How to rewrite an AI draft so it sounds like your brand

Why do AI drafts all sound the same?

Large language models write by predicting a likely next word, again and again. That is why their drafts are fluent and why they drift towards the middle: the safest opening, the most familiar metaphor, the tidy three-part list. Ask five tools for a page about your accounting software and you will get five versions of the same polite, slightly breathless text.

The result is rarely wrong in an obvious way. It is just unowned. It could sit on any competitor’s site without anyone noticing, and readers sense that quickly. They skim, they stop trusting the claims, and the page does nothing for the brand that paid for it.

So the job is not polishing. It is replacing the generic parts with things only you could say: your numbers, your process, your opinions, your customers’ words. AI can still save time on structure and coverage. The voice has to come from you.

Write the voice down before you edit

Editing by feel works for one writer on a good day. It falls apart when three people edit and a tool drafts. A short voice sheet, one page at most, gives everyone the same reference. It should cover:

• Three or four traits, each with a limit. “Confident, not boastful.” “Plain, not blunt.” Traits without limits are useless, because every brand claims to be friendly.
• Words you use and words you never use. Your product names, how you refer to customers, and a banned list of filler such as “seamless”, “cutting-edge” and “unlock”.
• Sentence habits. Typical length, whether you use contractions, how you address the reader, and how you handle numbers and dates.
• Two before-and-after examples. One paragraph in a generic voice, rewritten in yours. Examples teach faster than adjectives.
• Language rules. Which pages run in Danish, which in English, and whether you write British or American spelling.

Keep the sheet next to the editor and paste it into your prompts. It will not make the tool write like you, but it narrows the gap.

Edit in four passes, not one

Trying to fix structure, facts and tone at once is how drafts end up half edited. Separate passes are faster overall, because each one has a single question.

1. Structure. Does the order match what the reader needs? Delete the warm-up paragraph and the closing summary that repeats everything. Move the answer to the top. Merge sections that say the same thing twice.
2. Substance. Underline every claim. Replace each vague one with a fact, an example or a number you can stand behind, or cut it. Check anything that looks like a statistic, a law or a quote, since models produce confident errors.
3. Voice. Go sentence by sentence with the voice sheet. Swap banned words, adjust formality, and add the phrases your team really uses with customers.
4. Rhythm. Read it aloud. AI text often has sentences of near-identical length. Break some, join others, and let a short sentence land after a long one.

If the second pass removes more than half the claims, consider writing the piece yourself from the outline. That is a perfectly good outcome. Our AI text removal work often starts with exactly that decision.

A short before and after

Take a typical AI paragraph for the services page of a made-up Copenhagen accounting firm:

“In today’s complex financial landscape, businesses need a trusted partner. Our dedicated team leverages cutting-edge tools to deliver seamless, tailored solutions that empower you to focus on what truly matters: growing your business.”

Nothing in it is false, and nothing in it is useful. It names no service, no customer and no reason to pick this firm. Here is the same paragraph after the four passes, built on the kind of facts a real firm would supply:

“We do bookkeeping, VAT returns and annual accounts for about forty small companies in Copenhagen, most of them with fewer than ten staff. You get one named accountant, a reply within a working day, and a fixed monthly fee agreed before we start.”

The rewrite is shorter, more specific and easy to check. Every claim in it would come from the business, not the tool. That is the standard to aim for: if a sentence could appear on a competitor’s site unchanged, it is not finished.

The patterns that give a draft away

Certain habits appear so often in machine-written text that readers now notice them. None is wrong on its own. Together they read as filler.

Do not just run a find-and-replace. Removing the words while keeping the empty sentence structure leaves text that is still generic, only harder to spot. If a sentence has nothing in it once the filler is gone, delete the sentence. Ownership matters too: our phrasing clean-up service also looks at whether copy could be mistaken for something a tool produced for someone else.

Danish drafts need their own checks

Most tools write Danish well enough to pass a quick glance, but it often reads like translated English. Nordic readers notice. Look for:

• Tone of address. Danish business writing almost always uses “du”. Drafts that drift into a stiff, formal register, or that switch between forms, sound foreign.
• Compound words. Danish joins compounds into one word. Tools, like many people, sometimes split them in the English way, which looks careless to a Danish reader.
• Anglicisms and calques. Literal translations of English idioms, and English word order in subordinate clauses, are common giveaways.
• Understatement. Danish audiences tend to distrust superlatives more than English-speaking ones do. A claim that sounds normal in American marketing copy can sound like bragging in Copenhagen.

If you publish in both languages, do not generate one and machine-translate the other. Write or edit each with a native speaker, using the same facts and the same voice sheet. The pages can say the same thing without being word-for-word copies, which also reads better to search engines treating them as separate versions.

Making it honest and repeatable

A single good rewrite is useful. A process that produces good rewrites every week is what changes a content programme. Three habits help.

First, keep a record. Note which pieces started as AI drafts, who edited them and what sources were checked. That protects you if a fact turns out to be wrong, and it helps if you ever need to show how much human work went into a text.

Second, give the tool better input. Paste the voice sheet, your outline, real customer quotes and the facts you want included. The closer the draft is to your material, the less there is to replace.

Third, train the people, not just the prompt. Editors who understand why AI text sounds the way it does fix it faster. If the volume is high and the team is small, it can make sense to hand the first edits to specialist copywriting help and keep final approval in house.