Do AI detectors identify authorship—or punish humans for sounding like the machines trained to imitate us?
I have been writing poetry since before generative AI existed.
Prose was my first love language. My first introduction to writing. I grew up with Irish and Black poets at my disposal, books all over the house, and a mother who would spend hours with me at the library. Later, she introduced me to anthology contests. I submitted poems. Sometimes I won.
I have been writing for decades.
Which is why I have a question.
Is Pangram detecting authorship… or is it penalizing the ways human writers mature, revise, and become more polished?
Because apparently becoming a better, more concise writer now makes my humanity suspicious.
Let me show you what I mean.
In 2009, I wrote this:
Delicious Ambiguity (unfinished)
June 13, 2009
Delicious ambiguity is what I taste in your kiss, I long for it, I crave the way it slides across my mouth filling me with the warmth that only your lips can bring. You see it is this very equivocation I seek to bring out of you, and oh, how I love the vacillation, the near battle that takes place between each kiss.
There is a touch I was designed for and it rests obscurely in the simple passing of your thick strong hands along my cheek. I beg for a simple graze of this esoteric hand to lead me to an even deeper layer of uncertainty. As always you grant my request.
I must be a masochist, some would say to desire this sort of exsistence, however I must admit that I do long to hear the exacting sound of ambivelence in your voice when faced to describe us. I find comfort in the knowledge of the quandry you are in, torn between what you feel and what you know, your indecisive touch lingers far longer then those that were sure.
I love the feel of your speculative tongue as you seem to search for the answers in the darkest corners of my body and still you come out with nothing more than conjecture, hmmmm how I revel in your postulation.
(c) 2009 — Desiree B Lawrence
Pangram says that poem is 100% human.
Fair enough.
There are misspellings.
There are sentences I would absolutely break apart today.
There is a young writer in there reaching for every beautiful word she owns and apparently determined to use all of them before the poem ends.
I can see her.
I love her.
And I can also see the seventeen years of writing that happened after her.
Because in July, I wrote this:
When Fall Comes
July 28, 2026
When fall comes,
I miss you the most.
I sit by my window,
the cicadas singing their final song,
their chorus reminding the world
that another season is ending.
Outside, the vehicle that used to be yours is parked, but I can still remember the excitement when I heard you pulling into the drive.
I can still see the backyard
that once held our laughter—
the dancing,
the barbecues,
the fires that burned long after the sun had gone down.
Memory has a way of making places
feel inhabited long after everyone has left.
It is this time of year
that I miss you the most.
Not because I question the choice.
I wouldn’t change my mind.
Some decisions save us,
even as they break our hearts.
The choice had to be made.
But there is grief in the choices we make,
and sometimes grief does not ask to be healed.
Sometimes it simply asks
to be witnessed.
I am no longer consumed by it.
Most days it lives quietly,
tucked between ordinary moments.
But every so often,
it comes to sit beside me—
as faithful as the song of the cicadas,
reminding me that seasons change,
that love changes,
that people change.
Some grief is not meant to be overcome.
It is meant to be carried.
It sheds its leaves,
learns the language of quiet,
and waits patiently
for me to remember
that some losses
do not disappear.
They simply learn
how to become autumn.
Pangram says that poem is 100% AI.
I wrote both.
So what happened?
Did I become a robot sometime between 2009 and 2026?
Or did I become a better writer?
Because the second poem is more restrained.
I understand now that I do not have to explain every metaphor after I make it.
I know when to let a line breathe.
I know that sometimes the strongest sentence is the shortest one.
I know how to return to an image without beating the reader over the head with it.
I know how to edit myself.
I have spent seventeen more years learning how to write.
And apparently some of the evidence of that growth now looks suspicious.
Which brings me to a question I heard online that has been sitting with me ever since:
Do I sound like AI, or does AI sound like me?
Question posed by Uppity Negress
Because I think we have the direction of influence completely fu$%ed up.
Artificial intelligence did not invent human language.
It did not invent the essay.
It did not invent the em dash.
It did not invent parallel construction, lyrical repetition, metaphor, clean transitions, polished academic prose, conversational cadence or the sentence that lands exactly where you want it to land.
AI learned those things from us
From human writing.
From human language.
From human patterns.
From the things people have been saying, writing, publishing, arguing, documenting, teaching and creating for generations.
And now we have built software that looks at human writing and says:
You resemble the thing that learned how to write by studying humans.
Well…
yes.
That is evidence of resemblance.
It is not necessarily evidence of authorship.
And that distinction matters.
A detector does not sit beside me while I write.
It does not know what I have been thinking about for three weeks.
It did not watch me jot down a sentence in Notes while waiting in the carpool line.
