AI can draft social posts for clients quickly, but a person should approve every post before it is published. The workable setup is a written brand voice for each client, a queue where drafts wait for review, a content calendar, and clear limits on what the AI may do and spend.
The risk of AI in an agency is rarely a badly written sentence. It is a post that sounds like the wrong brand, states something untrue about the client, or goes out when nobody was looking. A human approval step handles all three, and it is cheap compared with the cost of one public mistake on a client’s account.
What AI is good at, and what it is not
Drafting is where AI saves time. A first version of a caption, three options for a headline, a week of ideas built around a promotion: these take minutes instead of an hour. Reviewing is where people still earn their fee.
AI is weaker on facts it has not been given. It will happily invent an opening time, a menu item or a discount. It also drifts toward a generic, enthusiastic voice unless you pin it down. Both problems are manageable with good inputs and a review step. Neither is manageable if drafts publish automatically.
Step 1: Write down each client’s brand voice
A brand voice document is the single most useful input. Keep it to one page per client:
- Who the audience is and what they care about.
- Three to five words that describe the tone, and three that do not fit.
- Words and phrases the brand uses, and ones it avoids.
- Offers, prices and opening hours that are true today.
- Hard limits: topics never to mention, claims never to make, competitors never to name.
Include two or three real examples of posts the client liked. Examples teach more than adjectives do. Update the page when the client’s offer changes, because stale facts produce confident, wrong posts.
Step 2: Build an approval workflow
An approval workflow is the path a draft takes from creation to publication, with a person’s sign-off in the middle. Keep it simple:
- A draft is created and lands in a review queue, marked as AI generated.
- A named reviewer reads it against the brand voice and the facts.
- The reviewer edits it, approves it, sends it back, or deletes it.
- An approved post is scheduled on the content calendar.
- If the client must sign off, the approved post goes to them next.
Two rules keep this honest. The reviewer must be a person who knows the client. And nothing publishes without passing step 3, at any level of automation you later allow.
Step 3: Use a content calendar
A content calendar shows what is going out, on which channel and when, across every client. It does two jobs. It stops a client being posted to five times in one day and not at all the next. And it makes gaps obvious, so you can ask the AI for drafts for the empty days rather than discovering them on Friday.
Plan around the client’s real calendar: launches, events, public holidays, quiet seasons. Give the AI those dates in advance.
Step 4: Set limits on autonomy and spend
Decide how much freedom the system has, and write it down per client. A sensible ladder:
- Suggest only: the AI proposes ideas, and a person writes the posts.
- Draft on a cadence: the AI writes drafts on a schedule, and a person approves them.
- Place approved drafts in the calendar: after approval, posts are scheduled automatically.
Approval stays in every rung. Cost deserves a limit too. AI use is billed by the amount of text processed, and a loop that runs unchecked can spend real money. Set a monthly cap, and make the system stop and tell you when it reaches it.
Know the channel rules
Platforms have their own conditions. TikTok, for one, expects a person to post the content. The workable pattern there is for the AI draft to arrive in the account’s TikTok inbox, where someone finishes and publishes it in the app. Check each platform’s current terms before automating anything, and keep a person in the loop on any channel where the rules are unclear.
Check the draft before approving
Reviewers work faster with a short list:
- Is every fact true today: prices, dates, hours, claims?
- Does it sound like this client and not like any client?
- Would the client be comfortable seeing it on their account?
- Does it avoid the hard limits in the brand voice page?
- Is the call to action one clear thing?
- Are any images and credits cleared for use?
Keep a record
Log who approved each post and when. If a client later asks why something was published, you can answer in a minute. Over time, note which edits reviewers make most often. They show you what to add to the brand voice page, and the drafts improve.
Where software helps
Kervol’s MUSE module follows this pattern. It drafts posts in each client’s brand voice, holds them in an approval queue where a person can edit, schedule and move them, and shows them on a content calendar with campaigns. It supports Instagram and Facebook, and TikTok through a handover to the account’s own inbox. Autonomy levels and monthly spend caps are built, and MUSE does not publish on its own. A social inbox, insights, and KOL and PR tools are planned.
Checklist
- A one page brand voice for each client, with true facts and hard limits.
- A review queue where every AI draft waits for a person.
- A named reviewer who knows the client.
- A content calendar covering every client and channel.
- A written autonomy level per client.
- A monthly spend cap.
- Channel rules checked, TikTok handled by handover.
- A log of who approved what.
Updated 21 October 2026