Approve or Reject: A Checklist for AI-Drafted Posts

Read each post against your own material

Keep your website, your notes and what you know about the business beside you. Each question below is one you can answer only by comparing the draft with them.

1. 1

   Did this actually happen?

   Catches: Invented company stories

   Ask a model for a story and it writes one that fits your company, using real phrases from your website inside an event that never took place. A company that argues against a common practice gets a story about the time its founders tried it. You are the only reviewer who knows your history, so check every past event, habit or experience against what happened.

   Fix: Cut the story and keep the point as an observation or a question.

2. 2

   Who says so?

   Catches: Research claims with no source

   "New data suggests", "recent studies show" and "clinical observations" with no study named. They read as confident and on-brand, which is why they get approved. In a health, legal or financial business they are claims a customer or a regulator can hold you to.

   Fix: Name the source, or cut the claim.

3. 3

   Is that date still ahead?

   Catches: Past dates presented as upcoming

   Drafts are written from your website, and your website keeps old pages. A model that isn't sure of today's date will present last January's deadline as coming up in September. Check every date in the draft against today.

   Fix: Cut the date, or update it from a current source.

4. 4

   Is this for your customers?

   Catches: Posts written for the wrong audience

   Most websites also hold pages for staff, support, careers and legal. A draft built from an IT help page ends up walking the public through a staff password reset. If a whole topic in your plan came from a page like that, every post on that topic will have the same problem.

   Fix: Reject the post, and remove the topic if it came from an internal page.

5. 5

   Is every detail true of you?

   Catches: Invented numbers, features and hashtags

   Read the small print: each figure, each product feature, each hashtag. A model adds a plausible hashtag the way it adds a plausible story. #opensource on a product that is not open source is a false claim in a single word.

   Fix: Cut anything you could not back up if a customer asked.

6. 6

   Does it do what its format promises?

   Catches: A format the body ignores

   A this-or-that that never offers two options, a "data point" post with no number in it, a myth-buster with no myth. When the format and the body disagree, the post usually drifts into a generic story, and the variety you planned for the week is gone.

   Fix: Rewrite it to the format, or reject it.

7. 7

   Would you say this out loud?

   Catches: Stock phrasing

   Some sentences could sit in anyone's post. They are fluent and say nothing, and they repeat: one week we reviewed had a line about "the dynamic" shifting in four different posts. If you would not say it to a customer across a table, it is filler.

   Fix: Rewrite it in your own words, or delete the sentence.

Read the week together

These two never show up in a single post. Open the whole week before you approve the first one.

1. 8

   Is one topic taking over the week?

   Catches: Topic pile-ups

   When each post's topic is picked on its own, four topics across six posts put three or more on one topic in nearly two weeks out of three. Each post reads fine. The week reads like a company with one thing to say.

   Fix: Keep any one topic to a third of the week, and swap the rest.

2. 9

   Does a hook, a line or a number appear twice?

   Catches: Repeats across the week

   Reviewing one post at a time hides repetition. Two posts can open differently and still share a sentence in the second paragraph, and one good statistic can turn up in four posts out of twelve. Read the week's drafts side by side before you approve any of them.

   Fix: Each hook and each figure once a week. Keep the strongest use and cut the others.

If the post has an image

• The image and the caption say the same thing.
• Words on the image pass the same checks as the caption. A headline repeats the post's claim in the largest type on the page.
• Your logo never sits next to a claim you would not defend.

The checks for the picture itself, from garbled lettering to invented emblems, are in AI images for Instagram that don't look AI-made.

Edit or reject

If deleting the bad line leaves a post worth publishing, edit it. If the bad line is the post, reject it.

The checklist

• ✓Every story, habit or past event actually happened.
• ✓Every research or "data shows" claim names its source.
• ✓Every date is checked against today.
• ✓The post is for customers, not staff.
• ✓Every number, feature and hashtag is true of you.
• ✓The body does what its format promises.
• ✓No sentence you would not say out loud.
• ✓No topic on more than a third of the week's posts.
• ✓No hook, line or figure used twice in the week.