Are online dementia tests accurate? What they can and can't tell you

Short answer: Some are, within limits — and most of what's online isn't a test at all. The best evidence is a 2024 Cochrane review of self-administered cognitive tests, which found sensitivity ranging from 55% to 100% and specificity from 45% to 100%, and concluded there is insufficient evidence to recommend any single one. Even the best-performing tool in that review flagged about a third of healthy people as possibly impaired. Which means the result most people actually get — a worrying score — is far more likely to be wrong than right. No online test can diagnose dementia. What a good one can do is help you decide whether to book an appointment.

This is educational, not medical advice. If you or someone close to you has noticed memory or thinking changes that are new, getting worse, and affecting daily life, that warrants a conversation with a clinician regardless of what any online test says.

Three different things get called "an online dementia test"

Almost all the confusion here comes from collapsing three unrelated instruments into one phrase.

1 · A validated cognitive screening test — a real instrument with published diagnostic-accuracy data, designed to detect current impairment and studied against clinical diagnosis. SAGE, from Ohio State University, is the best-known free example. It answers: is something changing right now?

2 · A risk assessment — a structured look at the modifiable factors shaping your future risk: blood pressure, hearing, activity, sleep, blood sugar. It measures nothing about your current cognition and is not a screening test. It answers: what could I change?

3 · A quiz — ten questions, an instant score, no citations, no validation, and frequently an email capture or care-home enquiry form at the end. It answers nothing, and it's the majority of what ranks for this search.

The first two are useful for different people. The third is why this article exists.

What the research actually found

The strongest available answer is a Cochrane diagnostic-test-accuracy review published in December 2024, which pooled 11 studies covering 2,303 participants across six self-administered tools.

Its headline numbers:

• Sensitivity: 55% to 100% — how often the test correctly flags someone who does have dementia
• Specificity: 45% to 100% — how often it correctly clears someone who doesn't

And its conclusion, stated plainly: "There is insufficient evidence to recommend the use of any single self-administered cognitive assessment tool" (Naeem, McCleery, Quinn et al., Cochrane Database of Systematic Reviews, 19 Dec 2024).

Two findings inside it are worth more than the headline. First, self-administered tools performed broadly similarly to clinician-administered screening tests — so the problem isn't that doing it yourself is uniquely unreliable. Screening tests in general are blunter than people assume. Second, only one tool had enough studies to meta-analyse: Test Your Memory (TYM) at a threshold of 42/50, which came out at 94% sensitivity (95% CI 90–96) and 66% specificity (95% CI 45–82).

Hold on to that second number, because it's the one nobody explains.

The number that actually decides what your result means

Sensitivity is the number every test markets itself on, and it's the less useful one for you. A test can hit 94% sensitivity trivially by flagging almost everybody. What determines whether your worrying result means anything is specificity combined with how common the condition is in people like you.

Here is that arithmetic, done openly. It's an illustration, not a prediction, and the assumptions are on the table.

Take 1,000 people aged 65–74 who take TYM. Alzheimer's dementia affects about 5.2% of that age group (Alzheimer's Association, 2026 Alzheimer's Disease Facts and Figures), so call it 50 people. Apply the review's pooled figures — 94% sensitivity, 66% specificity:

Of the 370 people told they might have a problem, 47 actually do — about 1 in 8. The other seven out of eight got a frightening result from a test that was working exactly as designed.

Now flip it. Of the 630 people who got a reassuring result, 627 were genuinely fine. That's the quietly important part: these tools are far better at reassuring you than at alarming you. A calm result is worth something. A worrying result mostly means go and get properly assessed — which, to be fair, is all a screening test ever claimed to do.

Two honest adjustments. People who seek out a memory test are more likely to have a problem than the general population, and that improves the odds — but not as much as you'd hope. Even if one in five people taking the test had dementia, roughly 41% of those flagged would actually have it. Still fewer than half. And in the other direction, the confidence interval on that 66% specificity runs from 45% to 82%, so the true figure could be considerably worse than this illustration assumes.

What about SAGE — the one doctors actually mention?

SAGE (the Self-Administered Gerocognitive Exam) is the strongest free option and the one a clinician is most likely to hand you. Developed at Ohio State University's Wexner Medical Center, it's designed to be done at home with pen and paper and brought to an appointment — which is the right way to use any of this. Its published validity data reports 95% sensitivity and 62% specificity for distinguishing mild cognitive impairment from healthy ageing, and 95% / 95% for distinguishing dementia from healthy ageing at a cutoff of 16 or below.

