The old approval problem was a slow one. Someone wrote next week's posts, sent them for review, and by Thursday the thread was eleven replies deep and nobody was sure which version of Tuesday's post was the final one.
The new approval problem is fast. An AI agent can draft a week of posts for five networks in the time it takes you to make coffee, and every one of them will read fine. That is exactly what makes them risky. A human writer who is unsure leaves a gap or asks a question. An agent fills the gap with something plausible and moves on.
I should be upfront about where I sit. Planpost is my product, a social scheduler that an agent like Claude, ChatGPT or Cursor can run. It is young, so I have no case studies to wave around. What I have is the review step I built into it, the habit I use around it, and an honest account of what it does not do yet.
Why review matters more with an agent
Agents make a particular kind of mistake. Not typos, which they rarely make, but confident errors that look exactly like correct work:
- A date that is off by a week because the brief said "next Thursday" on a Sunday.
- A statistic that sounds right and came from nowhere.
- A link to the wrong page, or to a page that does not exist yet.
- A tone that is fine in general and wrong for this week, the week something bad happened in your industry.
- A promise your product or your client's product does not keep.
Volume multiplies all of these. When you wrote three posts a week, a bad one was a bad one. When an agent writes twenty, a misunderstanding in the brief becomes twenty posts with the same flaw, scheduled at good times across seven networks.
And publishing is the one step you cannot really take back. You can delete a post, but not the screenshot someone took of it. Everything before publishing is cheap to fix. So the check belongs right before the post goes live, not after.
That is the case for turning approval on: when an agent drafts at volume, when an account carries a client's name or your reputation, and while you are still learning what your agent gets wrong.
What a reviewer actually needs
Strip the problem down, and the person approving a post needs four things. This was true when the drafts came from a junior copywriter, and it is still true when they come from a model:
- The post exactly as it will go out, not a summary of it or a screenshot of a draft.
- The time and the network it is scheduled for.
- A way to change something specific.
- One unambiguous way to say yes.
The fourth one matters more than it looks. Email let everyone avoid the difference between "yes", "yes, if" and "no". The worst sentence in content review is "approved with one small change": do you make the change and publish, or make the change and ask again? A review step should force the choice. Either the post goes out as it is, or it does not go out yet.
How to make Planpost hold an agent's posts
When you connect an agent to Planpost, the consent screen shows a switch called "Ask me before anything goes live". It starts off, so by default an agent's posts go out at the time it sets. Turn it on there, or any time later on the Agent page, where each connected app has its own switch and each API key has "Require approval before publishing" in its row. The agent cannot flip it itself. With it on, every post that agent schedules or publishes is held instead:
Home tells you how many posts are waiting for your approval, and the link goes straight to them. Each one shows its text per network, its media, and the slot it was meant for. You open it, change what needs changing, and approve it, which keeps the time it was scheduled for, or reject it, which makes it a draft. Until you approve it, it does not go out, whatever the agent was told.
One detail I care about: if a post's slot passes while it is still waiting, it does not publish late. The post goes back to being a draft, so you decide deliberately when it should run instead of having a Monday post appear on Wednesday because you approved it on Wednesday.
Chat inside Planpost works differently and does not depend on the switch. You say what to post, it saves a draft, and it publishes or schedules only after you confirm in the chat.
Who can approve
On your own account, that is you. If you are on a team, the owner and admins can approve; editors can draft and viewers can only look. The agent never approves its own work, and I think that line should stay bright. The whole point of the step is that a person who answers for the account has read the post.
A review habit that survives a busy week
A switch is only half of it. The other half is a habit, because a queue of posts waiting for review is useless if nobody looks at it until the slots have passed. What works for me:
Batch it. I brief the agent once for the week, on Monday morning, and review everything it drafted in one sitting the same day. Reviewing one post at a time as they trickle in is how review becomes rubber-stamping.
Read on the phone. Most people will see the post there, and a LinkedIn post that looks fine on a laptop can be an unbroken wall on a small screen.
Check the things agents get wrong. Dates, numbers, links, names, and any claim about what a product does. I read those first, then read the post as a whole for tone.
Fix the brief, not only the post. If I rewrite the same kind of opening three weeks in a row, the problem is my brief. I add a line to it ("no rhetorical questions as openers", "never mention prices") and the next week's drafts need less work.
Turn approval on for anything that matters. It starts off, and it is tempting to leave it off once the drafts get good. For anything with a client's name or your reputation on it, the minute it costs is the cheapest insurance you will buy.
If you post for clients
Here is the honest part. Planpost does not have a client sign-off link today, and there is no review flow that passes a post from one team member to another. The approval that exists belongs to the account owner: with it turned on, you approve what your agent drafts. If you run social media for clients, plan around that rather than discovering it later.
What you can do. On Pro you can invite a client into your team as a viewer, so they can see the calendar and the posts as they will go out. They cannot approve inside Planpost, so their yes still reaches you some other way, and you are the one who approves.
Or keep the client out of the tool entirely and send the drafts another way. If you do that, borrow the structure that keeps email from becoming the workspace:
- One message a week, with every post in it, not one message per post.
- Each post numbered, with its network and its time.
- A yes or no asked for each number, and "approved with one small change" treated as a no until the change is made and seen.
- Your decision recorded in one place, so that when someone asks in October why a post went out in September, you have the answer.
The honest caveat
Planpost is built for people who run their own accounts, and a freelancer or small studio who approves the work themselves. There are no customer quotes in this post and no invented ones either. If you need a client to approve each post inside the tool, Planpost cannot do that yet.
Where this fits in the week
Approvals sit on top of ordinary scheduling, so if your calendar is chaos, the review queue will faithfully show you the chaos. Get the base layer right first: how scheduling actually works and building a content calendar in one afternoon cover that ground. For the agent side, scheduling a week from Claude shows what the briefs look like, and the agentic scheduler page shows how to connect yours.
The email thread was never really about email. It was about the preview, the schedule and the decision living in three different places. With an agent writing the drafts, the same rule holds with higher stakes: keep the post, its slot and the yes in one place, and make sure the yes comes from a person.
