic's Claude Integrations (launched 2025) and tools like Gumloop or Relay.app handle the connection layer. The agent itself doesn't need to know how to call the Google Ads API — it just needs the integration to be there.

5. The content ideas — replaces the editorial meeting

The fifth prompt is the one that gets the most pushback, because editorial meetings are sacred to content teams. But I've come around on it.

@Claude read the last 20 blog posts in [Notion database / Drive folder]. 
Identify: 3 topics we covered well but haven't updated in 12+ months, 
3 trending questions our customers asked support about this quarter, 2 angles 
our competitors haven't covered. For each, return: working title, 2-sentence 
angle, and a source citation. No fluff ideas. If you can't back it with a 
source, leave it out.

That last line is the editorial discipline. "No fluff ideas. If you can't back it with a source, leave it out." Without it, the agent generates 12 generic topics that could've come from any blog's tag page. With it, you get a short list of defensible angles — and the editorial meeting becomes 20 minutes of "which of these 8 do we ship first" instead of 60 minutes of brainstorming.

This replaces the editorial meeting's generation phase. It does not replace the meeting's selection phase — that judgment, the "does this fit our brand and our audience" call, is the part humans should be doing anyway.

What to watch out for

Three things will bite you if you don't set them up front.

Guardrails on data access. The agent can see what it can see. If you give it the marketing channel, it should not see the legal channel. If you give it Notion, scope it to the marketing workspace. The 2024 Slack-Salesforce breach started exactly this way.

Token costs add up. Running prompts 3, 4, and 5 on a schedule means the agent is reading your competitor's entire blog, your full ad accounts, and 20 of your old posts. The bill is not huge, but it's not zero. Budget $30-80/month per active workspace for an SMB team.

The agent is the floor, not the ceiling. Every team I've seen fall behind with this pattern made the same mistake: they treated the agent's first draft as the final draft. The first draft is the floor. The copywriter, the strategist, the analyst — they push the work up from there. The point of a Slack-first marketing agent is to kill blank pages, not to kill craft.

The shift this is really about

The story isn't "AI replaces marketing roles." The story is that the cost of a blank page has collapsed. Strategists, copywriters, and analysts used to spend 70% of their time generating raw material and 30% applying judgment. That ratio is flipping. The raw material is now 30 seconds and a prompt. The judgment — the part that actually moves a campaign — is the part worth paying for.

If your team is still routing every request through a project manager, a brief template, and a kickoff meeting, the Slack-first agent is the cheaper, faster way to do the same work. The only thing you give up is the meeting.

What you do with the time you get back is the real question.