That dementia figure is genuinely good. But note the pattern running through the whole field: accuracy is markedly worse for mild cognitive impairment than for established dementia — and mild cognitive impairment is exactly what someone searching this at 1am is worried about. The subtler the change, the harder it is to detect, by any method. (See what mild cognitive impairment actually is and how to check your memory at home if you want to run one properly.)

One caveat, stated because it applies to us too: those SAGE figures come from the team that developed it, and developer-reported accuracy tends to look better than independent replication. That's a general rule in diagnostics, not an accusation — and it's why the Cochrane review's independent range matters more than any single tool's own numbers.

How to tell a real tool from a lead-generation quiz

Five checks. They take about a minute and they're the most practical thing in this article.

• Does it name its validation? A real instrument cites the studies behind it and reports its own sensitivity and specificity. If a page claims "clinically accurate" or "80–90% accurate" without a citation you can follow, treat the number as marketing.
• What happens when you finish? A screening tool tells you to take the result to a doctor. A lead magnet asks for your email, your phone number, or your postcode "to find care options near you."
• Who owns the site? A university medical centre, a national charity and a care-home operator have very different reasons for wanting you to take a test. All three can be useful; only one of them benefits from you scoring badly.
• Does it say what it can't do? Any legitimate tool states, unprompted, that it does not diagnose. Absence of that sentence is a strong signal on its own.
• Does it distinguish screening from diagnosis from risk? If a page uses "test", "screen" and "assessment" interchangeably, it hasn't thought carefully about what it's measuring.

Our own conflict of interest, stated plainly

We run an online tool. So the fair thing is to apply the checks above to ourselves.

Solenna's risk profile is not a cognitive test, and the accuracy figures on this page do not apply to it. It doesn't measure your memory, your attention or your processing speed, and it cannot detect impairment. It has no sensitivity or specificity, because sensitivity and specificity describe how well an instrument detects a condition — and detecting a condition is not what it does.

What it is: a structured review of your modifiable risk factors, based on the published evidence on what influences dementia risk across a lifetime. It tells you where you stand on the things you can change. It does not tell you whether you have dementia, and it cannot predict whether you will get it. If you have symptoms right now, it is the wrong tool and you should see a clinician.

We'd rather say that clearly than sell you a number.

When to stop testing and see a doctor

No online result — good or bad — should override the warning signs. If changes are new, getting worse, and interfering with daily life — getting lost somewhere familiar, struggling with a long-routine task, repeating questions within one conversation, or people close to you noticing — book an appointment regardless of your score. We've set out the full list in normal forgetfulness or an early warning sign?

Many of these have causes that are treatable and unrelated to dementia — thyroid problems, B12 deficiency, medication side effects, depression, poor sleep. That's exactly why they deserve a proper assessment rather than another quiz.

If you're worried but symptom-free, you're likely on the wrong page

Most people searching "are online dementia tests accurate" don't have symptoms. They have a parent who declined, a bad week of forgetfulness, and a question they can't put down. For that person a screening test is close to useless: it's built to detect a condition they almost certainly don't have, and as the arithmetic above shows, it's likelier to frighten them wrongly than to tell them anything.

The forward-looking question has better evidence behind it. Researchers following 12,409 American adults for about 26 years found that those who reached midlife without high blood pressure, type 2 diabetes or smoking averaged nearly 13 more years free of dementia than those with all three (Hu et al., Neurology Open Access, 5 Aug 2026) — an association rather than proof of cause, as the authors state, but a substantial one. More broadly, around 45% of dementia cases worldwide are attributed to modifiable risk factors: a population-level estimate under ideal conditions, not a personal guarantee.

None of that shows up on a memory quiz. It shows up in a risk profile.

How we research and review our content →

This article is educational and is not medical advice, diagnosis, or treatment. The accuracy figures cited — from the 2024 Cochrane review of self-administered cognitive assessment tools and the published SAGE validity data — describe cognitive screening instruments in general, not Solenna. Solenna's risk profile is not a cognitive test and does not detect or measure cognitive impairment, so those figures do not apply to it. The worked example using 1,000 people is an illustration based on the cited pooled figures and age-specific prevalence, not a prediction about any individual. Solenna is not a medical device and does not diagnose, prevent, treat, or cure Alzheimer's disease or any other condition. Risk reduction means lowering probability, not eliminating it; individual results vary and no outcome is guaranteed. If you are concerned about memory or thinking changes, consult a qualified healthcare professional.