AI writing in British English, as standard

By Jack Stovell X · Instagram · LinkedIn · published 2026-08-09 · updated 2026-09-30

You've told the model to use British English. It still writes like it's from Ohio.

Spelling is the easy half

"Use British English" fixes organise, colour, whilst. Fine. Job done, apparently.

Except it isn't done. The model swaps -ize for -ise and calls it a day, then hands you "I'm excited to share" and three exclamation marks in a row. That's not British. That's American enthusiasm with a UK spellcheck bolted on.

Register is the tell

British English isn't a spelling list. It's a whole different relationship with confidence, understatement, and how much you're allowed to sound pleased with yourself.

  1. Americans say "amazing results." You say "not bad, all things considered."
  2. Americans open with the good news. You bury it, politely, on paragraph two.
  3. Americans use exclamation marks. You use "quite."

A spelling toggle can't fake that. A voice profile can, because it's not guessing at "British", it's measuring how you actually write.

UK English as a setting, your English as the profile

ScriptGrain runs UK English as a first-class setting at generation, not a find-and-replace pass afterwards. That handles the accent.

Then it goes further. It runs a two-pass stylometric analysis across 45 attributes of your actual writing: sentence rhythm, contraction rate, hedging, how often you reach for "whilst" versus "while", whether you undersell or oversell. If your English says organise and means it, understates and means that too, the profile captures it and enforces it on every draft after.

Every draft you generate comes back with a voice-match score, so you're not guessing whether it landed.

Sterling pricing, no card required

Build one profile free: one full analysis, no card, see what it actually finds.

Priced in sterling. Run by a UK company. British English isn't a checkbox we added, it's the standard we built on.

Try it free

Build your free profile. One analysis, no card, five minutes.

You'll see your 45 attributes laid out plainly: where you hedge, where you understate, where you'd never in a million years say "I'm thrilled to announce."

Fair enough if you want to check it against something you've already written. That's exactly what it's for.

Spelling was never the hard part. Sounding like yourself, in your own English, is.

The British habits a model gets wrong

Start with the exclamation mark. A model trained on mostly American text treats it as a warmth dial, so a friendly email gets two, a product note gets three. British writing spends them like a nervous tipper. One, occasionally, and only when something really has earned it.

Then hedging. Americans hedge with "I think" or "maybe". You hedge with "rather", "a bit", "fairly", "I'd have thought". Compare the two:

"This approach is really effective and will transform your workflow!"

"This works rather well, and it should save you a bit of time."

Same claim. Different country.

Understatement is the real giveaway. "Not bad" can be top marks. "Quite good" is warmer than it sounds, and "a little disappointing" means someone should probably apologise. A model that reads these literally will flatten every one of them into praise.

Then the boring conventions, which bite anyway. Dates run day, month, year: 3 March, not March 3. Quotation marks are single by default, with the full stop outside unless it belongs to the quote. Sign-offs get drier too. "Best wishes" and "Cheers" beat "Excited to connect!" every time.

So the fix isn't a rule list. It's knowing which of these your own writing actually does.

Checking a draft for register as well as spelling

A spellchecker can tell you that "color" is wrong. It can't tell you the draft sounds thrilled when you never would.

Register needs measuring. ScriptGrain scores every draft from 0 to 1 against your profile, and the score comes back with per-feature deltas: the draft's value against yours, attribute by attribute. On-voice drafts typically land between 0.85 and 0.95. Partial matches sit at 0.5 to 0.8, and a clearly different voice falls under 0.45.

Here's the useful bit. Say a draft scores 0.72. The deltas might show an exclamation rate well above your baseline, a confidence-versus-hedging balance that's too bullish, and formality drifting up a couple of points. Those are register faults, and none of them would trip a spelling check. You'd fix them by hand, or let Polish revise the draft toward a target score (the default is 0.9).

The profile itself covers 45 attributes in 8 layers, including punctuation habits and contraction rate, plus how you build an argument. Extraction usually takes under a minute. The method is published, so you can see how the scoring works rather than taking a number on trust.

To be fair, a score isn't a verdict on quality. It tells you whether the draft sounds like you. Whether it's any good is still your call.

Writing for readers on both sides of the Atlantic

Plenty of people write for both audiences. A UK agency with American clients. A founder selling in Manchester and Massachusetts. One voice, two spellings.

That's what the English setting is for. Every draft can be written in British or American English (`english_variant` set to uk or us), and the profile keeps the spelling your samples use. The variant changes the accent. Your rhythm, your hedging, your habit of undersell or overstatement stay where they were.

Spelling is only the start of it. British and American English also split on vocabulary, punctuation and a handful of grammar habits, and some of those a machine can check while others it can't. We set the variant on the profile itself, so you never have to ask for it in a prompt. The stylistic markers you can measure are laid out in full, and that's where you'll see how it works.

There's a US edition too, priced in dollars, for readers who'd rather see their own currency. The default here remains British English, and the company behind it is a UK one.

A practical tip. Build the profile from writing in the variant you use most, then switch the setting per piece. If a US client wants "organize" and "color", flip it and generate. Don't rewrite the profile just because the audience moved.

One caution, and it's a fair one. Switching variant shouldn't switch personality. "Whilst" becomes "while" for an American reader, and the sentence should still sound like you wrote it. If the deltas show your confidence level jumping when the setting changes, that's a draft worth a second look.

Same voice, different passport.