The gatekeeper returns somewhere else

Perhaps this is where the argument becomes unfair to Substack.

A platform cannot inspect intention.

It cannot know whether I spent three weeks thinking about an article or twenty seconds asking a model to create one. It cannot look at polished English and discover whether the ideas underneath it began in my head, emerged through an argument with a machine, or arrived fully formed from a prompt I barely understood.

And I am not sure I want it trying.

The moment a platform begins deciding who has thought enough to deserve publication, something important has already been lost.

That was part of the attraction of the open internet in the first place.

You did not need permission.

You did not need the right university.

You did not need an editor to decide that your subject mattered.

You did not need to speak perfect English.

You did not need to live in London, New York or Paris.

You did not need to belong to the institution that historically decided who was allowed into print.

You could simply write.

For someone with knowledge but without access, that is not a small freedom.

Artificial intelligence can widen it further.

Someone who thinks in Arabic can publish in English.

Someone who has spent forty years repairing engines but never learned how to structure an essay can explain what those forty years taught him.

A doctor can make technical knowledge readable.

A researcher can translate work for people outside the field.

A person whose physical condition makes conventional writing difficult can speak into a machine and receive prose back.

A young writer can ask questions of an argument that no editor has time to ask.

It would be a strange defence of human writing if preserving it required rebuilding all the old barriers around who gets to participate.

So I do not want Substack to decide who is a real writer.

I do not want an AI detector standing at the entrance.

I do not want the platform examining every paragraph for evidence of forbidden assistance.

And I certainly do not want good English to become suspicious simply because a machine may have helped somebody reach it.

The freedom to publish should remain exactly that.

Freedom to publish.

But perhaps we have been confusing publication with something else.

Amplification.

Those are not the same right.

Anyone can stand in a public square and speak.

That does not create an obligation for everybody else in the square to gather around.

Anyone can print a book.

That does not guarantee placement in the front window of a bookstore.

Anyone can publish an essay.

That does not necessarily mean the platform must recommend it, circulate it, place it in feeds or treat its existence as sufficient evidence of value.

This distinction may become central.

The old gatekeeper controlled the printing press.

The new one may control attention.

That sounds dangerous, because it is.

Once a platform decides what deserves visibility, it inevitably makes judgments. Even if those judgments are encoded as recommendations, probabilities and engagement models rather than editorial meetings, they shape the culture that forms above them.

There is no neutral ranking system.

If the platform rewards frequency, writers will become frequent.

If it rewards reaction, writers will become provocative.

If it rewards time spent, writing will adapt to hold attention.

If it rewards subscriptions, writers will learn what converts.

If it rewards existing popularity, established writers become harder to displace.

Every metric eventually teaches the participants what kind of person to become.

AI simply makes adaptation faster.

A writer can now test ten titles.

Generate twenty Notes.

Produce three versions of the opening.

Respond to every trend.

Maintain constant presence.

The platform does not need to instruct anyone to do this. The incentive is enough.

Which means the question for Substack may not be:

How do we identify AI writing?

It may be:

What behaviour do we want our system to reward now that writing itself can be produced almost without limit?

That is a much harder question.

Perhaps posting more should not automatically mean appearing more.

Perhaps a writer publishing once a month and retaining readers for years tells us something different from a writer publishing three times a day and constantly replacing yesterday’s audience with today’s.

Perhaps the archive matters.

Perhaps correction matters.

Perhaps a reader returning voluntarily to the same person for two years matters more than whether one post briefly travelled through the feed.

Perhaps sustained subscription is a stronger signal than momentary engagement.

Perhaps thoughtful replies matter differently from reactions.

Perhaps the platform should become better at recognizing relationships rather than merely activity.

None of this requires knowing whether AI wrote paragraph seven.

It requires asking what sort of ecosystem makes disposable intellectual production less advantageous.

There is another possibility.

Make the process more visible.

Not compulsory confession.

Not a red label saying AI GENERATED, as though every use of a machine belonged to the same category.

Something more useful.

A writer might choose to say:

Original argument and examples: mine.

Research and counterargument: developed with AI assistance.

English editing: AI-assisted.

Sources checked by me.

Final responsibility: mine.

Another might say:

Translated from my original French draft with AI assistance.

