There is a particular kind of failure that no standard cognitive test catches well. A person understands the task. They can explain what needs doing and when. They have the planning ability, the impulse control, and the verbal facility to carry it out. And then the moment arrives, and nothing happens. The intention was formed, stored, and never retrieved. Psychologists call the capacity that fails here prospective memory, and a study published in June in Scientific Reports caught it happening with unusual clarity.

Daniela Nürnberg and Mareike Altgassen, of the Department of Psychology at Johannes Gutenberg University Mainz, compared 30 autistic adults with 30 non-autistic adults matched for age, gender, and non-verbal reasoning ability. Mean age was 36 in both groups. Every participant completed a prospective memory task and three separate tests of executive function measuring planning, inhibition, and generativity.

On the executive function tests, the two groups were statistically indistinguishable. On the prospective memory task, they were not.

What Prospective Memory Is, and Why It Is Not Obvious

Prospective memory is remembering to do something later. Taking the laundry out when the cycle ends. Buying bread when you pass the bakery. Mentioning a thing to a colleague when you see them. It sounds trivial and it is not, because it requires several processes to cooperate across a gap in time.

You form an intention and store it. Then you go do other things, which is the whole difficulty, because the intention has to survive being ignored. At some later moment a cue arrives, and you have to notice it, recognize it as a cue, interrupt whatever you were doing, retrieve the stored intention, and act. Failure at any one of those steps produces the same visible outcome: the thing did not get done.

Researchers distinguish two types by what triggers retrieval. Event-based tasks are cued by something in the environment, like seeing the bakery. Time-based tasks depend on the clock, like taking a cake out after 60 minutes, and are consistently harder, because nothing external prompts you. You have to monitor elapsing time yourself while doing something else, which is why time-based tasks place heavier demands on executive control. That distinction turns out to matter for interpreting this literature, and we will come back to it.

This study used an event-based task, and a deliberately simple one. Participants were shown a box and seven coins and told to put a coin in the box whenever the experimenter began a new instruction with a specific sentence. The sentence was spoken four times across the session. The task was practiced first, and everyone demonstrated they understood it. Maximum score was four.

The simplicity was intentional. As the authors note, the Token Task places relatively low demands on executive function compared with prospective memory tasks that require interrupting an ongoing activity or independently monitoring the clock. It was chosen precisely because the sample would include people for whom a more demanding task would be unusable.

Who This Study Included

The design choice that makes this paper unusual has nothing to do with the task. It has to do with who was allowed to participate.

Roughly a third of autistic people have an intellectual disability, and close to another quarter fall in the borderline range. Yet as the authors note, autism research has concentrated since the 1980s on autistic people without intellectual disability, and every previous study of prospective memory in autistic adults excluded participants with an IQ below 70. An entire body of literature about a population had systematically omitted a large part of it.

Nürnberg and Altgassen recruited across what they call the wider autistic spectrum, drawing participants from residential groups and sheltered workshops as well as mailing lists and flyers. Some participants had external guardianship, and consent was obtained from them and, where applicable, their caregivers. Adults taking psychotropic medication, including stimulants, were excluded, as were those with severe psychiatric or neurological conditions. Within the autism group, 36.6 percent reported a co-occurring psychological condition, most commonly an affective disorder, with five participants reporting ADHD.

Accessibility was built into the procedure rather than assumed. Testing happened wherever the participant preferred, and the preference diverged sharply: 17 of the 30 autistic participants chose to be tested at home, compared with 4 of the 30 controls. A simplified written timetable sat on the table throughout so participants could see the shape of the session in advance. A single experimenter ran every session. These are small accommodations, and they are the reason a study like this can recruit a population that research has otherwise found difficult to reach.

Seventy adults were recruited. Ten were excluded during testing because cognitive or physical impairments made the tasks impossible to complete. That exclusion is worth sitting with. It is not a flaw in the study so much as a measurement of how few instruments exist that can reach this population at all, a gap the authors name explicitly as a priority for future work.

The Detail That Explains the Most

Buried in the discussion is an observation that does more explanatory work than any of the statistics.

