How to Write Authentically With AI (Without Gaming Detectors)

Ten writing techniques, each with a before and after, for making AI-assisted work clearly yours. Better writing and real ownership, not detector tricks.
How to write authentically with AI comes down to one rule: the thinking has to be yours. Decide your point before you prompt, give the AI narrow jobs, then rewrite until every claim, example and judgement is one you chose and could defend out loud. Be open about what the tool did. Your voice follows from that.
This guide gives ten techniques, each with a short before and after, plus a step-by-step workflow for turning an AI draft into work you own. One thing up front: none of this is a method for disguising AI text. These habits make writing better and more clearly yours. If a passage came from ChatGPT, Claude or Gemini and you kept it, the honest move is to say so, not to sand off the fingerprints.
How to write authentically with AI: what does it actually mean?
Authentic writing is not "text no AI ever touched". Plenty of people type every word themselves and still produce prose that could have come from anyone. And a writer can start from an AI paragraph, tear it apart and rebuild it into something nobody else would have written.
What makes writing yours is whose judgement shaped it. Three questions get at that better than any detector score:
Could you explain why every paragraph is there? Not what it says, but why it belongs in this piece and in this order.
Do the examples and evidence come from you? Your reading, your notes, your data, your experience, sources you actually checked.
Could you describe your AI use to your teacher or editor without squirming? If the answer is yes, you are probably on the right side of the line. If you would rather they didn't ask, that tells you something.
Human writing, in the sense readers care about, is writing where a person made the decisions. Typing is just the delivery mechanism.
Why not just write to beat AI detectors?
Because detectors guess authorship from style, and that guess is shaky in both directions. Writing to satisfy one means aiming at a moving target that also misfires on honest work.
The track record is worth knowing:
OpenAI withdrew its own detector. In July 2023 OpenAI retired its AI text classifier "due to its low rate of accuracy." Its own evaluation had found it correctly flagged only 26% of AI-written text while wrongly labelling human text as AI 9% of the time.
Plain writing gets flagged. A 2023 Stanford study by Weixin Liang and colleagues, published in the journal Patterns, found that popular detectors consistently misclassified writing by non-native English speakers as AI-generated. The Markup reported that the seven detectors tested flagged that writing as AI 61% of the time, while rarely making the same mistake with native speakers.
Mixed documents are the hardest case. Turnitin said in 2023 that its sentence-level false positive rate was around 4%, and that wrongly flagged sentences were more common in documents mixing human and AI text, especially at the transitions. That is exactly the kind of document you produce when you edit an AI draft.
Small error rates hit real people. When Vanderbilt University switched off Turnitin's AI detector in August 2023, it pointed out that even a 1% false positive rate across its roughly 75,000 yearly submissions would mean around 750 students wrongly flagged.
Detectors are also getting better. A 2025 working paper from the University of Chicago's Booth School of Business found large differences between tools, with the strongest commercial detector producing almost no false positives in its tests and weaker free tools doing badly. So the tricks that fooled last year's detector may not fool next term's.
The bigger problem is what those tricks do to your writing. Swapping in synonyms, adding deliberate typos or running text through a "humanizer" changes the surface and leaves the thinking exactly where it was: with the tool. The result usually reads worse. And a passage you deliberately disguised is much harder to explain in an academic integrity meeting than one you disclosed.
Where does your voice go when AI drafts for you?
Two recent studies show what happens to a writer's voice when AI does more of the work. Neither is the last word, but both point the same way.
AI pulls writing toward the average. Cornell researchers presented a study at the CHI 2025 conference in which 118 participants from India and the United States wrote with and without an AI writing assistant. With AI suggestions, the two groups' writing became more alike, mostly because the Indian participants' writing drifted toward American styles. Lead author Dhruv Agarwal said Indian users started describing their own festivals and food "from a Western lens."
AI weakens the sense that the work is yours. In a June 2025 MIT Media Lab study led by Nataliya Kosmyna, 54 students wrote essays using ChatGPT, a search engine or no tools. Self-reported ownership of the essays was lowest in the ChatGPT group, and those students struggled to quote their own work: a reported 83% could not accurately quote a sentence from the essay they had just written, against about 11% in the other groups. The paper had not been peer-reviewed, and the authors asked journalists not to overstate it, but the ownership finding matches what many writers describe.
Those are the two forces to push against: genericness, and distance from your own text. Every technique below targets one or both.
10 writing tips to sound more human (and more like you)
Each tip has a one-line principle, a short explanation and a small before and after. The examples are invented to show the edit, not drawn from anyone's real essay.
1. Decide what you think before you open the chat
Principle: your position should exist before the AI's does.
