How to Keep AI-Drafted Posts On-Brand: A 10-Point Human QA Checklist

AI drafting tools have made the first draft cheap. They have not made the last check optional. When an agency sends a batch of posts to a client, the client does not care which parts were drafted by a person and which by an AI model; every mistake is the agency's mistake.

This article is for the reviewer: the person who reads the batch right before it goes out for client approval. It is a 10-point checklist you run on every post, in order, with a copy-paste version at the end.

When to run the checklist

Run it after internal editing and before the batch leaves the agency. Not during drafting, because you will end up rewriting instead of reviewing. Not after the client has seen it, because then you are doing damage control.

Two ground rules:

1. The reviewer is not the drafter. Even if the drafter was an AI, the person who prompted and edited it is too close to the copy. Fresh eyes catch more.
2. Have the brand brief open. You cannot check voice against memory, especially if you review for several clients in one sitting.

The 10-point QA checklist

1. Facts are true and current

AI models write fluent sentences whether or not the facts underneath are right. Check every:

• Date, time, price and location
• Product name, feature and availability
• Name and job title of any person mentioned
• Event detail (venue, registration link, deadline)

Test: Can you point to a source (the client's website, brief, or an email from the client) for every fact? If not, cut it or ask the client.

2. Claims are ones the client can stand behind

This is different from facts. A claim is a statement about quality or outcome: "the fastest", "trusted by thousands", "guaranteed results", "the leading provider".

• Remove superlatives the client has not used themselves.
• Remove numbers that do not come from the client.
• Watch for invented social proof: testimonials, customer counts, or awards that sound plausible but are not in the brief.

Test: Would the client be comfortable if a competitor or a regulator asked them to prove this sentence?

3. Voice matches the brief, not the AI's default

AI drafts often drift toward a generic, upbeat, slightly salesy tone. Compare each post with the brief's voice traits and sample lines.

• Sentence length and rhythm match the client's usual style
• Formality level is right for this platform and this client
• No phrases that sound like any brand at all ("We're thrilled to announce", "In a world where...")
• Humor only if the client actually uses humor

Test: Cover the logo and handle. Could this post belong to a competitor? If yes, it is not on-brand yet.

4. Banned words and phrases are gone

Every client has words they never use: a competitor's name, outdated product names, jargon they dislike, or terms with legal weight in their industry. Keep a banned-words list in each client's brief and search for it.

• Run a find (Ctrl/Cmd + F) for each banned term across the batch
• Check spelling conventions (UK vs US) and trademark styling
• Check emoji and exclamation mark rules

Test: Zero hits on the banned list. This is the one check you can make fully mechanical, so do.

5. The CTA is clear, correct and allowed

• One clear call to action per post, unless the format calls for none
• The CTA wording matches the client's usual style ("Book a call" vs "Let's talk")
• The action is something the client actually offers right now (no "DM us" if they do not monitor DMs)
• The CTA fits the platform (no "link in bio" on LinkedIn)

Test: If someone does exactly what the post says, will it work?

6. Links work and go to the right place

AI drafts sometimes include placeholder links, old URLs, or links that look correct but are not.

• Click every link
• Confirm the destination is the intended page, not the homepage
• Check UTM parameters if the client uses them
• Make sure no link points to a staging site, a competitor, or a page that does not exist

Test: Every link opened in a private browser window, and every one landed where it should.

7. Platform fit

A post that works on LinkedIn may fail on X or Instagram.

• Length: within the platform's limit, and the key point appears before any "see more" cut-off
• Formatting: line breaks render properly; no markdown symbols showing as raw text
• Hashtags: number and style match the client's practice on that platform
• Mentions and tags: correct handles that exist on that platform
• Visual reference: the copy matches the image, video or carousel it is paired with

Test: Preview in the scheduler or the platform's composer, not just in a document.

8. Legal, compliance and sensitivity flags

You are not a lawyer, but you are the last person who can raise a flag.

• Regulated industries (health, finance, legal, education, alcohol): any claim about outcomes, returns, cures or guarantees goes back to the client for sign-off
• Promotions and giveaways: terms and eligibility are referenced if the client's practice requires it
• Third-party content: images, quotes and music are licensed or the client's own
• Timing: nothing scheduled that could read badly alongside a news event, a holiday or a sensitive date for the client's audience
• Sponsored or partner content is disclosed according to the client's policy

Test: If you hesitated on a post for any reason in this section, it is flagged for the client, not silently approved.

9. Consistency across the batch

Review the batch as a set, not just post by post.

• No two posts open the same way
• No repeated phrases across posts in the same week
• The mix of topics matches the content plan
• Nothing contradicts another post in the batch (different prices, different dates)

Test: Read every opening line in a row. If they sound like a template, vary them.

10. Right client, right account

The most embarrassing mistake an agency can make is posting one client's content on another client's account. When you review several brands a day, it happens.

• The client name, product names and hashtags all belong to this client
• The post is in the right client folder, calendar and scheduler workspace
• No leftover references to another client from a reused template

Test: Read the brand name in each post out loud, then check the account it is scheduled to.

The copy-paste QA checklist

Paste this at the top of every batch review doc, or into your project management tool as a checklist.

AI-DRAFTED POST QA (reviewer completes before client send)
Client: ____________  Batch / week: ______  Reviewer: ______
Brand brief version checked against: ______

For every post:
[ ] 1. FACTS: every date, price, name, product and event detail sourced
[ ] 2. CLAIMS: no superlatives, numbers or proof the client hasn't provided
[ ] 3. VOICE: matches brief traits; not generic; passes "cover the logo" test
[ ] 4. BANNED WORDS: zero hits; spelling, styling, emoji rules followed
[ ] 5. CTA: one clear action, client-approved wording, actually works
[ ] 6. LINKS: every link clicked, correct page, UTMs correct
[ ] 7. PLATFORM FIT: length, line breaks, hashtags, handles, visual match
[ ] 8. COMPLIANCE: regulated claims, promos, licensing, timing, disclosure
[ ] 9. BATCH CONSISTENCY: no repeated openers, no contradictions
[ ] 10. RIGHT CLIENT: names, hashtags, workspace and account all correct

Flags for client (anything uncertain from 1, 2 or 8):
- ____________________________________
- ____________________________________

Posts returned to drafter (post # and reason):
- ____________________________________

Reviewer sign-off: ______  Date: ______

Making the checklist stick

A checklist only works if it is used every time, including on busy weeks.

1. Keep it to 10 points. Longer lists get skimmed. If you add a point, remove one.
2. Track returns by point. If most returned posts fail on point 3 (voice), the problem is upstream: the brief or the drafting setup, not the reviewer.
3. Feed fixes back into the brief. Every time a banned word or a bad claim slips through, add it to the client's brief so the next draft starts cleaner.
4. Separate "fix" from "flag". Reviewers fix voice, links and formatting. They flag facts, claims and compliance for the client. Do not let a reviewer guess on a legal question.
5. Time-box it. Decide on a target review time per post and watch for posts that take much longer; they usually point to a weak draft that should go back rather than be rewritten.

The cleaner the drafts coming in, the faster this checklist runs. Drafting tools that keep a stored brand profile per client tend to reduce the number of voice and banned-word catches, because the AI is not starting from zero each time. Impret works that way for agencies, but the checklist above applies whatever tool your team drafts with.

Next step

Try the free caption generator, then run its output through this checklist to see where your review time goes.