Approval-first marketing means the AI does the work and a person says yes before anything goes out. To make it work, every action arrives as a specific card (the exact post, ad and budget, email and list), approving takes one tap from wherever you are, anything that spends money starts paused, low-risk work can be auto-approved once you trust it, and every edit and skip teaches the system. You keep control without doing the work.
Key takeaways
- The risk with AI marketing isn't bad drafts. It's things going out that you never saw.
- Make every action a card with the exact content, audience, cost and timing on it.
- Approving has to take seconds, from your phone, or you'll stop doing it.
- Anything that spends money starts paused, with the amount stated before the ask.
- Auto-approve only low-risk, reversible work, and only after you've seen enough of it.
- Edits and skips are data. A good system learns from them and asks less over time.
The problem with fully automatic marketing
AI can now write a post, design an ad, build an email list and launch a campaign. The question isn't whether it can do the work. It's what happens when it does something you didn't want.
What goes wrong without approvals:
- Wrong tone. A post that sounds like no one at your company.
- Wrong facts. A price that changed, a feature you don't have, a customer named without permission.
- Wrong money. A budget typo, or a test that kept spending all weekend.
- Wrong people. A cold email to an existing customer, or to someone who opted out.
- Account risk. Automated posting on Reddit, Hacker News or LinkedIn DMs that gets your account limited.
Each of these is rare. Across hundreds of actions a month, rare things happen. And they happen under your name.
The fix isn't to stop using AI. It's to put one cheap step between the work and the world: your yes.
What a good approval card shows
An approval is only as good as what you see when you give it. Each card should show:
| Action | The card shows |
|---|---|
| Social post | The exact text and image, the network and account, the time it goes out |
| Article | Title, summary, the full draft one tap away, where it publishes |
| Ad | Creative, audience, daily budget, start state (paused), the metric it's judged on |
| Cold email batch | The email, the list size, who's on it and why, the sending inbox |
| Budget change | Old amount, new amount, and why |
Two rules make cards trustworthy:
- Money and risk go first. "$40/day, starts paused" belongs at the top, not in the details.
- Always offer a default. "I'd pause ad set B. OK?" means one 👍 is enough. A card that asks an open question takes real thought every time.
Make approving fast
If approving means logging into a dashboard, you'll do it twice and then stop. The approval needs to happen where you already are, in seconds.
What fast looks like:
- One place. Cards arrive in a chat you already check, such as WhatsApp, iMessage, Telegram or email.
- One tap. 👍 approves, ✏️ asks for a change, skip drops it.
- Batches. "Yes to all" for a morning's worth of routine cards.
- Plain-language edits. "Make it shorter" or "move it to Friday" should be enough. You shouldn't have to open an editor.
- A rhythm. One morning brief with three cards beats 20 pings across the day.
When approving takes seconds, it stops feeling like overhead. It feels like being the person who decides, which is the job.
Paused by default for anything that spends
Money deserves an extra rule: anything that spends starts paused.
- New ad campaigns and ad sets are created paused, with the budget on the card.
- Budget increases are their own card, with the old and new amounts.
- Scaling a winner is a proposal, not an automatic action.
- Ad spend is billed by the ad platform to your own account, so you can always see it there too.
This costs a few seconds per campaign and removes the worst outcome: waking up to a weekend of spend you never agreed to.
When to auto-approve, and when not to
Approving everything forever is safe but slow. Once you've seen enough of a type of work and it's consistently right, let some of it go through.
Good candidates for auto-approve
- Scheduled organic posts in a format you've approved many times.
- Reposting your own approved content to a second network.
- Routine reports.
Keep asking for
- anything that spends money or changes a budget,
- cold email batches (new people each time),
- articles and pages that publish to your site,
- anything mentioning a customer, a price, a promise or a competitor,
- anything on Reddit, Hacker News or LinkedIn DMs, where you should post yourself.
Turn auto-approve on per category, not globally, and review what went out once a week. If something slips, turn that category back to manual.
Learning from approvals, edits and skips
Every decision you make is information. An approval-first system should use it.
- Approvals say "more like this".
- Edits show exactly what was wrong: tone, length, a claim, a word you never use.
- Skips say "not this", and a pattern of skips says "not this kind of thing".
Over a few weeks that adds up to a picture of your voice and your judgment: you never use exclamation marks, you prefer customer stories to feature lists, you don't want to mention a certain competitor by name. Written down as learnings, they shape the next draft, so you edit less and approve faster.
The test of a good system is simple: does it ask better questions in week four than in week one?
A sample day, approval-first
Here's what a day looks like with an AI doing the work and a founder approving. This uses Shiftly, our fictional example.
- 8:02 The morning brief: "Yesterday: 9 sign-ups at $16 each. 3 things for today: 1. LinkedIn post on the 'spreadsheet' angle, 2. Blog: '7shifts alternatives for diners', 3. Pause the $22 CPA ad set."
- 8:05 The founder replies "👍 1 and 3. For 2 make it about diners, not restaurants."
- 8:06 "Done. Post goes out 9:10. Ad set paused. I'll send the diner draft by noon."
- 12:10 The article draft arrives as a card. The founder skims it and replies 👍.
- Friday A new test campaign card: three videos, $40/day, starts paused. The founder asks to hold it until Monday.
Total founder time: a few minutes. Nothing went out that the founder didn't see.
How Cosmo works this way
Cosmo is built approval-first. Every action he proposes (a post, an article, an ad, a cold email batch, a directory listing, a budget change) is an approval card in WhatsApp, iMessage (US), Telegram, email or the web app. Reply 👍, ✏️ or skip, or "yes to all". New ads are created paused. Auto-approve is opt-in, and only for low-risk work. On Reddit, Hacker News and LinkedIn DMs he drafts and you post.
He learns from every approval, skip and edit, and he can lead with a daily plan or follow your instructions. The rule is the same everywhere: nothing goes out without your yes. See how an AI CMO works and the approval workflow entry.
One short email when a new guide is out. At most one a week.