Ask for "an essay on remote work" and you get the median view of the internet. Write one rough sentence stating what you believe and two reasons, before any prompting. Then whatever the AI produces is material for your argument, not a replacement for it.
Before: Remote work offers both significant benefits and notable challenges for modern organisations.
After: Remote work suits teams whose output is easy to see, like support desks, and hurts teams that learn by overhearing, like new engineers.
The second version could be wrong. That is the point. It is a claim a person made, and the rest of the essay has a job: to prove it.
2. Use the example only you have
Principle: replace the example anyone could give with one from your notes, reading, work or life.
AI reaches for textbook examples because they are the most common ones in its training data. You have better material: the case study from week four, the figure from your lab's dataset, the client email that changed your mind. One rule here is non-negotiable. Never invent an experience to sound personal. If you have no first-hand example, use a specific source you actually read.
Before: For example, many companies have struggled with communication when moving to remote work.
After: The week my team switched to written stand-ups, three questions sat unanswered for two days because each of us assumed someone else would reply.
3. Show your reasoning, not just the conclusion
Principle: write the "because", including the step you almost skipped.
AI drafts often jump from evidence to verdict with a "therefore" in between. Your marker or editor wants to see how you got there. Spell out the middle step, especially if it is the one where you changed your mind.
Before: Therefore, the late-submission policy is ineffective.
After: The policy cut average lateness by a day, but the students it was meant to help, the ones with part-time jobs, were still late. It worked on paper and missed its target.
4. Say how sure you are, and why
Principle: specific uncertainty sounds like someone who checked; blanket hedging sounds like no one.
AI text tends to swing between confident assertion and vague hedges like "may potentially". Neither tells the reader anything. Say where your evidence is strong, where it is thin, and what would change your view. Honest uncertainty is one of the clearest signs of a real writer.
Before: This approach could potentially have a significant impact on outcomes.
After: I'd expect this to raise pass rates, but the only study I found covered a single school, so I wouldn't trust the size of the effect.
5. Vary the rhythm on purpose
Principle: let sentence length follow the thought.
Unedited AI prose often settles into sentences of similar length and shape, one after another. People write in bursts. A long sentence carries a complicated idea; a short one lands a point. Read your paragraph and look for three sentences in a row built the same way.
Before: The library extended its opening hours to support students during exams. This change was welcomed by many students who needed quiet study space. However, the extended hours also increased staffing costs for the university.
After: The library stayed open until 2 a.m. during exams, and students filled it. It also cost money. Someone had to staff those extra hours, and the bill came out of the same budget as the new books.
6. Cut the warm-up and the recap
Principle: start where the content starts and stop when it's done.
AI drafts love an opening that sets the scene and a closing that restates everything. Delete both and see what you lose. Usually it is nothing, and the piece now opens on an actual claim.
Before: In an era where technology shapes every aspect of our lives, the question of screen time has become increasingly important. This essay will explore the various perspectives on this complex issue.
After: Most screen-time limits fail because they count minutes instead of asking what the minutes are spent on.
7. Use the words you'd actually say
Principle: if you wouldn't say it to your tutor across a desk, don't write it.
Some words turn up far more in AI drafts than in most people's writing: "pivotal", "multifaceted", "underscores", "fosters", "plays a crucial role". None is wrong, but a page full of them sounds like no one in particular. The Cornell finding applies here too: name your own things in your own terms rather than translating them into a generic register.
Before: The festival plays a pivotal role in fostering a sense of community.
After: During Onam, my grandmother's whole street cooks together, and nobody keeps track of who brought what.
8. Take a side, then answer the strongest objection
Principle: balance is not the same as having no view.
Chatbots are tuned to be even-handed, so their drafts often end with "there are valid arguments on both sides." In an essay or opinion piece, that is a non-answer. State your position, name the best argument against it fairly, and explain why you still hold your view.
Before: There are valid arguments on both sides, and the best approach depends on the specific context.
After: Courses should allow AI for brainstorming but not for final prose. The strongest objection is that the line can't be enforced. I think it can, roughly, if students hand in their drafts and notes along with the essay.
9. Let the content choose the structure
Principle: not every idea comes in threes with a bold heading.
AI drafts have recognisable habits: lists of exactly three, the "it's not just X, it's Y" construction, a heading for every two paragraphs, a final paragraph beginning "Ultimately". If your argument has two parts, write two. If one point matters more than the others, say so instead of giving them equal billing.
Before: Effective feedback is timely, specific and actionable. It's not just about criticism; it's about growth.
After: Feedback that arrives after the next assignment is due is useless however good it is. Fix timing first. Specificity only matters once students get comments in time to use them.
10. Read it aloud and see if you can defend it
Principle: if you can't explain a sentence without the screen in front of you, it isn't yours yet.