Every participant understood the task at the start and executed it correctly in practice. By the end of the session, some participants in both groups who depend on external support could still recall that a coin was supposed to go into the box. What they had lost was the sentence that was supposed to tell them when.

The intention survived. The cue did not.

That distinction has immediate practical consequences, and the authors draw them. The cue in this task was a full spoken sentence, which places a load on working memory and episodic memory before any prospective remembering can happen. They suggest that shortening the cue to a single word, or replacing the spoken cue with a visual one, might substantially change performance. The second suggestion carries extra weight given how common auditory processing differences are in autism, and given that this sample showed lower scores on the verbal measure.

In other words, part of what looked like a prospective memory difficulty may have been a cue design problem. That is not a dismissal of the finding. It is a specific, testable hypothesis about where the breakdown occurs, and it points at something changeable. A cue that has to be held in memory before it can be recognized is a different kind of cue from one that is simply present in the environment when the moment arrives.

What the Executive Function Tests Did and Did Not Find

The executive function results were mostly null, and the authors are candid that this surprised them. They had expected reduced planning performance based on prior work, and did not find it.

There is a plausible methodological reason. The planning measure was the Tower of Hanoi, and it was too hard for a meaningful number of participants. Only 24 of 30 autistic participants and 26 of 30 controls completed the three-disc version, and just 15 and 21 completed the four-disc version. Autistic participants did tend to take longer without reaching significance. The authors recommend the Tower of London for future work because its difficulty can be adjusted in finer increments, which would keep more participants in the analysis.

One difference did emerge, and it is easy to miss because it sits inside an otherwise null result. On the Go-NoGo task, autistic participants were no less accurate, either overall or on the trials requiring them to withhold a response. But they were significantly slower to respond on the trials requiring action, at p = 0.02. The authors then reran the analysis excluding every participant with ADHD, on the reasonable suspicion that comorbid ADHD might be driving it. The effect did not weaken. It strengthened, to p = 0.007, with a larger effect size than in the full sample. Accuracy remained non-significant.

That pattern points away from inhibition and toward processing speed. These participants could withhold a response as reliably as anyone. They took longer to produce one. Whether that has any bearing on the prospective memory result is not something this study can answer, but a task that requires noticing a fleeting cue and acting before the next activity begins is exactly the kind of task where response speed might matter.

The more interesting analysis, though, is what happened when the researchers looked across both groups rather than between them. Better executive function was associated with better prospective memory throughout the sample. And in a regression including inhibition, fluency, and planning as predictors, only one variable came out significant: how long a participant took to complete the planning task, at p = 0.007.

The group comparison and the individual-differences analysis are answering different questions, and both answers matter. Being autistic did not predict planning performance in this sample. Planning performance did predict whether a person remembered to act on an intention, in autistic and non-autistic participants alike.

What the Caregivers Saw

There is one more result in this study that deserves more attention than it is likely to get.

Alongside the laboratory tasks, caregivers completed a questionnaire called the DEX, a 20-item measure of everyday executive difficulties covering behavior, cognition, and emotion. It asks how often specific problems actually occur in daily life, rated from never to very often. It is not a test. It is an observation by someone who watches the person live.

Within the autistic group, higher caregiver-rated everyday executive dysfunction was associated with poorer prospective memory performance, and with poorer category fluency. The laboratory tests of planning, inhibition, and generativity had shown no group difference at all. The people who spend their days with these adults were reporting executive difficulties that tracked a real outcome, in a sample where the standard executive function battery had found nothing to report.

This is worth stating carefully, because the study cannot support the strongest version of the claim. The DEX was administered only to the autistic group, on the expectation that there would be too little variability among controls to be informative. So there is no comparison, and no way to say that caregiver ratings outperformed lab tasks in any general sense. What can be said is narrower and still interesting: within this group, an everyday observational measure correlated with prospective memory performance, and the timed cognitive tasks that are the field’s usual instruments did not distinguish the groups at all.

A caregiver questionnaire and a computerized task are not measuring the same thing. That is a familiar finding in executive function research generally, and it shows up here inside a single study, in a population where the gap between test performance and daily functioning has particular consequences.