Another:

This piece began as an experiment: the initial draft was generated, then substantially rewritten.

Another might say nothing.

The declarations would not prove honesty.

No disclosure system can.

But they could begin creating a vocabulary more sophisticated than human and AI.

Because the important question is not whether a machine was present.

It is what role the machine played.

We already understand this distinction elsewhere.

A photographer can use autofocus without surrendering authorship of the photograph.

A filmmaker can use digital color correction.

A musician can use software to remove noise.

A novelist can work with an editor.

A translated book remains connected to its author even though another person supplied every sentence in the reader’s language.

Tools do not erase authorship merely by entering the process.

The problem begins when the tool no longer assists the work but supplies the relationship to the subject that the work appears to represent.

And that is precisely the part no disclosure box can verify.

So perhaps responsibility matters more than purity.

You put your name above the article.

What does that name promise?

Not that your fingers pressed every key.

Perhaps not even that you invented every argument without assistance.

It should at least promise:

I know what this says.

I understand why it says it.

I accept responsibility for publishing it.

If you challenge it, I am still here.

That last sentence may become increasingly important.

The article ends.

The author should not.

This is where comments, conversations and continuity begin to matter in a way they did not when the artifact itself carried more evidence of the labor behind it.

A reader asks:

You argue here that X follows from Y, but six months ago you argued something very close to the opposite. What changed?

That is not an AI-detection test.

It is something better.

It asks for history.

A real answer might be:

I was wrong.

Or:

I still believe the earlier argument, but I was using the word differently.

Or:

This book changed my view.

Or simply:

I had not noticed the contradiction.

Those answers tell us something about the mind behind the archive.

Not that it is brilliant.

Not that it is human in some technologically pure sense.

That it is present.

This may be what the future reader learns to look for.

Not whether the prose contains machine fingerprints.

Whether the writer has a trajectory.

Does one essay remember another?

Do questions recur?

Do positions change for reasons?

Does experience accumulate?

Does the writer sometimes remain silent?

Does something written today belong to the same intellectual life as something written two years ago?

A generated article can appear instantly.

An intellectual history cannot.

Or at least not in the same way.

A machine can fabricate an archive, certainly. It can generate a hundred old-looking essays and construct consistency among them. It can manufacture replies. It can simulate evolution.

There is no final technological fortress here.

Anything visible can eventually be imitated.

But trust has never depended on mathematical certainty.

We trust people because we encounter them repeatedly under different conditions and the relationship survives.

The doctor does not prove competence with every sentence.

The journalist does not begin from zero with every article.

The writer does not earn our attention afresh from an anonymous blank page every morning.

Reputation is accumulated evidence.

And perhaps that is where paid writing still has something AI cannot cheaply replace.

I can ask a model for an essay on almost anything.

Soon the result may be excellent.

What I cannot ask for in the same way is:

What does this particular person, whose work I have followed for years, now think after everything they have already thought?

That question contains history.

It contains expectation.

It contains disagreement.

It contains memory.

It contains the possibility of surprise precisely because I already know something about the person being surprising.

That may become the real object of subscription.

Not words.

Not even information.

Continuity of mind.

If that is true, then Substack’s defence against abundance may not be to become better at detecting machines.

It may be to become better at helping readers recognize people worth returning to.

That would preserve the original freedom.

Anyone could still publish.

AI could still be used deeply, lightly or not at all.

A writer would not have to prove linguistic purity.

A non-native speaker would not be penalized for sounding too polished.

A person with genuine knowledge would not need permission from an editor to enter.

But the platform would acknowledge something the age of infinite prose makes unavoidable:

publication is no longer scarce enough to be the valuable act.

Attention is.

Trust is.

Return is.

And perhaps most of all, time is.

The strange thing is that this brings us back to an old form of reading.

Before feeds, before recommendations, before notifications, people often returned to writers because they wanted to know what that writer would make of something.

Not because there were no other words available.

Because the words belonged to a mind they had decided was worth following.

AI may make that distinction more valuable, not less.

But only if platforms resist the temptation to treat every new piece of content as the thing they are actually selling.

The article may become cheap.

The relationship cannot.

And that is where the reader begins to matter more than the detector.