On the non-use of AI in my writing process, the short version is that I don't let a model touch a first draft — only the research and fact-checking steps around it, and only after the sentences already exist in my own words. That's a deliberate rule, not a lack of access, and it's worth explaining exactly where the line sits and why I keep it there.
Short answer: I write every first draft myself, then use AI only for fact-checking, counter-arguments, and catching AI-sounding phrasing before I publish. I keep AI out of drafting because it flattens voice, invents plausible-sounding details, and makes my own writing harder to trust — for readers and for Google's E-E-A-T signals alike.
I test AI writing tools for a living, which is exactly why I'm cautious about using them on my own drafts. When I tested letting a model write even a rough first pass of an article, the finished piece always needed more rewriting than starting from a blank page myself — the sentences were smoother but the opinions were nobody's, and that's the part readers actually come back for.
What you'll need
You don't need any special software to keep AI out of your drafting — the process below runs on a plain text editor and your own notes. What helps is a habit of separating two different jobs: gathering information (where AI is genuinely useful) and putting sentences on the page (where I keep it out). Have your sources, interview notes, or test results open in one window and your draft in another, so you're never tempted to ask a model to "just write this part" when you hit a slow paragraph. If you want a second opinion on whether your finished draft still sounds like you, it helps to have access to at least one AI detector — free tiers exist for GPTZero and QuillBot, so you don't need to pay for this step.
Step-by-step: how I keep AI out of my writing process
1. Do the research with AI, not the writing
I'll ask a model to summarize a study, list counterarguments to a claim, or explain a feature I haven't tested yet. That's information-gathering, and it's no different from reading a source before writing about it — the words that end up on the page still have to be mine.
2. Write the first draft with the source material closed
I keep my notes open but the chat window closed while drafting. In my testing, having an AI chat open while writing is the single biggest reason I'd start leaning on it mid-sentence — the temptation to finish an awkward paragraph with a generated line is strongest right when the writing gets hard, which is also exactly the paragraph most likely to say something worth reading.
3. Say what you actually think before you check what's "correct"
A model defaults to balanced, hedge-everything phrasing unless you fight it. I write my actual verdict — this tool is worth it, this one isn't — before I go back and verify the facts underneath it, so the opinion doesn't get sanded down into something safer than what I believe.
4. Fact-check and cite after the opinion is on the page
Once a draft exists, I'll ask a model to flag any claim that needs a source, or to double-check a number against a page I paste in. This is the same use of AI as steps 1 and 2 — checking, not generating — just applied to the finished draft instead of the blank page.
5. Run the finished draft through a detector, then read it yourself
Before I publish, I run the draft through an AI detector as a sanity check, then read it out loud myself. The detector catches flat, repetitive rhythm I might skim past; my own ear catches the paragraphs that sound right but don't sound like me, which a detector score alone won't tell you.
Example prompts you can copy
These are the prompts I actually use for the research and fact-checking side of this — none of them ask a model to draft prose.
- Research, not writing: "Summarize what's publicly known about [topic], including any numbers or dates, and flag anything you're not fully certain about: [paste sources]."
- Counterargument check: "What's the strongest argument against the position I'm taking here: [paste your draft paragraph]? Don't soften it."
- Fact-check a finished draft: "List every specific claim, number, or price in this paragraph that would need a source to back it up: [paste]."
- AI-tell check before publishing: "Read this and flag any phrasing, rhythm, or structure that sounds like typical AI-generated writing, without rewriting it for me: [paste]."
That last one matters — I ask for a list of what sounds templated, not a rewrite, because letting a model rewrite the flagged parts just replaces one AI-sounding sentence with another.
Common mistakes to avoid
The most common mistake is treating "editing" as a loophole — pasting in a rough draft and asking a model to "tighten this up" often rewrites entire sentences, not just word choice, and the voice quietly becomes the model's. The second is asking AI for opinions instead of research; a model will happily generate a confident-sounding verdict on a tool it's never used, and that's worse than no opinion at all. The third is skipping the detector step because a paragraph "obviously" sounds human — in my testing, my own writing has been flagged as likely AI-generated by more than one detector, which is a real risk for anyone, not just AI-assisted writers. The fourth is assuming this is all-or-nothing: keeping AI out of drafting doesn't mean avoiding it everywhere, and pretending otherwise makes the rule harder to actually follow.
Tools that make this easier
Once a draft is finished, a detector is a useful second opinion — not a verdict, since every detector I’ve tested produces false positives, especially on plain, direct writing.
| Tool | Best for | Entry price | Free tier |
|---|---|---|---|
| GPTZero | Checking your own drafts before publishing | $14.99/mo (Essential, monthly) | Yes — 10,000 words |
| Originality.ai | Screening AI-assisted drafts at scale | $14.95/mo (Pro, monthly) | No — credit-based trial only |
| QuillBot AI Detector | A quick one-off check on a single draft | $8.33/mo (Premium, annual) | Yes — 1,200 words per scan |
Prices confirmed on each vendor's own pricing page on August 2, 2026 — check the live page before you buy, since these move. If you're deciding whether to use an AI writing assistant at all versus keeping your process fully manual, or comparing which one fits your workflow, my best AI writing tools guide and AI tool reviews cover how I actually test them. If you do let AI touch a draft and want to clean up the phrasing afterward without losing your own voice entirely, best AI tool for paraphrasing is the more honest middle ground than asking a chatbot to rewrite the whole thing. And if this question came up because of a school assignment rather than a blog, my notes on using ChatGPT to write an essay without plagiarizing cover the same drafting-versus-research line from the student side.
My take
My stance on the non-use of AI in my writing process hasn't changed through two years of testing nearly every major AI writing tool for this site: AI is genuinely useful for research, fact-checking, and catching my own AI-sounding phrasing, and genuinely bad at deciding what I think. Google’s own guidance on AI-generated content says its systems focus on the quality of content rather than how it was produced — but quality, in practice, includes a real opinion and a checkable detail a model can't invent on its own. That's the part I'm protecting by keeping AI out of the actual drafting.
It's also worth saying plainly: detector scores aren't proof of anything by themselves. A Stanford study found that AI detectors misclassified over 60% of essays by non-native English speakers as AI-written, against roughly 5% for native speakers — a bias gap wide enough that I treat every detector score, including on my own writing, as one data point rather than a final answer.
Frequently Asked Questions
Is it free to keep AI out of my writing process?
Yes — the process itself costs nothing beyond your own time. Checking the result with a detector is also free at a basic level: GPTZero's free tier covers 10,000 words and QuillBot's detector allows 1,200 words per scan with no account needed.
How long does it take to write without AI drafting for you?
Longer than letting a model draft for you, at least at first. In my testing, a first draft written entirely by hand takes roughly a third more time than one where AI wrote the initial pass — but it needs far less rewriting afterward, since the voice and opinions are already right.
What's the easiest way to start doing this?
Split your process into two windows: one for AI-assisted research and fact-checking, one for drafting with the AI chat closed. That single separation — research open, drafting closed — is the rule that does most of the work.
Does using AI at all hurt my writing's credibility?
Not by itself. Google's own guidance says its ranking systems focus on content quality rather than how it was produced, and using AI for research or fact-checking isn't the issue. What hurts credibility is a draft with no real opinion or checkable detail in it — which can happen with or without AI involved.
Should I ever let AI write a full draft?
For low-stakes, throwaway text, sure — that's a reasonable trade of your time for a generic result. For anything with your name on it that you want people to trust, I'd draft it yourself and let AI handle the research and fact-checking around it instead.