✂️ Stop AI Writing Like AI: The Plain-English Fix

Something quietly happened over the last couple of years: everyone learned to smell AI writing. Your customers, your boss, the parents in the group chat - they might not be able to say why a message feels machine-made, but they feel it, the way you can tell a stock photo from a holiday snap. And the moment they smell it, something worse happens: they stop reading properly, because it reads like something nobody actually meant.

Here’s the good news. AI text isn’t unfixable, and you don’t need to become an editor. The robot voice comes from a handful of specific, learnable habits - I call them the Four Tells - and once you can name them, one paste-in prompt strips all four from any text in seconds. Spot, paste, fixed.

(Quick signpost: this guide fixes any AI text so it sounds human in general. Making it sound like you specifically is a different job, covered in a different guide, linked at the end.)

Tell 1: The greatest-hits vocabulary

What it is: AI has favourite words the way a wedding DJ has favourite songs, and it plays them at every event. Words like seamless, robust, comprehensive, crucial, streamline, landscape (when nobody’s outdoors), navigate (when nobody’s driving), and openers like “In today’s fast-paced world”. Individually innocent. Collectively, a flashing sign.

How to spot it: read the text and ask “would anyone say this word across a table?” Nobody has ever told their neighbour about a robust scone or a seamless school run.

The non-obvious bit: don’t go hunting with somebody else’s list of banned words - the internet is full of them and they age fast, plus half the entries might be words you genuinely use. The test isn’t “is this word on a list?”, it’s “would the writer say it out loud?” That test never goes out of date.

Tell 2: The rule of three

What it is: AI adores triplets. Everything is “faster, simpler and more reliable”. Every benefit comes in threes, every paragraph rounds off with a tidy trio. It’s a rhythm borrowed from speechwriting, and in an email about the school cake sale it sounds deranged.

How to spot it: count. Once you notice the threes, you cannot stop noticing them - in AI text, in adverts, everywhere. (Sorry. This is permanent.)

The non-obvious bit: real people list two things, or four, or one thing said twice because they’re annoyed. When you break a triplet, don’t just delete the third item - usually the sentence wanted to be one plain claim all along. “It’s faster” was the whole point; the other two were upholstery.

Tell 3: The essay reflex

What it is: you ask a question, you get a dissertation. Headings, an introduction that restates your question, a balanced exploration of considerations, and a conclusion beginning “Ultimately…”. You wanted “yes, Tuesday works”.

How to spot it: the length is wildly out of proportion to the ask. A one-line question that returns four paragraphs has failed, however good the paragraphs are.

The non-obvious bit: the essay reflex is the tell people forward to their mates as a joke. Someone asks “can you do Thursday instead?” and gets back a structured overview of scheduling considerations. If your reply couldn’t be read aloud in one breath where one breath would do, it’s still wearing the robot suit.

Tell 4: The butler sign-off

What it is: the over-formal frame around an otherwise normal message. “I hope this email finds you well.” “Please don’t hesitate to reach out should you require further assistance.” “I trust this clarifies matters.” Nobody’s inbox has ever been improved by being found well.

How to spot it: look at the first line and the last line - the tells cluster at the edges. The middle of AI text is often fine; the bookends are pure butler.

The non-obvious bit: when you cut the butler, resist replacing him with forced chirpiness (“Hey hey! Happy Monday!!”), which is just a different costume. The human version is usually nothing at all: start with the point, end with “Thanks, [name]”. Endings are allowed to just… end.

The De-Robot Prompt

Here’s the fix for all four tells in one go. Paste it, then paste your text.

Nick this one:

Rewrite the text below so it sounds like a real person wrote it. Rules:
1. Cut stock phrases and buzzwords - any word I wouldn't say out loud across a table goes. If it sounds like a press release, rewrite it.
2. Break up lists of three. If a sentence makes three claims, keep the one that matters.
3. Match the length to the job. If the point fits in two sentences, two sentences is the answer - no headings, no intro, no summing-up.
4. Kill the formal bookends. No "I hope this finds you well", no "don't hesitate to reach out". Start with the point, end simply.
5. Keep every fact, number, date and name exactly as they are.
6. Plain words over posh ones: "use" not "utilise", "help" not "facilitate", "buy" not "purchase".
7. Keep the intent intact: a firm message stays firm, a request stays a request.
Then show me only the rewrite, no commentary.

Here's the text: [PASTE]

And the one-liner for when you’re mid-conversation and the robot voice creeps in:

The quick version:

Again, but how you'd say it to a friend across a table - half the length, none of the polish.

The quick version works because Claude already knows what it just wrote - you’re not teaching it style, you’re giving it permission to relax. You’ll be surprised how often the second attempt is the message you’d have written yourself on a good day, if you’d had the patience.

A worked example

Say you run a small bakery and Claude drafted your new-opening-hours announcement. It came back with: “Exciting news! We’re thrilled to announce that we’re enhancing our opening hours to serve you better, offering greater flexibility, convenience and choice. We can’t wait to welcome you! Don’t hesitate to pop in!”

Through the De-Robot Prompt: “We’re now open until 6pm on weekdays - lots of you asked for time to grab a loaf after work, so here it is. Saturdays stay the same. See you at the counter.”

Same facts. One sounds like a leaflet nobody reads. The other sounds like the person who actually makes your bread.

A dose of honesty

The De-Robot Prompt can overshoot: for a solicitor, a landlord dispute or a formal complaint, some starch is correct, so tell it “keep it professional, just make it human”. It also can’t fix content problems - if the draft says nothing, a natural-sounding version of nothing is still nothing. And you should still read every rewrite before sending, because “sounds human” and “says what I meant” are different tests, and only you can run the second one.

The one to start with tonight

Memorise the quick one-liner. It’s nine seconds of typing, it works mid-chat on anything, and it fixes maybe 80% of robot voice without you ever needing to name a single tell.

This gets any text sounding like a person. To get it sounding like you - your phrases, your rhythms, your personal ban list built from your own messages - the next step is Teach AI To Write Like You.