This is the direct answer to the MIT ownership finding. Close the document and explain your argument to a friend in two minutes. Then pick one sentence you are proud of and say it from memory. Any sentence you stumble over, or can't explain when asked "meaning what?", needs rewriting or cutting.
Before: This underscores the multifaceted nature of the issue.
After: Two things drive this: rent and the length of the commute.
How to add your own voice to AI draft text, step by step
The ten tips work best inside a process that keeps you in charge from the start. Here is a workflow I'd recommend for an essay, report or article where AI use is allowed:
Write a scrappy brief before you prompt. Your claim in one sentence, two or three reasons, any examples you already have, and what you are unsure about. Ten minutes of bad notes beats an hour of polishing someone else's argument.
Give the AI narrow jobs. Ask it to critique your outline, list counterarguments, explain a concept you are stuck on, or draft one paragraph you plan to react to. Paste your brief into the prompt so its output starts from your ideas.
Keep AI text visibly separate. Whatever tool you write in, make sure you can still tell which passages came from the model. In Assist Writer that means pasting with Ctrl+Shift+V so the text is marked as AI.
Edit for argument first. Delete anything you don't believe. Reorder paragraphs into your logic, not the model's. If a section exists only because the AI included it, ask whether your piece needs it.
Edit for evidence second. Swap generic examples for your own (tip 2). Check every fact, quote and citation against the original source, because language models can produce references that look real and aren't.
Edit sentences last. Now apply tips 4 to 9: rhythm, vocabulary, hedging, structure. Doing this first is the common mistake; polishing sentences you later delete wastes your time.
Test and disclose. Do the read-aloud test from tip 10, compare the result with your original brief, and describe your AI use the way your course or publication asks.
What if one AI paragraph is better than anything you'd write?
It happens, especially with definitions or summaries of background material. You have two honest options. Keep it and say it came from AI, or rewrite it from your understanding and accept that it might be a little plainer. The dishonest third option, keeping it and passing it off as yours, is the one these techniques can't fix.
How to make sure your essay sounds like you
Teachers who know your work notice a sudden change in voice long before any software does. The goal is not to sound "less like AI"; it is to sound like the person who wrote your last five assignments, only better.
Make a one-page voice sheet
Take two or three things you wrote before you used AI much: old essays, long emails, a blog post. Note what you actually do:
How you usually open (a question, a claim, a story?)
Your typical sentence length, and whether you use dashes, semicolons or brackets
Words and phrases you use often, and ones you never use
How you disagree with a source (politely, bluntly, with humour?)
What you tend to over-explain or skip
Keep the sheet next to you while editing. When an AI sentence clashes with it, the sentence goes, not the sheet.
Check against your old work
Put a paragraph of the new essay beside a paragraph of an old one. If they read like different people, find the reason. Often it is vocabulary (tip 7) or the disappearance of your own examples (tip 2). Sometimes the new piece is genuinely better because you have improved, which is fine. Growth is not the same as replacement.
If English is not your first language
Don't polish your voice away. The Stanford study above suggests that plainer vocabulary and simpler sentence patterns are part of what gets non-native writing misflagged, and the Cornell study suggests AI suggestions push writers toward an American default. Neither problem is solved by making your prose fancier. Clear, direct writing in your own words is good writing. If you are worried about a false accusation, keep evidence of your process instead of changing your style.
Keep the evidence of your process
Notes, outlines, early drafts and a record of what the AI contributed all show how the work came together. They matter far more than any single score if someone ever questions your authorship.
Which edits make writing yours, and which only hide AI?
Not all editing is equal. Some changes move the thinking from the tool to you. Others only change what the text looks like.
Edit | What it changes | Does it make the work more yours? |
|---|---|---|
Running text through a "humanizer" or paraphrasing tool | Surface wording | No. The ideas and structure are still the tool's, and meaning often gets garbled. |
Swapping words for synonyms by hand | Surface wording | Barely. Same argument, slightly different paint. |
Adding typos or odd punctuation on purpose | Surface wording | No, and it makes the writing worse. |
Replacing generic examples with your own | Substance | Yes. This is often the biggest single improvement. |
Rebuilding the argument in your order | Structure and reasoning | Yes. The logic becomes yours. |
Adding your reasoning and specific caveats | Thinking | Yes. Shows the reader how you got there. |
Cutting filler and recaps | Length and clarity | Partly. Better writing, though not new thinking. |
Disclosing AI use and keeping a record | Transparency | It doesn't change the text, but it makes your share visible. |
A useful rule: if an edit could be done by a tool without understanding your topic, it probably isn't making the work more yours.
What mistakes make writing feel less authentic?