It did not hear the first version I spoke out loud.
It did not watch me delete six paragraphs because I realized I was making the same point twice.
It did not see what I wrote in 2009 and compare it with what I write now.
It does not know my literary history.
It does not know my process.
It sees the artifact.
The finished language.
And from that, it makes an inference about how that language came into existence.
That becomes especially complicated when we start talking about disability and accessibility.
Because I dictate.
I have dictated for years.
When I was in college, I used Dragon.
I wasn’t introduced to dictation because ChatGPT arrived and suddenly I wanted a shortcut around writing.
I use dictation because sometimes speaking is the way I write.
I am dyslexic as hell.
I have never been an exceptional typist.
My mother could type something like 225 words a minute. Growing up, I remember hearing those keys CLICKING.
That woman could TYPE.
Me?
I think I topped out around 75 words a minute despite doing every typing test known to humanity.
And eventually I was like, okay.
Apparently we will not be pursuing a career in competitive typing.
Also, I am not cutting my nails.
Hear that clicking? (imagine it lol)
Not happening.
But seriously: there has always been a gap between how quickly my brain produces language and how efficiently my fingers can put it on a page.
Dictation closes some of that gap.
Spellcheck closes some of that gap.
Text-to-speech closes some of that gap.
Accessibility technology closes some of that gap.
These are not shortcuts around authorship.
They are tools that allow authorship to happen.
And this week I decided to test it.
I turned on Apple Dictation. (Instead of Dragon so there would be no copy/paste)
I opened Substack and began speaking.
I spoke.
No ChatGPT.
No AI rewriting.
No “make this more concise.”
No “clean this up.”
No generative editing.
Just:
my brain → my mouth → speech-to-text.
Then I took those words and ran them through Pangram.
100% AI.
Now tell me what exactly was detected.
Because I was THERE. (so were you if you watched the video above)
I know where the words came from.
They came out of my fu$%ing mouth.
And that is where this stops being a cute little internet debate about whether people overuse em dashes and becomes something I think we need to take seriously.
Because what happens when proof of humanness becomes a performance?
Do I need to leave the misspellings?
Should I make the sentence clunkier?
Stop using parallel construction?
Maybe throw in a few awkward transitions?
Should I stop revising before I publish?
Should the dyslexic writer decline spellcheck so her mistakes can authenticate her?
Should the disabled writer stop dictating because clean transcription is suspicious?
Should accomplished writers deliberately make themselves worse so everyone knows they’re real?
This is pushing the anti-intellectualism that we are currently witnessing, not to mention media literacy.
And who gets harmed first when we create systems like that?
It usually isn’t the corporation.
It isn’t the military contractor.
It isn’t the billionaire building AI infrastructure.
It isn’t Apple adding artificial intelligence to the phone you’re holding while asking me if I used AI.
It is the regular hard-working person.
That is the part of this conversation that keeps bothering me.
Because let me be clear:
I understand the critique of AI.
There are serious questions about labor displacement.
There are serious environmental concerns.
There are serious questions about where enormous data infrastructure gets placed and which communities absorb the costs.
There are questions about intellectual property and whose work trained these systems.
There are questions about surveillance.
There are questions about the war machine.
There are questions about extraction.
There are questions about how yet another technological revolution inside of a colonial capitalist system will inevitably reproduce the logic of that system unless we actively interrupt it.
Those are real conversations.
I want to have them.
What I am not interested in is our favorite little American pastime where we discover a massive systemic problem and immediately decide the solution is to regulate each other.
Now everybody is the cop.
Did you use AI for that?
Did you use AI for this?
Is that picture AI?
Was that paragraph AI?
Explain your process.
Show me your drafts. (I have 150 by the way)
Prove you thought the thought.
Prove you wrote the sentence.
Prove to me that your labor is pure enough for me to consume.ESPECIALLY when it comes to the marginalized writer.
And I am watching this happen while people ask me these questions from devices that are themselves using artificial intelligence all day long. (you are not giving up your smartphone for a flip phone)
Your phone summarizes your messages.
Search engines use AI.
Your photo software uses AI.
Your accessibility features use machine learning.
The corporation is integrating it into your operating system while somehow the moral emergency becomes whether a disabled independent writer used a tool to organize a paragraph.
Make it make sense.
And yes, this has economic consequences.
I recently watched my paid subscriber count fall from roughly 467 to 428.
That is 39 people. (roughly $3,900–$4,680 in annualized revenue, depending on the subscription plans)
More than eight percent of that paid subscriber base.
I cannot prove why each person left.
I am not going to claim that every one of those cancellations happened because somebody scanned my writing.
But the timing has absolutely made me pay attention.
