How to avoid AI detection in writing, the honest way
A lot of advice on how to avoid AI detection in writing is about tricks: swapping letters for look-alike characters, running a paraphraser three times, adding typos on purpose. Each one makes the text worse for the person reading it, which defeats the point. The approach that holds up is plainer. Find the habits that make a draft read as machine-written, and edit them out.
Readers and detectors notice the same things: predictable wording, even rhythm, stock phrases and a shape they've seen a thousand times. Below are ten of those habits with a before and after for each, all written for this guide. If you have a long draft and not much time, you can humanize AI text with LumenWrite first, which keeps every fact, number and citation in place, and then use this page for the final pass by hand.
Ten habits that make writing sound like AI, and how to fix them
One of these in a paragraph means little. Several at once is what makes a reader stop trusting the text.
1. The warm-up opening
AI drafts like to open by saying the topic matters. That first sentence could sit on top of a thousand other pieces, which is why it reads as written for none of them. Open with the most specific thing you have: a fact, a problem, a moment.
- Before: “Time management is an essential skill that plays a key role in both personal and professional success.”
- After: “Last Tuesday I checked my email 41 times before lunch and answered six messages.”
2. Stock words that say little
Chat models lean on a small set of polished words: showcase, underscore, comprehensive, multifaceted, intricate, a wide range of. None is wrong on its own. Together they make a sentence sound finished without saying anything. Replace each with the plain word or, better, with the specific thing it stands in for.
- Before: “The report showcases a comprehensive range of strategies that underscore the multifaceted nature of customer retention.”
- After: “The report lists five ways to keep customers. Two of them, faster refunds and a named account manager, did most of the work.”
3. Transitions that connect nothing
Additionally, Furthermore, Moreover, It is worth noting that, When it comes to. Delete one and reread the sentence. If the logic still holds, the word was filler. When you do need a link, use the short ones people say out loud: but, so, still, because.
- Before: “Additionally, it is worth noting that when it comes to budgeting, consistency is key.”
- After: “A budget only works if you check it every week, including the dull ones.”
4. Sentences that all run the same length
Read an AI paragraph aloud and you'll hear it: sentence after sentence of about the same length, built the same way. People write in bursts. Split a long sentence, merge two short ones, and let a four-word line land after a long one.
- Before: “The new schedule gives staff more flexibility in their working hours. It also helps the company reduce its overtime costs each month. Customers benefit from longer opening hours on weekdays. Overall, the change has been positive for everyone involved.”
- After: “Staff now pick a start time between 7 and 10. Overtime fell, mostly because the late shift stopped running over. Customers noticed too. The shop stays open until 8 on weekdays, and those evening hours are now its busiest.”
5. Everything in threes
Three adjectives, three benefits, three verbs in a row. Groups of three sound complete, so models reach for them by default. Real points rarely come in neat sets. Keep the one that matters and give it the room the other two were taking.
- Before: “The app is fast, intuitive and reliable, helping teams save time, reduce errors and boost productivity.”
- After: “The app's best feature is search. Type part of a client's name and their last invoice is on screen before you finish.”
6. The “not X, it's Y” setup
“It's not just a tool, it's a mindset.” The pattern invents a claim nobody made so it can correct it. Once in a long piece, it can work. Once a paragraph, it gives the draft away. Say the real point directly.
- Before: “Onboarding isn't just about paperwork. It's about making new hires feel like they belong.”
- After: “On day one, new hires get a laptop that already works and lunch with their team. The paperwork waits until Wednesday.”
7. Nouns where people should be doing things
AI drafts turn actions into abstract nouns: implementation, utilization, optimization. The people and the verbs drop out of the sentence. Ask who did what, and write that.
- Before: “The implementation of the new system resulted in a significant improvement in the efficiency of invoice processing.”
- After: “Since we switched systems, the finance team pays invoices in two days instead of five.”
8. Hedging every claim
“It could be argued”, “may potentially”, “in some cases”. One hedge where you're unsure is honest. Three in a sentence says the writer has no view. Decide what you think, say it, and hedge only the part you really don't know. Academic writing needs finer control here, which our guide on humanizing academic writing covers.
- Before: “Remote work may potentially offer some benefits for certain employees in some situations.”
- After: “Remote work suits people with long commutes. It suits new hires less, because so much of what they need to learn comes from conversations they overhear.”
9. Examples that could be about anyone
“Many businesses use social media to reach customers” is a category, not an example. A real example has a name, a number, a place or a moment. A model can only make those up, so this is where your own knowledge does the most good.
- Before: “For example, many small businesses have used social media to grow their audience.”
- After: “The bakery across from my office posts a photo of the first tray out of the oven every morning. By 8 a.m. the line is out the door.”
10. The recap at the end
“Ultimately”, “In summary”, “By following these steps, you can”. The last paragraph repeats what the reader just read and adds a moral. Cut it and see what's missing. Usually nothing is. End on the last concrete point, or on what happens next.
- Before: “Ultimately, by following these tips, you can create a workspace that boosts your focus and helps you achieve your goals.”
- After: “Start with the chair. Everything else on this list costs less and matters less.”
How to make writing sound less like AI in one editing pass
Fixing habits one at a time is slow. This order catches most of them in a single read.
- Cover the first and last paragraph. Read the draft without them. If nothing is missing, cut them or rewrite them from scratch.
- Read it aloud. Your ear catches flat rhythm and stock phrases faster than your eye. Mark every sentence you'd never say to a person.
- Search for the usual suspects. Use Find (Ctrl+F, or Cmd+F on a Mac) for additionally, furthermore, moreover, notably, showcase, underscore and comprehensive. Rewrite each sentence you find instead of swapping in a synonym.
- Break the rhythm once per paragraph. Make at least one sentence clearly shorter than the ones around it.
- Add one thing only you know to each section. A number from your own work, a name, a mistake you made, something said in the meeting.
- Check every fact. If AI wrote the first draft, check each number, name, quote and source against where it came from. Better style doesn't make a wrong fact right.
- See which sentences still stand out. Paste the text into our free AI detector to see which sentences read as AI, then go back to those lines. Use it as a pointer, not a verdict.
A checklist for your final read
Run through this before you send, submit or publish.
- The first sentence belongs to this piece and couldn't open anyone else's.
- No paragraph ends by restating what it just said.
- Every paragraph has at least one short sentence, and no three sentences in a row run the same length.
- No Additionally, Furthermore, Moreover or “it is worth noting”.
- Lists of three appear only where there really are three things.
- Each general claim has a concrete example.
- Verbs do the work in your sentences, not nouns like “implementation”.
- Hedges sit only on what you're really unsure of.
- Bullets, bold text and headings appear only where the format needs them. An email rarely does.
- The piece stops at its last point.
- Every fact, number and source has been checked.
Can you make AI writing undetectable?
No one can honestly promise that. Detectors use different models and thresholds, they change with updates, and two of them can give the same text very different scores.
They also misjudge human writing. When we ran 40 texts through ZeroGPT on October 2, 2026, it caught all 20 raw AI texts and cleared the 14 modern human ones. It also scored six famous older texts, including the Gettysburg Address and Pride and Prejudice, at 73 to 100% AI. Our ZeroGPT test has the full results. If the Gettysburg Address can fail a detector, no edit can guarantee a pass.
What editing does reliably is make the writing better for the people who read it. Detectors look at many of the same patterns, but each weighs them differently, so no edit comes with a predictable score.
If the writing is for a class, editing doesn't change what your school allows. Follow your institution's rules on AI use, and disclose AI help where they ask you to. These edits are for drafts where AI help is permitted, and for your own writing when it comes out stiff.