CBT-I feels worse before it gets better: what the research says

For some people, the early weeks of CBT-I bring more daytime tiredness before sleep gets better. The American Academy of Sleep Medicine (AASM) describes these effects as concentrated early and short-lived, and it says improvement is typically gradual. ^[1] That same stretch is also when many people in trials stop attending, so it helps to know what to expect.

This article describes what studies found about the early weeks of CBT-I. It does not set out how a treatment is run, and it is not a reason to change anything you have been told by your own clinician.

What the research says

The guideline and the trials agree that the early weeks can be the hardest. The American Academy of Sleep Medicine lists daytime fatigue, sleepiness, irritability and attention problems as CBT-I's main downsides, says they cluster at the start when the behavioural parts begin, and describes improvement as gradual but lasting beyond treatment. ^[1] Two small lab studies of this stretch disagree. In one, 16 adults restricting their time in bed recorded about 70 to 90 minutes less sleep a night at first, with slower reactions and more sleepiness in weeks 1 to 3, all back to baseline by three months while diaries and insomnia scores improved. ^[2] In another 16 adults followed for two weeks, sleep time held steady and sleepiness, reaction time and simulated driving did not change significantly. ^[3] Four well-designed trials of the same approach found moderate-to-large gains in falling asleep and night waking, and a small gain in total sleep. ^[4] In plain terms, a rough early patch is common, usually temporary, and sits alongside real improvement.

The studies behind this

• AASM guideline (2021): the main downsides of CBT-I are daytime fatigue and sleepiness, irritability and attention problems. They are mostly limited to the early stage, when the behavioural parts begin, and typically ease by the end of treatment. The guideline adds that improvements are typically gradual but last beyond the end of treatment. ^[1]
• Kyle laboratory study (2014): 16 adults with insomnia were followed through the first weeks of the sleep-restricting part of treatment, used on its own. Recorded sleep fell by an average of about 70 to 90 minutes a night at the three lab checks, reaction-time lapses rose, and sleepiness scores were higher in weeks 1, 2 and 3. By three months, sleepiness and reaction times were back at baseline. Over the same period, diary-reported sleep and overall insomnia severity improved. ^[2]
• Whittall study (2018): 16 other adults were followed for two weeks on a similar approach. Their diary-reported sleep time held steady, and sleepiness, reaction time and driving-simulator scores did not change significantly. The authors say this applies only when sleep duration stays about the same, and call for studies with a control group. ^[3]
• Miller review (2014): four adequately designed trials of the sleep-restricting part used alone found moderate-to-large improvements in diary measures of falling asleep and night-time waking, and a small improvement in total sleep time. ^[4]

Why would it feel worse first?

We do not know, and researchers say so. A 2018 review looked for evidence on how the sleep-restricting part works and found 15 studies of generally poor quality. None was designed to test whether the proposed mechanisms explain the improvement. ^[5]

It is reasonable to read the early dip as a cost of changing long-held patterns, but that is our interpretation, not a proven mechanism.

What do people find hard?

Studies of people who stop early point to everyday life more than to the treatment itself.

• Online CBT-I trial (2025): 233 people were offered an eight-module program. 44.2% were classed as dropouts. The reasons most often given were distractions from daily life, content that did not seem useful, and difficulty resuming after a break. ^[7]
• First-week signal: in that trial, people who made more diary entries in week one were less likely to drop out. None of the baseline characteristics tested predicted dropout. ^[7]
• E-mail CBT-I in students (2023): 57% of 158 students did not complete all seven sessions. Among those who stopped, 57.8% did so before the fourth. ^[8]
• Group CBT-I (2008): in 528 patients, nothing measured at baseline predicted who would not finish. Among the 211 who did not finish, shorter baseline sleep and more depressive symptoms were linked to stopping early. ^[9]
• Completion by format (2025 meta-analysis): about 84% finished all sessions with individual therapy, about 71% with supported online programs and about 58% with fully automated online programs. ^[11] If you are weighing up a self-guided option, see Can you do CBT-I on your own?
• Patient accounts: in a nurse-delivered trial, the 16 patients interviewed described difficulty keeping going, negative attitudes towards treatment and low confidence. ^[6]

Does sticking with it matter?

Probably, though the evidence is observational. In that interview study, people who found the process positive showed improvements in sleep efficiency (a diary measure of how much of the time in bed is spent asleep), and those who struggled did not. If you want to see your own number, the free sleep efficiency calculator uses the same arithmetic as the app. That link between finding the process positive and improving sleep efficiency cannot show which came first. ^[6]

A 2013 review of 15 studies of adherence found both consistent patterns and inconsistencies across studies, which the authors say complicates any judgement about how much adherence drives success. ^[10] The AASM notes that people may get discouraged when results are not immediate, and advises clinicians to set realistic expectations at the start. ^[1]

Because the guideline describes improvement as gradual, a week with no visible change fits the pattern researchers describe. That is our reading of the guideline, not a rule about any one person's course. ^[1]

Who should be careful

The AASM guideline says the sleep-restricting part of CBT-I may be contraindicated for people in high-risk jobs (for example drivers or heavy-machinery operators), people prone to mania or hypomania, and people with poorly controlled seizure disorders, because of the sleep loss involved. ^[1] If any of these apply, talk with your doctor before starting.

On driving and machinery, the two small studies disagree: one found slower reactions during the early weeks, the other found no change on a driving simulator. ^[2][3] The guideline's caution about driving and machinery jobs exists because sleepiness at the wheel or at a machine is a safety risk for anyone, not something to push through. The authors of the Kyle study describe their findings as relevant to the safe delivery of treatment, and this is something to raise with your clinician rather than work around on your own. ^[1][2]

Where the evidence is thin

• The two sleepiness studies each had 16 participants and no comparison group, and they disagree. ^[2][3]
• Dropout figures come from different formats (online, e-mail, group), so one number does not fit every program. Two of the studies were secondary or exploratory analyses. ^[7][8]
• Predictors of dropout are inconsistent from one study to the next. ^[7][8][9]
• Few studies ask people directly what the early weeks are like. ^[6]

Where does this leave you?

The research suggests a harder early stretch is common, tends to ease, and is not a verdict on whether treatment will work for you. Small Hours is a six-week CBT-I program that works from your own sleep diary, and the diary is free. You can sign up at trysmallhours.com. If you are struggling, see also Sleep restriction making you more tired?

General education, not medical advice.