Because this is not abstract to me.
This is my income.
This is how I support the work I do, and my family.
This is what happens when the cultural conversation around AI shifts from:
How are institutions deploying this technology?
to:
Which individual human can we catch using it?
And now platforms are putting detection tools directly into the relationship between writer and reader.
Think about what that changes.
Before, you read something I wrote.
Maybe you loved it.
Maybe you hated it.
Maybe you thought I was brilliant.
Maybe you thought I needed an editor.
Maybe something sounded different than my last piece.
You assessed the work, and decided you wanted to follow along…
Now there is another possibility.
You can ask software to tell you whether I am telling the truth about creating it.
And suddenly there is a third party sitting between writer and reader whispering:
But are you sure she wrote this?
That is not a small cultural shift.
Especially when the software can look at a poem I wrote myself and confidently call it machine-generated.
Especially when it can look at words I literally dictated from my own mouth and classify them as AI.
Especially when another poem, written by a younger version of me who misspelled existence, ambivalence and quandary, passes the humanity test with flying colors.
There is something almost perversely funny about it.
Apparently the more visibly imperfect I was, the easier I was to recognize as human.
And I refuse that bargain.
I refuse the idea that I must preserve mistakes as evidence of authenticity.
I refuse the idea that disabled people should abandon accessibility tools so our process looks sufficiently laborious.
I refuse the idea that using technology automatically erases authorship.
And I refuse the moral flattening that treats every interaction with artificial intelligence as equivalent.
Because authorship is not keystrokes.
It never was.
My mother typing 225 words a minute was not more of an author than me because her fingers could keep up with her thoughts.
A writer using a pencil is not more authentic than a writer using a typewriter.
A writer using a typewriter is not more authentic than a writer using Microsoft Word.
Spellcheck did not eliminate authorship.
Voice dictation did not eliminate authorship.
Editing does not eliminate authorship.
Collaboration does not eliminate authorship.
And the questions around generative AI deserve much more intellectual rigor than:
machine touched it = human didn’t make it.
There are questions worth asking.
Who developed the idea?
Who made the argument?
Who is accountable for what the piece says?
What labor did the technology replace?
What labor did it enable?
What was generated?
What was edited?
What was transcribed?
Who is profiting?
Who is being exploited?
Who bears the environmental cost?
Whose work trained the technology?
Those questions require us to think.
“SCAN FOR AI” lets us pretend we don’t have to.
And maybe that is what bothers me most.
We are outsourcing discernment.
Again.
We are taking a deeply complicated conversation about art, labor, disability, capitalism, technology, ethics and authorship and handing it to an algorithm so it can give us a percentage.
Then treating the percentage like knowledge.
But a probability is not provenance.
Resemblance is not authorship.
And a machine saying I sound like a machine that learned how to sound like people like me is not proof that I did not write the fu$#ing thing.
I have almost two decades of my writing sitting online.
You can watch the voice change.
You can watch the syntax tighten.
You can watch the ideas deepen.
You can watch the young woman who wanted to use every word she knew become the middle-aged woman who finally understands that sometimes:
Some decisions save us,
even as they break our hearts.
is enough.
That growth belongs to me.
The messy writing belongs to me.
The polished writing belongs to me.
The dictated writing belongs to me.
The assisted writing belongs to me when the thinking, argument and authorship are mine.
And yes, when I use generative AI, I am capable of telling you that too.
But I will not allow a detector to become more authoritative about my own creative process than I am.
So I am going to keep writing.
I am going to keep dictating.
I am going to keep revising.
I am going to keep getting better.
And I am not going to deliberately make myself worse just so a machine recognizes me as human.
Which leaves me with only one question.
My poem from seventeen years ago was human.
The poem I wrote this year was AI.
The words I spoke directly from my own mouth were AI.
So tell me, Pangram:
At what point did I stop being human?
If this essay made you think, challenged something you have accepted as neutral, or gave language to something you have been trying to name, join me as a paid subscriber.
Paid subscriptions do more than unlock additional writing. They directly support the labor that makes this work possible—the research, writing, teaching, community building, and continued development of liberation-centered resources outside of institutional gatekeeping.
If you value independent thought, rigorous critique, embodied practice, and conversations that refuse easy answers, become a paid subscriber and help sustain the work.
Paid subscriptions are $10/month or $100/year.
Equity Partners are $150/year, helping create access for others while sustaining the larger Liberation Education ecosystem.
And if cost is the barrier, scholarships are available at Scholarships@DesireeBStephens.com.
No proof required.
No trauma story owed.
Just ask.
And YES, apparently this article too is AI
The only reason it is not 100% is that I decided to test what physically typing some of this would produce.
