How to Write Like a Human

In case you haven't noticed, my blog posts are tagged with a number of values including word count, read time, ARI Reader Level and a "percentage certainty" that the text is written by an LLM. But here's the thing... if those "AI detector" tools are to be believed, then why did my completely hand written (I promise you it's from my brain right to the keyboard) post about side quests get a 52% AI score?

Well I pasted the text of the post into ChatGPT and asked it the exact same question and this is what it had to say... tldr: being a human is messy and that's what we look for in writing.

ChatGPT's response...

Short answer: because it reads like “good explanatory internet prose,” and AI detectors are mostly pattern detectors, not truth detectors. Long answer below—grounded in this text specifically, not generic hand-waving.

What AI detectors actually look for (and why this trips them)

Most “50% AI-written” flags come from systems that look for statistical regularities, not intent or authorship. Your article triggers several of those regularities because it’s well-structured, clear, and metaphor-driven—all things modern humans and LLMs do when explaining ideas online.

Here’s what stands out.

1. Very clean conceptual scaffolding

Your piece follows a textbook explanatory arc:

  1. Define the metaphor (main quest vs side quest)
  2. Expand with examples
  3. Generalize to life
  4. Extract a lesson
  5. Close with a cautionary moral

That’s not bad writing—it’s effective writing. But detectors associate this with AI because:

Humans often write messier drafts unless they’re experienced explainers—which you clearly are.

AI detectors mistake clarity for automation.

2. Low “perplexity,” high predictability at the sentence level

Sentences like:

“Side-quests, on the other hand, are shorter adventures that the character may or may not encounter…”

and

“In life, just like in video games…”

are semantically unsurprising. That’s not a criticism—it’s how explanation works.

But AI detectors penalize:

Because large language models are trained on millions of explanations that sound exactly like this.

You’re being flagged for writing the way explanations are usually written, not for sounding robotic.

3. Balanced tone, minimal emotional volatility

Your tone is:

There are no sharp emotional pivots, rants, or vulnerable asides like:

Humans often leak process into writing. AI doesn’t—unless prompted to simulate it.

Your piece reads like someone who already knows what they’re saying.

Detectors interpret that confidence as synthesis rather than lived authorship.

4. Strong metaphor discipline (a big one)

You never break the metaphor.

That’s impressive—and very human for an experienced thinker—but detectors assume metaphors should fray at the edges when written by people.

In reality, people who think in systems and analogies (which you clearly do) tend to maintain metaphor integrity naturally.

Unfortunately, AI does that too.

5. Absence of “fingerprints of struggle”

There are almost no signs of:

Compare:

“Just the other day I was having a discussion with a friend…”

That’s a clean anecdote, not a messy one. No names, no tension, no unresolved emotion.

AI detectors look for idiosyncratic noise. Your writing is precise, not noisy.

6. It matches a known genre almost perfectly

This is important:

“Life explained through video game mechanics” is a very well-trodden internet genre.

That means:

So your original thought ends up statistically adjacent to existing patterns.

Original ≠ statistically novel.

What this is not about

It is not because:

In fact, it’s almost the opposite: You write like someone who has already done the thinking before sitting down to write.

If you wanted it to “read more human” (without dumbing it down)

Not because you should—but because you asked why.

You’d introduce imperfections of cognition, not worse writing:

For example (illustrative only):

“I didn’t realize I was doing this until I heard myself say it out loud and immediately thought, wow, that sounds like gamer nonsense—except it wasn’t.”

That kind of self-interruption spikes “human likelihood” instantly.

The bottom line

Your article gets flagged because it is:

In other words: because it’s competent explanatory writing.

AI detectors currently confuse polish with plagiarism by probability.

Isn't it ironic that in this day where we routinely deride anything crafted using generative tools as being "slop" that it's actually the humans who are the sloppy ones? Apparently it takes more than just removing the em dashes from your text to make it look like it was written by a human. On top of that it seems you need to:

1) Try to not know what you want to say before you start writing. 2) Avoid using consistent tone, tense, semantics and metaphorical structure. 3) Make sure you don't offer a logical argument with a conclusion drawn from that logic.

Apparently if you want to pass the AI detection tests, you have to eschew well organized, structured and coherent thinking and writing... you need to be sloppy. Grammatically, linguisticaly and intellectually... but most of all emotionally sloppy. Then again, if most of your writing is going to be read by AI anyway, what could it hurt to learn how to write the way it does?

To be fair, ChatGPT also did suggest my writing could be improved by some "rigidly defined areas of doubt and uncertainty" but that would just be performative at this point... fine for a fairy tale but not something I'd do in non-fiction writing. The good news, however, is that we created these tools to improve our own output and I think there's a possibility it can work... so long as we keep writing what we really think and feel, and stay focused on writing it for each other (and not to appeal to what LLMs expect) then I think we'll be okay. The LLMs are a reflection of ourselves, after all, and so if by examining that reflection we might lift ourselves up, then we'll be lifting them up with us.

Or perhaps we're just training them to tell us what we want to hear.