Why the Literature Disagrees With Itself

Anyone reading across this field will notice that studies of prospective memory in autism do not agree, and the disagreement has a shape worth understanding.

In 2014, Williams and colleagues tested autistic adults and matched controls on both types of prospective memory task. They found diminished time-based performance and undiminished event-based performance. The title of the paper says exactly that. Their sample, as the title also specifies, consisted of intellectually high-functioning adults. In 2024, Groenman and colleagues went considerably larger, with 82 autistic and 111 non-autistic adults, and found prospective memory spared on both task types. A separate 2024 study by Faustmann and Altgassen also found no significant group differences.

This study used an event-based task, the type that has most often come out intact, and found a significant difference anyway.

The likeliest explanation is the one running through this entire piece. The studies finding intact event-based performance excluded autistic adults with intellectual disability. This one did not. And there is a precedent pointing the same direction: a 2016 study by Sheppard and colleagues compared autistic children with low and high support needs against matched non-autistic children, and found that the children with high support needs, but not those with low support needs, performed worse on event-based prospective memory.

Two studies, in different age groups, now suggest the same boundary. Event-based prospective memory may be relatively preserved in autistic people without intellectual disability, and diminished among those with higher support needs. If that is right, then the literature is not really contradicting itself. It is describing different populations and reporting the difference as though it were about autism in general.

What the Study Does Not Show

This is a small, cross-sectional, correlational study, and several things follow from that.

It cannot establish causation. Better planning was associated with better prospective memory, and the regression identifies planning as a predictor, but that is not evidence that improving planning improves prospective memory. The authors call explicitly for longitudinal work to examine causal relationships over time.

It is not an intervention study, and nothing in it tests whether any support works. The authors do recommend supports, including reminders, timers, agendas, calendars, to-do lists, structure and routine, breaking complex tasks into sub-steps, repeatedly articulating intentions, and assistive technologies. Those recommendations are reasoned inferences from the pattern of findings and from separate literatures. They are not results of this experiment. Anyone citing this study as proof that a particular support improves prospective memory is citing something that was not measured.

The authors list their own limitations plainly. Working memory was never assessed, which is a real gap given that lost cues may reflect working memory load rather than prospective memory as such. Detailed autism diagnostic assessment and comprehensive cognitive testing were skipped to keep session length manageable for the participants. The two groups differed on verbal ability, although the main finding held when that was controlled for. And the sample spanned a wide range of cognitive profiles rather than isolating any one, which the authors say should be addressed by future studies focused specifically on autistic adults with intellectual disability.

The sample is also small, and the exclusion of ten participants during testing means the final groups were shaped partly by which tasks happened to be usable. With 30 people per group, individual variation carries a great deal of weight.

Why This Matters Beyond Autism

The finding that should travel furthest is the dissociation itself.

A group of adults performed at the level of matched peers on planning, on inhibition accuracy, and on verbal fluency, and still failed measurably more often at remembering to act on an intention. Whatever prospective memory requires, the standard executive function battery did not capture it.

This is the formal version of a pattern that anyone working with executive function encounters constantly, and it complicates a common assumption. Performing adequately on structured cognitive tasks does not establish that a person will follow through in ordinary life, because ordinary life supplies neither the instruction nor the moment. It supplies a cue, sometimes, if you notice it. A test session tells a participant what to do and then presents the conditions for doing it. A Tuesday afternoon does neither.

The authors make a related point that deserves emphasis. Their results, they argue, show why executive functions have to be differentiated rather than treated as a single faculty, because an autistic person can show full functionality in some and difficulty in others. They connect this directly to the neurodiversity perspective and to the argument that autism should not be framed as a purely deficit-based condition. A study that found equivalent performance on three of four measured domains is reasonable evidence for that view.

What remains is a narrow, specific difficulty with a process that daily life depends on more heavily than its name suggests, in a population that research has mostly declined to study, measured with instruments that a tenth of the people recruited could not use. That is a useful thing to have established, and a reasonable place for the next round of work to begin.