The English Teacher’s Dilemma: Is AI Changing English?

English teachers spend much of their professional lives explaining rules.

Yet the language they teach is the product of centuries of people changing those rules.

English once distinguished between thou and you in ways that have largely disappeared from everyday standard usage.

Words have acquired new meanings. Pronunciations have shifted. Grammatical constructions have emerged, spread and sometimes become accepted.

English has never been a finished product.

It is a constantly evolving set of conventions shared, negotiated and sometimes contested by the people who use it.

And those people do not all speak the same English.

British English differs from American English. Indian English has established features of its own. Regional dialects preserve constructions that standard written English may reject.

Meanwhile, millions of people use English internationally without necessarily communicating with native speakers at all.

Consider an Italian engineer discussing a technical problem with a Japanese supplier.

Their objective is not to sound British or American.

Their objective is to understand one another.

So why do we teach standard English?

Because shared conventions matter.

They make communication more predictable, particularly when people do not know one another.

They are important in education, professional writing, examinations and countless situations where readers and listeners have established expectations.

But there is a distinction worth preserving.

Standard English is a valuable convention. It is not the entirety of the English language.

And that distinction becomes particularly important when AI enters the classroom.

Here is where the story becomes more surprising.

We might reasonably expect AI to influence emails, essays and reports.

After all, people are actively asking it to rewrite those texts.

But what about the English people actually speak?

Research published in 2025, Model Misalignment and Language Change: Traces of AI-Associated Language in Unscripted Spoken English, examined approximately 22.1 million words from unscripted science and technology podcasts.

The researchers identified an increase in certain words associated with language-model output following the public release of ChatGPT.

Comparable synonyms did not show the same pattern.

This matters because the researchers were examining spoken language, not simply text produced directly by AI.

A separate 2026 investigation, Empirical Evidence of Large Language Model’s Influence on Human Spoken Communication, analysed a much larger body of podcast material and included an experimental study involving 496 participants.

It reported evidence that interactions with language models could influence subsequent human vocabulary choices.

These findings need careful interpretation.

They do not prove that AI is responsible for every observed change in vocabulary.

Nor do they establish that AI is already rewriting English grammar.

But they support a genuinely interesting possibility.

Language produced by AI may be finding its way into language produced by people.

We are accustomed to thinking of AI as a system that learns patterns from human language.

Increasingly, we may need to think about the reverse process.

Humans are also learning patterns from AI.

Return to our English teacher.

A student writes:

“We discussed about the problem.”

AI corrects it:

“We discussed the problem.”

The correction is appropriate for standard English.

But consider the assumptions behind it.

Which variety of English is being used as the reference point?

Which communicative context is being considered?

And what happens when a construction is common within a particular community of international English users?

Research on English as a Lingua Franca has long challenged the idea that successful international communication must always be measured against native-speaker norms.

More recently, researchers examining generative AI from a Global Englishes perspective have raised concerns that AI systems may privilege dominant varieties of English while insufficiently representing the diversity of English used around the world.

This is particularly important in language education.

An AI system might identify an expression as incorrect because it differs from the standard variety it has been trained or instructed to favour.

That does not necessarily tell us whether the expression is understandable.

It does not necessarily tell us whether it is conventional within a particular speech community.

And it certainly does not tell us whether the speaker has communicated successfully.

None of this means teachers should abandon grammatical correction.

The student still needs to know that discuss about is not standard in this construction.

But perhaps the explanation should go further.

We can teach learners that a construction is non-standard without suggesting that it is meaningless.

We can explain why one form is expected in an examination or professional document while recognising that communication may succeed despite departures from that form.

Correctness, appropriateness and intelligibility are related, but they are not identical.

AI makes that distinction more important, not less.

I suspect we may see an increasingly visible distinction between two broad tendencies.

The first is AI-assisted formal English.

Professional emails, reports, academic papers, administrative communication and other written documents may become more standardised.

Their language may become more predictable, grammatically conventional and internationally recognisable.

The second is the English of everyday human interaction.

The English of humour, relationships, identity, music, social media and spontaneous conversation.

This language will continue changing because people continue changing.

But the distinction is unlikely to be absolute.

People carry expressions from informal conversation into professional writing.

AI encounters emerging forms of language.

And humans may adopt expressions they first encountered in AI-generated text.

This is a hypothesis about how English might develop, not a settled scientific prediction.

But the educational implications are already worth considering.

If AI can produce highly polished written English for almost anyone, what does a polished piece of writing tell us about its author?

A student may submit an excellent essay without being able to explain the reasoning behind it.

A professional may send a sophisticated email but struggle to discuss the same subject spontaneously.

A learner may produce grammatically accurate sentences with AI assistance while still finding it difficult to construct those sentences independently.

The finished text may be excellent.

But the finished text and the writer’s independent linguistic competence are not the same thing.

The English someone can produce with assistance is not necessarily the English they can use when communication happens in real time.

Let’s return to the student’s original sentence.

“I am working here since five years.”

Today, an English teacher would normally correct this in a standard-English lesson.

But suppose a particular construction becomes widespread among a community of English users.

Suppose it is consistently understood.

Suppose speakers begin treating it as an ordinary feature of their shared variety.

At what point does it stop being an error and become part of the language?

There is no universal numerical threshold.

Language change is not simply a matter of counting how many times an expression appears.

A form may become conventional within one speech community while remaining non-standard in another.

Its acceptance depends on social use, recognition, identity and the expectations of the people communicating.

AI introduces an additional complication.

An expression may become more common because AI repeatedly generates it.

Or less common because AI repeatedly removes it.

If an AI system begins treating a previously non-standard construction as acceptable, does that make it correct?

Not automatically.

A language model is not a linguistic authority simply because it produces a confident answer.

But neither can we ignore the possibility that millions of people may begin adopting the expressions it generates.

If AI influences what people say, and widespread usage influences what counts as conventional, how much linguistic authority could AI eventually acquire?

Let’s finish with the most provocative possibility.

Could English eventually develop expressions or grammatical conventions that originated with AI rather than human speakers?

In principle, AI can generate novel expressions.

People may encounter those expressions, find them useful and begin using them.

If a speech community adopts a form consistently, it could eventually become conventional.

But we should distinguish a plausible mechanism from a demonstrated historical change.

The research discussed here provides evidence that AI can influence human vocabulary choices.

It does not yet establish that a new, widely accepted grammatical convention of English originated independently with AI and subsequently spread through human communities.

That would require careful evidence tracing its emergence and adoption.

Nevertheless, the possibility is extraordinary.

For centuries, dictionaries and grammar books have largely described, organised or prescribed language produced by people.

AI is different.

It does not merely describe language.

It generates language that people may imitate.

And when enough people begin imitating a form, the distinction between the language produced by machines and the language produced by humans becomes increasingly difficult to maintain.

We tend to assume that AI is learning English from us.

But increasingly, we are also learning English from AI.

Who will be teaching whom?