No. No AI model, app or chatbot can tell you the date you will die. What the best research models can do is rank people by risk: they are better than chance at saying who in a large group is more likely to die within the next few years. The most-cited example, the Danish life2vec model, was right about 79% of the time on a balanced test set where guessing would score 50%, and its authors stress that it does not predict the time of death. For a single person, the honest answer is a range of ages with probabilities, not a date.
What "predicting death" actually means in research
When a study says a model "predicts mortality", it almost always means one of two things. Either the model sorts people into higher and lower risk of dying within a fixed window (say, the next 4 or 10 years), or it estimates a survival curve: the chance of still being alive at each future age. Neither gives a date. Performance is usually reported as a C-index: the share of random pairs of people in which the model correctly says which one dies first. A C-index of 0.5 is a coin flip, 1.0 is perfect, and good medical risk scores tend to land between about 0.7 and 0.8.
Three kinds of AI that make headlines
| Model | What it used | What it predicted | How well |
|---|---|---|---|
| life2vec (Denmark, 2024) | Registry life events of about 6 million people: jobs, income, education, health contacts | Whether people aged 35–65 died in 2016–2020, from data up to 2015 | 78.8% correct on a test set balanced 50/50 between deaths and survivors |
| AIRE AI-ECG (UK/US/Brazil, 2024) | A single electrocardiogram; trained on 1.16 million ECGs from 189,539 patients | A personal survival curve and risk of death from any cause | C-index 0.775 in the development hospital, 0.638–0.773 in five outside groups |
| Epigenetic clocks, e.g. GrimAge (2019) | DNA methylation from a blood sample | Time to death, as a "biological age" in years | A strong predictor of lifespan in several cohorts, but at the level of groups |
life2vec treated each person's life as a sequence of events, the way a language model treats words (Savcisens et al., Nature Computational Science 2024). On the authors' own FAQ they explain that the widely quoted 78.8% accuracy comes from a test set of 100,000 people, half of whom died, so random guessing would score 50%. In real life more than 95% of people that age survive four years, which is why "accuracy" alone is misleading. The model and data are held securely at Statistics Denmark; there is no public demo, app or chatbot, and the authors warn that websites offering a "life2vec death calculator" are fraudulent (life2vec.dk).
AIRE, from Imperial College London and partners, reads a standard heart tracing and returns a survival curve, so it estimates time-to-death risk as well as risk (Sau et al., Lancet Digital Health 2024). It is a clinical research tool aimed at spotting people who need closer follow-up, not a consumer countdown.
Epigenetic clocks such as GrimAge estimate an age from chemical marks on DNA. People whose GrimAge runs ahead of their real age died sooner on average (Lu et al., Aging 2019). Commercial tests based on these clocks exist, but the result is a risk signal, not a forecast for one person.
What about death clock apps and chatbots?
Consumer "death clock" apps ask about your age, habits and family history and then show a date. Under the hood, they work like any life expectancy calculator: they start from population life tables and add or subtract years for your answers. The date is simply the average age of death for people like you, converted into a calendar day. We could not find independent, peer-reviewed validation showing that any such app predicts individual deaths better than standard tables plus well-known risk factors.
General AI chatbots have no special knowledge of your body. If you ask one how long you will live, it will reason from the same published life tables and studies, and it may get numbers wrong. Treat any single date, from an app or a chatbot, as a talking point rather than a prediction.
Why no model can give you a date
- The spread is wide even for identical people. For a 65-year-old man in the US today, US Social Security data give about a 64% chance of reaching 80, 24% of reaching 90 and about 1% of reaching 100. The middle half of that group dies somewhere between about 77 and 90. A model can shift that curve, but not collapse it to one day.
- Much of what kills people is unpredictable. Accidents, infections and new diseases are not in anyone's data today.
- The future changes the odds. Quitting smoking, treating blood pressure or a new drug all change risk after the prediction is made.
- Models are trained on the past. A model built on Danish registries from 2008–2015 describes that population in that period, not a person in another country in 2030.
What you can usefully estimate
The useful version of the question is not "when will I die?" but "what range should I plan for, and what moves it?". That is what our life expectancy calculator shows: an estimate built from population life tables, adjusted for your habits, with your chances of reaching 80, 85, 90 and 100 and the ages to plan for if you live longer than average. The method, including its limits, is written out on our methodology page. For a single question such as "will I see 2050?", try Will I Live to See It?.
Quick answers
Can AI tell me my death date?
No. Research models estimate risk or a survival curve for groups of people. Any app that shows a single date is turning an average into a day on the calendar.
How accurate is life2vec?
It was right 78.8% of the time on a test set balanced between people who died and people who survived within four years, against 50% for guessing. It does not predict when someone dies, and it is not available to the public.
Are death clock apps accurate?
They are as good as the life tables and risk factors behind them. We found no independent, peer-reviewed evidence that they predict individual deaths better than standard methods.
What is the most reliable way to estimate my life expectancy?
Start from your country's official life table for your age and sex, then adjust for major factors such as smoking, activity, weight and chronic disease, and read the result as a range rather than a single number.
Last reviewed: 29 October 2026. This article is educational and is not medical advice. If you are worried about your health, talk to a doctor.