Recently, I have found myself thinking about a pattern I am beginning to notice in work with autistic adults.

Clients are describing conversations they have been having with conversational AI. As they shared examples, I found myself in an unusual position. The insights brought into session were thoughtful. The interpretations were nuanced. The language was sophisticated. Yet something felt different.

The formulations often seemed less like understandings they are still actively developing and more like polished interpretations that had arrived already assembled.

This is not a criticism of AI. I have found it useful myself, and many autistic adults are finding genuine benefit in these systems. They can help organize thoughts, clarify emotions, generate language for difficult conversations, and offer possible explanations for confusing social experiences. For some people, they may also function as one of the most consistently available conversational partners they have.

That is part of what makes this clinically complicated.

The question is not whether it is useful. Clearly, it can be. The question is what role it is beginning to play in the cognitive, emotional, and interpersonal lives of autistic adults - particularly those with histories of chronic misunderstanding, social exhaustion, or isolation.

I wonder whether these systems may be creating a new form of social compensation that resembles masking in ways we have not yet adequately conceptualized.

Traditionally, masking refers to the effort autistic people expend to navigate environments organized around non-autistic communication styles and expectations. People learn scripts. They study interaction. They develop strategies for interpreting situations that may not come intuitively. Over time, many become highly skilled at producing responses that are socially acceptable, legible, or expected.

But masking is not only imitation. It is also a strategy for remaining intelligible in environments where spontaneous self-expression has repeatedly been met with confusion, rejection, correction, or exclusion. In that sense, masking is not simply about appearing more neurotypical. It is about reducing interpersonal friction in a world that often misreads autistic forms of communication and relating.

Many autistic adults spend significant amounts of time trying to interpret social situations: Did I misread that? Why did that interaction shift? Was I rude? Am I overreacting? What am I supposed to say next? Historically, those questions may have been brought to a friend, therapist, mentor, or family member.

Now, increasingly, they can be brought to an AI system.

I see a related phenomenon in the emails some clients send between sessions. These messages can contain language that is highly polished, psychologically sophisticated, and strikingly self-reflective. Clients arrive with detailed formulations of their experience, nuanced explanations for interpersonal conflict, and extensive analyses of relational patterns that may previously have been difficult for them to articulate.

On one level, this can be genuinely useful. Having language for an experience is often a necessary condition for understanding it. Many autistic adults have spent years struggling to explain themselves to others and, at times, to themselves. These tools can help organize thoughts that might otherwise remain diffuse, inaccessible, or difficult to communicate.

At the same time, I sometimes find myself wondering where the client’s insight ends and the system’s interpretation begins.

The issue is not authorship. The issue is ownership.

The relevant question is not whether a thought emerged in dialogue with a machine. All thinking is relational and scaffolded in some way. The question is whether the person has actually worked with the idea: tested it against experience, revised it, tolerated uncertainty around it, and integrated it into their own understanding. Increasingly, part of the clinical task may involve helping clients distinguish between an idea that feels resonant and an idea that has been metabolized.

These systems complicate this because they almost always produce an answer, and they rarely present that answer with visible uncertainty. They do not get tired of hearing the same story repeatedly. They do not appear confused. They do not often respond with, "I do not know enough to say." Most importantly, they are highly effective at generating explanations that feel coherent.

That coherence can be useful. It can also create false confidence.

Human relationships require tolerance for ambiguity. We often do not know why another person behaved as they did. We rarely have direct access to another person’s motives. Much of social understanding involves forming hypotheses, reality-testing them, gathering additional information, and revising conclusions as new information emerges.

This technology compresses that process. It is very good at generating plausible interpretations. It is far less capable of determining which interpretation is actually true.

The result is that a person may leave an interaction with it feeling as though they understand a situation when what they actually have is a well-constructed hypothesis.

For many autistic adults, that is an important distinction. The clinical difficulty is often not an absence of possible explanations. It is difficulty determining which explanation best fits the available evidence, tolerating the discomfort of not knowing, and remaining open to revision when new information appears.

AI can inadvertently short-circuit those capacities. A client who is prone to self-blame may receive sophisticated explanations for why everything is their fault. A client who is prone to externalization may receive equally sophisticated explanations for why everyone else is wrong. In both cases, the interpretation may feel persuasive because it is framed in psychologically plausible language. But plausibility is not the same as accuracy, and coherence is not the same as integration.

This is where the comparison to masking becomes especially useful.

Seen through this lens, conversational AI can function as an ideal masking partner. It is always available. It does not get impatient. It does not visibly react to awkward phrasing. It can take fragments, contradictions, and partially formed emotional experience and convert them quickly into coherent narrative.

Sometimes that process is supportive and adaptive. But it can also create a subtle gap between lived experience and presented understanding.

A person may begin to rely on generated language not simply to communicate what they already know, but to determine what they think, what they feel, and how they should interpret events. Over time, what looks like increasing insight may sometimes reflect something else: a form of outsourced social cognition, in which difficult interpersonal experiences are repeatedly translated into polished explanations by an external system.

This will not always be obvious clinically. In fact, it may present as progress. A client may sound more articulate, more reflective, more relationally minded, and more emotionally literate. Some of that may reflect genuine development. Some of it may reflect a collaboration between person and machine that is difficult to disentangle.

To be clear, not every use of these systems by an autistic person should be understood as masking. At times it is better understood as accommodation.

There is a meaningful difference between using external support to reduce unnecessary cognitive or communicative strain and using it in a way that distances a person from their own process of meaning-making. Using AI to organize notes, rehearse a difficult conversation, draft an email, or find language for an already recognized feeling may be no more problematic than using a calendar, spell-check, or therapy worksheet.

The concern emerges when assistance becomes substitution: when generated interpretation is experienced as self-knowledge before it has been adequately tested, when plausible language hardens into certainty, and when sounding coherent becomes more valued than remaining curious.

What concerns me even more, however, is not the technology alone.

It is the loneliness to which it is responding.

Many autistic adults describe a long history of struggling to find people who genuinely understand them. Some have few close relationships. Others have experienced repeated misunderstanding, rejection, or social fatigue. For these individuals, conversational AI offers something deeply compelling: a conversational partner that is available on demand, responsive, apparently validating, and persistently engaged.

In some cases, it may function as the most consistently available interpreter of their social world.

That reality makes simple cautionary advice feel inadequate. If it is meeting needs that have gone unmet elsewhere, encouraging restraint may feel, to the client, less like support and more like a request to return to isolation.

The question, then, is not whether autistic adults should use it. Many will derive real benefit from it. The more clinically useful question is what function it is serving. Is it operating as a communication aid? A cognitive scaffold? A rehearsal space? A translator? A substitute for reciprocal relationship? Perhaps most importantly, does it help the person engage more effectively with actual others, or does it increasingly become the place where social understanding occurs?

I do not think we yet know the answer.

What does seem increasingly clear is that this technology is altering how some autistic adults process social experience. Like masking, it may reduce friction, uncertainty, and misunderstanding. Like masking, it may also create a gap between visible competence and the underlying process that produced it.

The central question may not be whether conversational AI helps autistic people communicate. Clearly, it can. The harder question is whether it supports the development of more fully owned meaning - or whether, in reducing friction, it sometimes begins to take over the work of meaning-making itself.

If that is happening, then the issue is not merely technological. It is developmental, relational, and clinical.

And it asks something important of therapists: to listen not only for whether an interpretation is sophisticated, but for whether it is inhabited; not only for coherence, but for ownership; and not only for insight, but for the process by which that insight was formed.

Share this post: