I Needed to Check if Our Freelancer's Content Was AI Generated — Here's What Actually Worked

It started with a Slack message from our SEO manager on a Tuesday morning: "Hey — have you noticed blog traffic is down 23% since March? I ran some of the recent posts through a detector and... we might have a problem."

We'd been working with the same freelance content writer for six months. Reliable. Always on time. Great rates. But now I was staring at a screenshot showing several of our posts flagged at 89% AI-generated. We'd already read about how Google treats AI-generated content and the traffic implications — so this wasn't just a contractor issue. It was potentially an SEO crisis with 40 published posts in question.

If you're in a similar situation — content manager, professor reviewing submissions, editor who received a suspicious draft, or just someone who got a document and genuinely needs to know if a human wrote it — here's what I actually learned from going through this process properly.

How Do You Check if Something Is AI Generated?

To check if something is AI generated, run the text through at least two independent AI detectors and compare results at the paragraph level, not just as a whole document. AI content tends to score high on statistical predictability — detectors measure how "surprising" the word choices are across the piece. No single tool is definitive, so consistent results across multiple tools carry far more weight than any one score.

That's the short answer. The longer one matters if you're about to do something with the result — like confront a contractor, flag a student, or pull content from your site.

Why One Tool Is Never Enough

The first thing most people do is paste text into a free detector, see a percentage, and treat it as a verdict. That's a mistake.

AI detectors work by analyzing statistical patterns — word predictability, sentence entropy, something called perplexity. The problem is those signals aren't clean. Human writers who happen to write formally, or who use structured templates, can trigger high AI scores. And AI content that's been lightly edited can slip through with low scores. This is why AI detection false positives are a genuine issue — one tool's "definitely AI" is another tool's "probably human."

That inconsistency isn't a flaw in one product. It's the fundamental state of this technology right now.

The Multi-Tool Process That Actually Worked

After going through 40 posts, here's the approach I'd recommend:

• Use at least two independent detectors. If three separate tools all flag the same paragraphs, you have a meaningful signal. If they disagree, you don't have a conclusion yet — you have a question worth digging into.
• Check sections, not just full documents. AI-generated content often appears in chunks: a robotic intro, a more natural middle, a stilted conclusion. Breaking text into paragraphs and running each one separately reveals patterns that a whole-document score will average out and hide.
• Compare writing style across multiple samples from the same source. Human writers have consistent quirks — sentence rhythms, word preferences, ways they handle transitions. If your freelancer's first submission reads completely differently from their sixth, something changed. That's worth noting.
• Check revision history where available. Google Docs and Microsoft Word both log edits. If a 1,200-word article shows a single paste event and zero revisions, that's not proof of anything — but it's a data point.

The Tool That Gave Me Usable Results

I tested six different detectors across the content backlog. Most gave me wildly different scores for the same article — sometimes a 30-point swing on identical text. The one that consistently gave me the most explainable output was WriteMask's free AI detector.

What made it useful wasn't just the score — it was that it highlighted specific sections that triggered the AI flags rather than handing me a percentage and leaving me to guess. That granularity is what you need when you're making a real decision. "This document is 74% AI" tells you almost nothing actionable. "These three paragraphs in particular read as highly predictable" tells you exactly where to look and what to ask about.

Understanding What the Score Actually Means

AI-generated text follows statistically predictable patterns by design. A language model picks words based on what's most probable in context — that's the mechanism. Human writers make "surprising" choices, use unusual rhythms, go on tangents, contradict themselves slightly. Detectors measure how surprising or unsurprising the word choices are across a piece. For a deeper breakdown of the technical side, this explainer on how AI detectors work covers perplexity and burstiness scoring in plain terms.

What to Do With the Results

Here's the part most guides skip: a detector score is not proof. It is a probability signal.

If you're a manager confronting a contractor, or a professor confronting a student, a high AI score tells you something worth investigating — it doesn't tell you the person is guilty. Most institutional AI policies and most contract law frameworks care about demonstrable evidence, not algorithmic probability. Knowing the difference matters if the conversation is going to go anywhere constructive.

If you're checking your own writing because you used AI assistance and want to know how it reads before submitting: that's a completely different situation. Tools like WriteMask are built for exactly that — detecting what flags, then helping you revise it so the final piece reads as genuinely yours. WriteMask passes content through AI detectors at a 93% rate after humanization, which is why it's worth running both the detection and the revision in the same place.

The Bottom Line

To check if something is AI generated: use multiple detectors, analyze at the paragraph level, and treat results as signals to investigate — not conclusions to act on. One tool, one score, one decision is the approach that creates problems, whether you end up accusing someone wrongly or missing something that actually matters.

Start with the free AI detector and look at what it flags at the section level. That's usually enough to know whether you need to dig deeper — or whether you can put the whole question to rest.