Nobody Talks Like That

Writing check

Make your writing
clearer, stronger
and more human.

A practical writing check that spots the words, phrases and patterns that make your text sound generic, padded or AI-generated.

What it does

  1. Spot the problems

    Find the lines that weaken your writing.

  2. Understand why

    See what's wrong and why it reads that way.

  3. Make it better

    Get clear, practical suggestions.

  4. Build your skills

    Learn the rules in the book and improve over time.

Not an AI detector

It reads the writing, never the writer. What comes back says which lines sound generic and why, and nothing about who typed them.

How it differs
0 words At least 40 words to check. 300 or more gives a fairer reading. 1,000 free words to use

A high score is a reason to edit and never a reason to accuse anyone, because nothing here can tell you who wrote a sentence.

What a finding looks like

Every line it marks comes back in the same shape. Here's one in full, then four more.

Opening · failed

In today's fast-paced world, staying on top of your finances has never been more crucial.

Why it failed. The first sentence is a warm-up. It buys time to sound weighty, and a reader's eye slides straight past it looking for the actual claim.

Try insteadCut it and open on the first sentence that says what the thing does. Nothing is lost, because the line carried nothing.

Evidence · failed

Studies show that people who track their spending save more.

Which studies? A claim with nobody's name behind it is an opinion in a borrowed coat, and readers have learned to hear it as 'I couldn't be bothered to look'.

Try insteadName the study and the figure, or own it as your view: 'In my experience, people who track their spending save more.' If you can do neither, cut the claim. It never gets a made-up source.

Sentence shape · failed

No spreadsheets. No complexity. Just insight.

A run of denials and then a reveal that reveals nothing. It's a writer saying what something isn't because they haven't yet worked out what it is.

Try insteadOne plain sentence with the real number in it: 'It sorts every line of a statement into [how many] categories.' The number is yours to supply.

Ending · failed

Ultimately, financial freedom is a journey. Start yours today.

A moral stapled to the last line to give the piece a shape it hadn't earned. Readers can feel the reach, and they trust everything above it a little less.

Try insteadCut it and end on the last useful fact, which is usually the price or how to start.

The whole piece · failed

No name, no figure and no date anywhere in eight hundred words.

This is the one that decides it. Generated text is generic because it has no lived experience to draw on, so it writes the average of everything it has read.

Try insteadPut in the number, the person, the month it happened. One real detail rescues a whole paragraph.

Every rewrite uses only what's already in your writing. Where the fix needs a fact only you have, it leaves a gap for it rather than inventing one, because a made-up figure is worse than the line it replaced.

The rules

Twenty six of them, in four groups, and four more that count in your favour. They stay on the server; what comes back to you is the edit rather than the rulebook.

The words

The words people reach for when they've nothing particular to say, from the stock phrase to the borrowed authority of a study nobody names.

Sentence shape

The shapes sentences keep falling into, such as saying what a thing isn't before what it is, or a colon used as a drum roll before a reveal.

Openings and endings

The warm-up first sentence that delays the point, and the moral stapled to the end to give a piece a shape it hasn't earned.

The whole piece

The faults you only see standing back from it: no figure, name, date or place anywhere, and every paragraph the same size, so nothing seems to matter more than anything else.

Four that count in your favour

These are the ones most guides leave out: a detail only you would know, a voice, an uneven shape, and a claim you could be wrong about. Those four pull your score down, by as much as half, which is why a piece with a few lapses and plenty of real detail can still land in the clear.

The full set is written out in the book. If you've read it and counted twenty seven, you counted right. One of those, placeholder text left in a finished piece, is a proofreading slip rather than a habit of voice, so the checker leaves it out.

How this differs from an AI detector

ToolThe question it answersWhat you get backWhat it's for
AI detectorsDid a machine write this?A percentageSuspicion
AI humanisersCan I get this past a detector?Reworded textEvasion
ThisDoes the writing hold up?The lines that fail hardest, and whyPublishing something with your name on it

AI detectors try to work out whether a machine wrote something. No test can settle that for a single piece of writing, and the research so far says the people they flag hardest are academics and anyone working in a second language. This doesn't try. It reads the writing itself, points at the lines that read as generic or assembled, and says what to change, so what comes back is editorial feedback rather than a verdict on who wrote it. Humanisers do a job I'd rather not be in, which is helping text slip past a checker without getting any better. This one holds your writing to a standard I wrote down, shows you the lines that miss it by the most, and you're welcome to argue with the standard on the log.

A grammar checker won't catch any of this either, because none of it is an error. 'In today's fast-paced world' is flawless English. So is a claim with no source behind it, and a paragraph that happens to be the same length as the four before it. Grammar checkers are built to make writing correct and clear, and correct and clear is exactly what generated text already is.

How it works

Every rule gets a score out of three, and so do the four that count in your favour. The arithmetic turns that into a number out of a hundred where lower is cleaner, plus a band, and anything from 0 to 19 is fine to send. The first few habits carry most of the score, because a reader only needs a few to make their mind up, and the four in your favour can pull it down by as much as half.

It runs on a language model, so two runs on the same piece can land a few points apart. If both land in the same band, trust it. If they don't, the piece is sitting on a boundary and wants another pass.

What it costs

The first 1,000 words are free and there's nothing to sign up for. Paid checks aren't switched on yet. When they are, the price will be written here in plain figures before you're asked for a card, it'll be a monthly subscription for as many checks as you need at up to 3,000 words each, and you'll be able to cancel from this page.

Privacy

This site doesn't store what you paste. Your draft goes to the model provider, Anthropic, to be scored under terms that don't allow it to be used for training, and the result comes back to your screen. The count of free words you've used is kept in your own browser, and if you subscribe, the pass that proves it is kept there too. Payment is handled by Stripe, so card details never touch this site.