Every article on this says the relationship is bidirectional and then stops, which leaves you exactly where you started. The evidence supports a more useful answer.
The direction people miss
That anxiety disturbs sleep is obvious. The reverse is better evidenced than most people realise.
A meta-analysis of twenty-one longitudinal studies found that people with insomnia at baseline had roughly 2.6 times the odds of developing depression later (95% CI 1.98 to 3.42) compared with people sleeping normally.
For anxiety specifically, a Norwegian cohort of over 25,000 people followed for eleven years found that insomnia at the first survey was associated with an odds ratio of 3.4 (95% CI 3.1 to 3.8) for developing an anxiety disorder by the second. For people with insomnia at both timepoints it was 4.9.
These are associations, not proof of causation. But the temporal order is the point: the insomnia came first, and the anxiety followed.
Which means treating insomnia purely as a symptom - something that will clear up once the real problem is dealt with - is not well supported by the data.
Why "fix the anxiety first" often stalls
It is the intuitive order and there is a specific reason it disappoints.
Insomnia may start with anxiety and does not stay there. Within weeks it acquires its own maintaining machinery: extra time in bed to compensate, a bed that now reliably produces alertness, a body clock loosened by a variable rising time, and the effortful trying that makes sleep less likely, not more.
None of that machinery is anxiety. Treating the anxiety does not dismantle it, which is why people finish a course of therapy feeling better in themselves and still lying awake at 2am - and conclude, wrongly, that the insomnia must be permanent.
It is not permanent. It has just become its own problem and needs its own intervention.
What happens when sleep is treated directly
Several large randomized trials have tested exactly this, and the results are consistent enough to state plainly - and modest enough that overstating them would be dishonest.
The largest was a trial of digital CBT-I in over 3,700 university students. At ten weeks the effect on insomnia was large. Depression and anxiety symptoms also improved - with small effect sizes. Real, measured, and much smaller than the sleep effect.
A second trial of over 1,700 people compared digital CBT-I with sleep hygiene education and looked at functioning and wellbeing rather than sleep. Both improved, and the analysis found that improvement in insomnia mediated between about 45% and 84% of those gains - meaning the sleep improvement was the route through which the rest happened.
Pooling across twenty-two randomized trials of digital CBT-I, anxiety symptoms improved with a standardised mean difference of -0.29 (95% CI -0.40 to -0.19) and depression symptoms by -0.42 (95% CI -0.56 to -0.28). Small to moderate. Reliably different from zero.
One thing worth being straight about, because it is often reported carelessly: a trial that tested whether an online insomnia program could prevent episodes of major depression did not find that it did - nine cases against thirteen in the control arm, not a significant difference. Improving depressive symptoms and preventing a depressive episode are different claims, and only the first is supported.