Practice Your Praxis
Liberation is not just about having the right critique.
It is about what we do once we can see the system clearly.
So before you leave this conversation thinking only about whether Pangram is accurate, I want you to bring the question back into the places where liberation actually has to live:
Self. Home. Work & Community.
SELF
Where have you started outsourcing your own discernment?
When a detector, algorithm, institution, headline, credentialed expert, or authority gives you a percentage or a label, do you treat it as information to consider—or as the final word?
Notice where certainty feels safer than curiosity.
Notice where you have learned to trust the system’s classification of a person more than the person themselves.
And notice where you may be performing your own humanity in order to remain legible to systems that were never built to hold the fullness of you.
Your practice:
Catch yourself the next time a score, label, or algorithmic judgment feels like proof. Ask: What do I actually know? What am I assuming? What information is missing?
You do not have to become less articulate, less polished, less disabled, less assisted, less complicated or less yourself in order to prove that you belong here.
HOME
How are we teaching the people closest to us to understand technology, authorship and truth?
Because media literacy cannot simply mean teaching our children how to spot “AI.”
It has to mean teaching them how to think.
Can they distinguish between evidence and inference?
Can they understand that a tool can be useful and still be wrong?
Can they hold the possibility that technology can create access for one person while creating harm somewhere else?
Can they ask better questions before accusing someone of dishonesty?
And perhaps most importantly:
Are we raising people who know that imperfection is not the price of being believed?
Your practice:
Have a conversation at home about something an algorithm has decided for you recently—a recommendation, a search result, a grade, an AI detection score, a social media feed—and ask together: Who built this? What is it measuring? What can it actually know?
That is media literacy.
That is critical thinking.
That is liberation education.
WORK & COMMUNITY
This is where the individual conversation becomes structural.
Where is your workplace, school, organization, institution or community beginning to use AI detection—or other algorithmic systems—to make judgments about people?
Who gets questioned?
Who gets believed?
Who gets asked to prove their work?
Who has access to tools that make their labor easier, and whose use of those same tools becomes suspicious?
What happens to disabled people who rely on dictation, spellcheck, transcription, text-to-speech or other assistive technology?
What happens to multilingual writers?
What happens to Black and brown writers whose language, cadence or intellectual traditions may not fit whatever a machine has decided “human writing” is supposed to look like?
And when harm happens, is your institution interrogating the system—or simply disciplining the individual?
Your practice:
Before adopting an AI detector or any automated decision-making tool in your workplace or community, demand a human process around it.
Ask:
What exactly does this tool measure?
What does it not know?
What happens when it is wrong?
Who gets to appeal its judgment?
How are disability and accessibility protected?
Are people being given due process before consequences are imposed?
Are we using technology to support human discernment—or replace it?
Because “the software said so” is not an ethical framework.
The Conclusion Is the Practice
There will always be another tool.
Another technology.
Another institution promising to make a complicated human question easier by giving us a score, category, checkbox or answer.
Our work is not to reject every tool.
Our work is to refuse to surrender our discernment to them.
Liberation requires us to tolerate complexity.
To ask better questions.
To hold institutions accountable without turning one another into suspects.
To recognize that technology can simultaneously create access and reproduce harm.
To understand that two things can be true at once—and still require us to decide what ethical action looks like.
That is praxis.
Theory tells us what the system is doing.
Praxis asks us what we are going to do about it.
And that is why I do not want you simply leaving my writing with another thing to agree with.
I want us practicing together.
Practice Your Praxis — Every 19th at 12 PM ET
Members and paid subscribers meet with me every month on the 19th at 12 PM ET for Practice Your Praxis, our live space for taking the theories, critiques and conversations we engage here and moving them through:
Self.
Home.
Work & Community.
We meet on the 19th intentionally, holding this practice in relationship with Juneteenth and the unfinished work of liberation.
Bring the questions.
Bring the contradictions.
Bring the place where what you believe and what you actually practice have not quite caught up with one another yet.
You do not need to arrive with the answer.
We come together to practice finding it.
In solidarity and liberation,
Desireé B Stephens, CPS-P
Somatic Educator | Counselor | Community Builder
Founder, Make Shi(f)t Happen
Writer of Liberation Education
Steward of Selenite & Sage Healing Homestead Co.
LIBERATION IS A LIFESTYLE.
Reading and remembering, yet again, that tools utilized within the power structures and systems we traverse will reflect the vested interest of those few power hoarders at the top. Thus we the standard set at mediocrity, undermining the success of diverse voices and the immediate invocation of policing each other to maintain this system.
The mind boggles at using AI to assess if something is AI; to then trust the AI assessment and determine the work as unworthy — because it is AI.