Is this AI?

What the checker looks for

It reads your draft the way a sceptical reader on X does, and flags three kinds of tell:

• Phrasing that reads as machine-written. Stock openers, filler and tidy turns of phrase that show up far more in AI drafts than in real replies. Each flag says why it was raised, and the sentence it came from is highlighted.
• Formatting tells. Lines that start with emoji or bullet points, a pattern people now read as AI. In reply mode it also checks length, links, hashtags and emoji.
• Quoting the post back. If you add the post you're replying to, it checks whether your reply repeats its wording instead of reacting to it.

How long should a reply on X be?

Short. Replies get read in a scroll, under someone else's post, so the ones that land are a single thought. These are the rules reply mode uses:

• Under 140 characters is the comfort zone. Over 140 gets a warning; over 240 is flagged as too long. Characters are counted the way X counts them, where a link always counts as 23.
• No links. A link in a reply reads as promotion.
• One hashtag at most, and no more than three emoji. An emoji dropped mid-sentence reads as decoration.
• Don't repeat the post back. Add the post you're replying to and the checker flags any run of words copied from it. Saying the author's point again isn't a reply.

How the score works

You get a human score out of 100. A "warn" flag lowers it; a "drop" flag means most readers would spot the draft as AI, so rewrite before posting. No flags means nothing in the draft matched a known tell. That isn't proof a person wrote it, but it won't read as AI to most people.

It's the same check Humanish runs on every draft it writes for you. It's free, needs no login, and takes up to 600 characters.