Elli Cares
Back to Insights

September 4, 2023

AI-powered cognitive assessments and early detection

An older man sitting across from a clinician in a doctor's office

Your dad has started repeating stories. Your mum can't remember whether she paid the power bill. Is it normal ageing, stress, or something that needs a closer look? For many families, the hardest part is not knowing, and not knowing who to ask.

Doctors have traditionally answered that question with pen-and-paper memory tests and a good conversation. Now, AI and digital tools are starting to play a part. Some are tests you take on a tablet. Others look for patterns in how people use their phones or smartwatches.

This article explains what AI-powered cognitive assessments are, what the evidence says about them so far, where they fall short, and what to do if you're worried about someone you love.

Why early detection matters

Mild cognitive impairment, often called MCI, is a stage where memory or thinking changes are noticeable but don't yet stop someone living independently. It can be an early sign of Alzheimer's disease, though not everyone with MCI goes on to develop dementia.

It's also very often missed. A 2026 scoping review in Alzheimer's & Dementia notes that MCI remains undiagnosed in more than 90 percent of people who have it.

Finding changes earlier gives people time. Time to rule out other causes that can be treated, such as medication side effects, low mood, sleep problems or vitamin deficiencies. Time to plan, make decisions while they can still lead them, and put support in place. And for some people, time to be assessed for newer Alzheimer's treatments, which are designed for the early stages of the disease.

What is an AI-powered cognitive assessment?

The term covers a few different kinds of tools.

Digital versions of familiar tests. Some tools take established paper tests and move them onto a tablet. The Mini-Cog, for example, combines a short word-recall task with drawing a clock. A digital version records not just the final drawing but how it was drawn, including pauses and hesitations, and uses machine learning to help score it.

Purpose-built digital tests. Other tools are new tests designed for screens. The Alzheimer's Association's cognitive assessment page for clinicians notes that the FDA has cleared several digital cognitive testing tools for marketing, and that computerised tests have the advantage of being given exactly the same way each time. The FDA has also cleared an AI-based test from Cognetivity Neurosciences that can be taken on an iPad in about five minutes.

Passive signals from everyday devices. Researchers are also studying whether phones and smartwatches can pick up changes in thinking over time, without anyone sitting a test. In the Intuition study, reported in Nature Medicine in 2025, more than 23,000 US adults used an iPhone and Apple Watch for up to two years while researchers collected activity, sleep, device use and regular cognitive tests. The early results suggest that remote cognitive testing is reliable at large scale. The larger question, whether passive signals alone can detect MCI, is still being analysed.

What AI and digital tests do well

There are real reasons for optimism.

  • Consistency. A digital test is given the same way every time, which reduces differences between clinicians and clinics.

  • Speed. Many take only a few minutes, which matters in a busy primary care appointment.

  • Reach. Some can be completed at home or in a local clinic, which could help people who live far from specialists.

  • Tracking over time. Repeating the same test regularly can show change more clearly than a single snapshot.

Real-world use is growing. A 2025 study in the Annals of Family Medicine tested a digital cognitive assessment in seven primary care clinics over a year. More than 1,700 patients aged 65 and over completed it. The authors concluded that these tools could be a useful first step in finding people who need a fuller assessment.

Where the evidence is still thin

The same research also shows the limits.

In the Annals of Family Medicine study, fewer than 11 percent of eligible visits resulted in a completed test. In more than half of visits, doctors chose not to offer it, often because of time pressure. And almost half of the patients who did the test scored as impaired or borderline. That doesn't mean all of them had dementia. It means a positive screen is the start of a conversation, not a diagnosis.

The 2026 scoping review was more cautious still. It looked at 79 self-administered digital cognitive tests and found that only four met the researchers' criteria for community and primary care screening. Common problems included study groups that didn't reflect the wider population, inconsistent reporting of accuracy and limited comparison with biological markers of Alzheimer's. The authors concluded that the evidence base is not yet complete enough for widespread clinical use.

There are other questions too:

  • Bias. A tool trained mostly on one group of people may be less accurate for others, including people with different languages, education or cultural backgrounds.

  • Anxiety. An unexpected "impaired" result can be frightening, particularly without a clinician there to explain it.

  • Privacy. Passive monitoring collects a lot of personal data. People should understand and agree to what's being collected.

None of this means AI tools aren't useful. It means they work best as one part of an assessment led by a clinician, not as a replacement for one.

AI tests and blood tests: a changing picture

Cognitive tests are only one part of a fast-moving field. In 2025, the FDA cleared the first blood tests to help assess Alzheimer's disease. According to Alzheimer Europe, Roche's Elecsys pTau181 test was cleared in October 2025 for use in primary care, to help rule out Alzheimer's-related changes in adults 55 and over who are showing signs of cognitive decline. It followed Fujirebio's Lumipulse test, which was cleared in May.

These blood tests are designed for people who already have symptoms, not as screening for healthy people, and results are meant to be read alongside other clinical information. Over time, digital cognitive tests may help decide who needs a blood test, and blood tests may help decide who needs specialist care.

Our article on new breakthroughs in dementia testing covers more of these developments.

Elli Cares doesn't diagnose dementia or run cognitive tests. What it does is help families and care teams notice change over time. Daily wellbeing check-ins, medication reminders and family observation notes build a simple record of how someone is managing, and a downloadable report makes it easy to bring that picture to the doctor. See how Elli supports seniors and families.

What to do if you're worried about someone's memory

If you've noticed changes, there's no need to wait for a new app or test. A few steps help:

  1. Write down what you've noticed. Note specific examples, when they happened and whether they're getting more frequent. Our guide to the 10 warning signs of dementia can help you know what to look for.

  2. Look for other changes. Changes in mood, sleep, appetite, walking or confidence can all be relevant. Even a change in sense of smell has been linked to later memory problems, as our article on the smell test and dementia explains.

  3. Book an appointment with their doctor. Go together if you can, and bring your notes. Ask whether a cognitive assessment makes sense, and which one.

  4. Ask about other causes. Many treatable conditions can affect memory. A good assessment will check for them.

  5. Ask what comes next. If a test suggests a problem, ask what further checks are recommended and who will explain the results.

If you're unsure where someone sits on the spectrum, our guide to early versus mid-stage dementia explains what different stages can look like.

Looking ahead

AI won't replace the doctor who listens carefully to a worried family. But it may help that doctor see more clearly, sooner. Quicker, more consistent tests, combined with blood tests and a better record of everyday life, could mean fewer people waiting years for answers.

The research is moving quickly, and some of the most promising tools still need better evidence. For now, the most useful thing families can do is notice, write things down and start the conversation early.

Text