Afternoon Caffeine and Sleep: How Late Is Too Late

What the evidence actually supports

One next step: Pick a caffeine cut-off at least six hours before your usual bedtime and keep it for two weeks before judging it.

The short version

Caffeine does not simply keep you awake while it is in your system. A single 400 mg dose taken six hours before bed measurably reduced sleep in a controlled trial, and pooled data from many studies show later intake pushes sleep onset later and shortens total sleep time. The catch is that people vary a lot in how fast they clear caffeine, so a single universal cut-off hour is not supported. The practical move is to pick a personal cut-off a fixed number of hours before your own bedtime, then judge it by how you sleep.

Six hours is not a free pass

The cleanest single study here gave 12 people 400 mg of caffeine (about two strong coffees) at bedtime, three hours before bed, or six hours before bed, in a double-blind crossover design. Even the six-hours-before dose reduced total sleep time compared with placebo.

That matters because six hours is exactly how most people describe their own habit: 'only after lunch, never in the evening.' The study says that habit can still cost you sleep, even when you feel fine falling asleep.

[C01]

What the pooled evidence adds

Two systematic reviews reach the same broad conclusion from much larger evidence bases: caffeine before sleep lengthens the time it takes to fall asleep and reduces how long you sleep. A 2023 review with meta-analysis found the effect is dose-dependent, and that timing shifts the size of it.

An older but still widely cited systematic review of both cohort studies and trials added the point that keeps coming up: the relationship is not one-size-fits-all.

[C02]

Why there is no universal cut-off hour

Caffeine's half-life differs substantially between people because of genetics, age, liver function, smoking status, pregnancy and some medications. Two people can drink the same coffee at the same time and have meaningfully different amounts still circulating at midnight.

That is why the honest advice is not 'no caffeine after 2pm'. It is: choose a buffer measured backwards from your own bedtime, then check it against how you actually sleep for a couple of weeks.

[C03]

What the evidence does not say

Nothing here proves that caffeine causes insomnia as a diagnosis, and none of it applies to children, pregnancy, or people managing a medical condition with stimulants. The reviews also do not establish that cutting caffeine will fix sleep problems that have another cause.

If you cut caffeine and sleep does not improve in a few weeks, the caffeine was probably not the main driver.

[C02] [C03]

Caffeine myth check

Four things people repeat about caffeine and sleep, and what the evidence actually supports.

Make it one decision

Cautions and limits

Read this before trying anything

• Caffeine affects people differently, and pregnancy, some medications and liver conditions change how it is cleared.
• This page does not cover caffeine withdrawal; cutting abruptly can cause headaches for a few days.
• If you use caffeine to stay awake for safety-critical work (driving, night shifts), do not change your routine based on this page alone.
• None of this replaces advice from your own clinician.

This page is not written for

• children and teenagers
• people who are pregnant or breastfeeding
• people with a diagnosed arrhythmia, panic disorder or insomnia disorder
• anyone whose clinician has given specific stimulant advice

Talk to a clinician if

• you have persistent trouble sleeping that is not explained by caffeine timing
• you feel palpitations, chest pain or significant anxiety after caffeine
• you cannot function without caffeine, or cutting it causes severe withdrawal
• sleep problems are affecting your work, mood or driving

Claims, sources and what they can carry

Every claim links to the source(s) behind it. Weights rank source types (Cochrane/systematic reviews 1.0, recognised guidelines 0.9, RCTs 0.8, observational or preprint 0.5, mechanistic 0.2). A weight is not certainty and is not GRADE.

Evidence ledger for this entry

Verification in this run was record/abstract level via NCBI E-utilities. Full text was not read and no retraction database was queried. Sources with a recorded verification gap are flagged in place.