Tired of being flagged by AI? I started ‘watermarking’ my writing to prove I’m human

AI detection tools are getting more popular and more aggressive. I've tested and reviewed many of them, but the problem is, they’re not always right.

If you’re a real human writer (hi!) who’s ever had a draft flagged as “AI-generated” just because you write cleanly, quickly, or consistently, you already know how frustrating this can be, especially if you regularly use tools such as Grammarly, aka Superhuman. It’s not just annoying — it can mess with your credibility.

So I started doing something new: I “watermark” my writing.

Not with an obvious signature or a cheesy hidden message, but with subtle, repeatable, human-only markers that make it easier to prove a draft came from me. If you are a writer or write regularly for school or work, you might want to consider trying it, too.

Here’s the workflow I use now to protect my reputation in the age of suspicion.

What does it mean to 'watermark' your writing?

With so much AI content flooding the internet, many creatives are left scratching their heads, wondering how to protect their work. While I can't promise someone isn't going to steal your writing prompts, what I can do is help you embed small signals into your draft that:

• are easy for you to explain later
• are hard for AI to replicate consistently
• show evidence of your real process (not just the final output)

You can think of it as proof-of-work for writers. You’re not trying to “beat” AI detectors. You’re building receipts that prove you're a human being (what a world!).

Step 1: Build a personal “human marker” list (and reuse it)

The easiest watermark is one you naturally use over and over.

Pick 3–5 signature elements that feel like you, and start weaving them into your work consistently. Some of my favorite authors, like Nora Ephron and Stephen Kin,g already do this.

Examples of human markers that work well:

• specific slang you actually use
• a recurring sentence structure
• a personal habit in your formatting
• a slightly chaotic but readable rhythm

You can think of it as your distinct voice. When you speak to a friend on the phone, they know it's you because of your unique articulations, speaking rhythms, and mannerisms. It's the same thing here. You're aiming for a recognizable footprint in your work.

It's a way to say, “I write like this on purpose. Here’s how it shows up across my work.”

Step 2: Add one intentional “structural quirk” per draft

If you ask me, this is where watermarking gets fun. AI tends to write in a predictable pattern: intro, list, conclusion, smooth transitions, no surprises. But we humans, we zigzag even when we don't mean to.

So now I add one deliberate structural choice that’s easy to defend. You can try it with:

• easy structural quirks that look natural
• a micro-story in the middle
• a tiny detour section:
• add a quick side note before moving on to a new topic

This creates texture — the kind that AI cannot imitate, and consistently helps you stand out.

Step 3: Plant a “human-only detail” AI wouldn’t guess correctly

This one matters because it’s the hardest thing for a bot to fake. As humans, creativity is where we shine. Again, so wild that "being human" is something we have to try to highlight more of, but in 2026, this is the AI ear.

Add one detail that’s true, specific, and low-stakes, like:

• what you were doing when you wrote it
• what you almost titled the piece
• the exact moment you thought something new

This isn’t “proof” on its own, but it makes your writing stand out as human and not manufactured.

Step 4: Keep timestamp receipts (your future self will thank you)

Although AI doesn't train on your writing unless you enable it, AI is still very new and we don't know what types of leaks, hacks or prompt injections are out there. And, as someone who would seriously lose it if an LLM started spitting out my own work, here's what I do to keep a process trail.

What to keep (minimal effort, maximum proof):

• a screenshot of your first outline
• a version history link (Google Docs is perfect for this)
• a quick note showing what changed and why
• a draft exported as PDF with a timestamp
• Bonus tip: If you write in Google Docs, turn on version naming, such as "Draft 1- messy" and "Draft 2- tighter intro."

Of course, you don't need to do this for every email or piece of written work. But you might want to consider doing this for bigger creative works, such as essays or novels you are working on.

Step 5: Write “too human” for AI detectors

I once had an English professor tell me that we limit what we can do if we don't have the words to describe it. I often think about that when I read anything written by AI. You may have noticed that anything written by ChatGPT, Gemini or even Claude is still pretty basic.

When you elaborate or use words that a chatbot just wouldn't use, you immediately stand out as a human and a writer.

AI detectors tend to flag writing that is:

• too polished
• too symmetrical
• too evenly paced
• too neutral

That tone shift alone can make your draft feel undeniably authored.

Final thoughts

Here’s the uncomfortable truth: AI detection tools can be wildly inconsistent. Two detectors can look at the same paragraph and disagree completely. I realize that's probably not helpful if you're a student who just turned in a term paper or professional cover letter, but the act of essentially 'watermarking' your work makes it easy to show how you wrote it, when you wrote it, and why certain choices (including the words) were made.

Because honestly, that’s something AI can’t retroactively generate convincingly. And you’re not proving you didn’t use AI — you’re proving you’re a writer. I wish writers didn’t have to do this. But we’re in a weird moment where being clear, fast and competent can make your work look suspicious.

So instead of stressing every time a detector throws out a false accusation, I started treating my writing like a craft and a paper trail. Here's a simple rule to follow: if you're going to be judged by machines, leave human fingerprints everywhere.

Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.

More from Tom's Guide

• I use the 'unicorn prompt' with every chatbot — it instantly fixes the worst AI problem
• ChatGPT Is quietly replacing Google's most important page, study finds
• I didn’t realize ChatGPT could do this — 10 features hiding in plain sight