Almost nobody multitasks. What people call multitasking is rapid switching between tasks, and the switching is where the cost lives.
The exception is genuinely automatic activity. You can walk and talk, because walking demands almost nothing of the system that handles language. Put two tasks in front of the same limited resource, and one of them waits.
Where the cost actually comes from
The standard laboratory finding is the switch cost: people are slower and less accurate on the trial right after a change of task than on a trial that continues the previous one. Rubinstein, Meyer and Evans described this in terms of executive control, with two stages that have to run before the new task can proceed. Goal shifting, deciding to do the other thing. Rule activation, loading the rules that thing requires and unloading the previous ones.
Rule activation is the expensive part, and it is why switching between two similar tasks is often worse than switching between two different ones. Similar tasks compete for the same rules.
Two things follow, and they are frequently confused.
The cost is paid at the boundary, not spread through the work. This is the same event-boundary machinery behind the doorway effect and the resumption problem described in attention residue. What you lose is not throughput during the task, it is reconstruction time on re-entry.
The cost is often invisible from the inside. You do not experience four small reloads. You experience a normal afternoon that somehow produced very little, which is the illusion of competence wearing a different hat.
The part usually left out
The most repeated claim in this area is that heavy media multitaskers have measurably worse attention. It comes from Ophir, Nass and Wagner's 2009 PNAS paper, which found that people who habitually run several media streams at once were more susceptible to irrelevant information and showed larger switch costs. It is a good study and it launched a decade of research.
That decade was not kind to the strong version of the claim.
A 2017 meta-analysis by Wiradhany and Nieuwenstein pooled 39 performance-based tests across 12 studies. Ten showed heavy multitaskers as more distractible. Three showed the opposite. The remaining 26 showed no significant difference. The pooled effect was small, around d = 0.17, and the authors noted it might shrink further once small-study effects are accounted for.
So the honest summary: switch costs in a controlled task are robust and easy to reproduce. The claim that habitual media multitasking damages your general attention is weak, correlational, and has not replicated well. You should be suspicious of anyone quoting a number like "40% of your productive time" as though it were measured in your actual job. It was not.
None of which makes the practical advice wrong. It makes the reason for it different: you should stop switching not because it is rewiring your brain, but because reconstruction is expensive and you pay it every single time.
The reconstruction problem
Here is the version that survives the deflation.
Serious work runs out of working memory, which holds very little and holds it briefly. When you switch away, that contents does not persist. When you come back, you rebuild it. Where was I, what was I about to do, what did that variable mean.
Interruptions are not costly in proportion to their length. A ten second glance at a message can cost several minutes, because the ten seconds happened to demolish the state you had spent twenty minutes assembling.
This is why deep work is defined by uninterrupted duration rather than total hours, and why an afternoon of four one-hour blocks beats eight hours cut into fifteen-minute pieces.
What to do instead
Write the state down before you leave it. The cheapest fix by a wide margin. One line about where you were and what comes next, so re-entry is reading rather than remembering. This is the whole argument for external memory applied to a five minute interruption.
Give interruptions a destination. When something intrudes mid-task, the reflex is to handle it, because handling it takes eight seconds. Capture it instead and keep going. See the capture habit and the note inbox.
Batch by rule set, not by topic. Since rule activation is the cost, group work that shares rules. All the email at once. All the code review at once. Alternating between them is worse than either block.
Treat meetings as a switch, not a pause. The intention you formed in the meeting competes with the boundary at the end of it. Write the actions before you stand up: how to take meeting notes.
Stop counting hours and start counting blocks. Fragmentation is the variable that matters. Time blocking and the Pomodoro technique are both, underneath, just devices for making blocks contiguous.
The one honest caveat
Some work is genuinely interrupt-driven. Support, clinical shifts, parenting, on-call. Telling someone in those roles to protect two hour blocks is not advice, it is a description of a job they do not have.
For that shape of day the lever is not fewer switches, it is cheaper ones. Capture friction becomes the single most consequential property of your system, because you will cross that boundary forty times before lunch. Every one of them is a chance to lose something.
More in attention and focus, cognitive load theory, and context switching.