Inventing personal stories. Adding a made-up anecdote to "sound human" is fabrication, and it is worse than a generic example. Use real experience or a real source.
Overcorrecting into fake casualness. Sprinkling in "honestly", slang or exclamation marks does not create voice. If you don't write like that normally, it reads as a costume.
Editing only the opening. Many people rewrite the first paragraph carefully and leave the middle untouched. The result has an obvious seam where your voice stops. Turnitin's 2023 note that false flags cluster at human-AI transitions suggests a half-edited document can even cause trouble for the sentences you did write.
Trusting the AI's facts. A sentence in your voice that contains an invented statistic is still wrong. Checking sources is part of owning the text.
Imitating an "academic voice" that isn't yours. Formal does not mean inflated. Most markers prefer a clear sentence to an impressive one.
Never using AI's suggestions at all. Being authentic doesn't require refusing help. A counterargument you hadn't considered, or a clearer way to explain a concept, can make the work better as long as you decide what to do with it.
How can you show which parts of your writing are yours?
The techniques above make your contribution real. The remaining problem is making it visible, especially when a teacher, client or editor wants to know what the AI did.
That is what we built Assist Writer for. It is a browser-based editor that records how text entered your document, rather than guessing from style afterwards:
AI text is marked when you paste it. Paste with Ctrl+Shift+V (Cmd+Shift+V on Mac), or right-click and choose "Paste as AI Text", and that text is shown in a colour spectrum. Normal typing and normal paste count as your own text.
Your rewriting shows. As you edit AI text, its colour gradually shifts toward normal text, so a paragraph you rebuilt using the steps above looks different from one you left alone.
A stats bar shows the split. You can see the human versus AI share of the document as you work.
It fits how you already write. There is a Markdown editor with live preview and a linter, and a rich-text editor, plus a choice between a five-colour spectrum and solid blue for AI text.
It is not an AI detector and it does not generate text. The record is only as honest as the person writing: normal paste counts as your own text, so the system relies on you marking AI text when it goes in. If you want to write in your own voice and be open about the help you had, that is the right trade.
You can try the Assist Writer editor without an account; in demo mode, documents stay in your browser. A free account stores up to five documents in the cloud. If you want to export a document to PDF or Word with the author colours and human/AI stats kept, that is part of Pro. The quick guide to shortcuts, AI paste and editor modes covers the rest in a few minutes.
Write the argument yourself, let AI do narrow jobs, rewrite until you can defend every line, and keep a record. That combination is harder to fake than any style, which is exactly why it works.
Frequently asked questions
Can AI detectors tell if I edited an AI draft?
Sometimes, and not reliably. Detectors estimate authorship from writing patterns, so heavy rewriting can change their score, while light edits often don't. Turnitin said in 2023 that its sentence-level false positives cluster where human and AI text meet, so mixed documents are their weakest case. The safer approach is to rewrite for substance, disclose what AI did, and keep notes and drafts that show your process.
Is it cheating to use AI if I rewrite everything in my own words?
It depends on the rules you are working under, not on how thoroughly you rewrote. Some courses and employers allow AI for brainstorming or editing, others ban it for graded work, and many require disclosure either way. Check the policy for that specific assignment or client. If AI use is allowed, rewriting in your own voice plus an honest note about what the tool did is usually what is expected.
Do AI humanizer tools make writing authentic?
No. Humanizers and paraphrasing tools change surface wording while leaving the ideas, structure and examples exactly as the AI produced them, and they often garble meaning along the way. The work doesn't become more yours. Detection tools also keep improving, so a trick that works this term may fail next term, and deliberately disguised text is much harder to explain than AI use you disclosed.
How do I find my own writing voice if I'm not confident in it?
Start with writing you did before relying on AI: old essays, long emails, messages where you explained something you care about. Note how you open, how long your sentences run, which words you use, and how you disagree. That list is your voice. Then use AI for narrow tasks and edit its output against the list, keeping your examples, reasoning and plain words.
Why does my own writing get flagged as AI?
Detectors look for predictable, uniform text, and some human writing fits that pattern. A 2023 Stanford study found popular detectors consistently misclassified writing by non-native English speakers as AI-generated. Formulaic assignments, short texts and heavily templated academic prose can trigger the same mistake. Making your prose artificially fancy won't fix it; keeping drafts, notes and a record of any AI use gives you real evidence.
Does Assist Writer detect AI-generated text?
No. Assist Writer is not an AI detector and does not guess authorship from style. It records how text entered the document: text pasted with Ctrl+Shift+V is marked as AI and shown in colour, while typing and normal paste count as your own. As you edit AI text its colour shifts toward normal, and a stats bar shows the human and AI split.