AI social posts that do not sound like AI

Readers can spot AI social copy in about a second, and it is almost never because of a factual error. It is the texture: every sentence the same length, every claim the same size, an opening that restates the topic and a closing that summarises what was just said.

Here are the patterns, and what to do about each.

The tells

1. No specifics

AI drafts default to the general case, because that is the safest completion. "Consistency builds trust with your audience" is unfalsifiable and unmemorable. "We cut that homepage from 340 words to 96 and demo requests went up" is neither.

Fix: put the specifics in the prompt. Numbers, dates, names, the actual thing that happened. A model cannot invent your March revision-round count, and it will happily use it if you give it.

2. Even sentence rhythm

Human writing is lumpy. Three words. Then a long sentence that wanders somewhere before it lands. AI writing tends to a uniform 12-to-18-word cadence that reads competent and forgettable.

Fix: in editing, break one sentence in half and merge two others. It takes ten seconds and does more than any prompt instruction.

3. The summary ending

Models like to close by restating. On social, a summary ending is dead space — the reader already read it, ten seconds ago.

Fix: delete the last sentence. Do it as a reflex. Then check whether the new ending asks for something or lands a point; if it does neither, delete another.

4. Hedging

"Can help", "may improve", "one of the most effective ways". Hedges make copy safe and toothless, and they are a model's default under uncertainty.

Fix: search your draft for "can", "may" and "often" and either commit to the claim or cut it. If you cannot commit, you do not have the post yet.

5. Vocabulary that gives it away

Certain words now read as machine-written on sight: delve, leverage, unlock, elevate, navigate the landscape, in today's fast-paced world, it's not just X, it's Y. The construction "it's not just a caption, it's a conversation" is a particularly strong tell.

Fix: keep a banned-word list in your brand profile and check drafts against it.

6. Emoji as punctuation

A rocket at the start of every line is a formatting habit, not a voice.

Fix: at most one emoji, and only where it replaces a word.

The prompt changes that matter most

1. Give it your actual material. A brand profile with your audience, topics and refusals produces sharply better drafts than a tone adjective. So does pasting the source paragraph you want the post to come from.
2. Say what to avoid, not just what to do. "No summary ending, no hedging, no emoji, do not restate the topic in the first line" removes more slop than any positive instruction.
3. Ask for a shape, not a length. "One-line hook, one specific claim with a number, one question" beats "write 100 words".
4. Generate three, keep one. Variance across drafts is more useful than iterating on a single one. Take the best opening and the best ending from different versions.

The edit that does the most work

Change one thing only a human could know. A time, a place, a name, a number, a small failure. One concrete detail per post is enough to break the pattern entirely — and it is also the sentence people reply to.