The two-minute rule is good advice, and you were right to install it. It was written about a paper in-tray. In the original, doing an item removes it. In email, doing an item often puts a new one in, because the person you answered answers back. That one difference is most of why an obvious rule gives you days where you handled forty things and none of the three that mattered.
- The two minutes price your tracking, and Allen says so directly. They mark where doing an item costs the same as storing it. So your threshold follows your own tracking cost, and for most people that now lands under two minutes.
- You learn a message is a two-minute message by reading it. By then the reading is spent, and every incentive in front of you argues for finishing.
- Sorting by size is a ranking. It’s the one ranking in your day that carries no information about what things are worth.
- A reply goes to a person who can reply back. In the largest study of email conversations, the single most common reply time was two minutes, so the likeliest result of your quick answer is another message shortly.
What the rule actually says
It comes from David Allen’s Getting Things Done, published in 2001. It lives inside the processing step. You pick up an item from the in-tray. You decide what the next action on it is. Then, in the book’s words, “if an action will take less than two minutes, it should be done at the moment it’s defined.”
The part that gets dropped in retelling is the reason. Allen is explicit that the number is a break-even. Two minutes is “more or less the point where it starts taking longer to store and track an item than to deal with it the first time it’s in your hands.” On the Getting Things Done podcast he puts the same thing in one line. A sub-two-minute action “would take you longer than two minutes to look at it again and review it and reflect on it later on. So this is a pure efficiency factor.”
Two things follow from that. Both are in the book, and both go missing in the advice.
The threshold moves. Allen calls two minutes “in fact just a guideline.” With a long window to process, you can push the cutoff to five or ten minutes. When you want to get to the bottom of everything quickly, you might “shorten the time to one minute, or even thirty seconds.”
The rule also has a scope. “You shouldn’t become a slave to spending your day doing two-minute actions,” Allen writes. “This rule should be applied primarily when you are engaging with new input; for example, processing your in-tray.”
So the rule as written is narrower than the rule as practiced. What most people installed is a general permission. If something looks quick, do it now, whenever you happen to see it. That wider version is the one that breaks. The book’s version stayed inside the processing pass.
Two minutes is a break-even, not a duration
The threshold is the point where doing costs the same as tracking. So the honest way to set it is to price your own tracking. That price has moved a long way since 2001.
In a paper system, deferring an item took real work. You wrote a line on a list and put the paper somewhere. Later you read the line, remembered what it meant, and found the paper again. Two minutes is a plausible break-even against that.
In a mail client, deferring a message you’ve already read is a keystroke. It goes away and comes back at a time you pick. The storing part of the cost has essentially gone to zero. Where a set-aside returns and what that costs is snoozed email. Turning a message into something you’ll act on later is just as cheap now, and that’s the subject of a task in five seconds.
Follow Allen’s logic and the conclusion sits awkwardly with the rule’s reputation. Cheaper tracking should lower the threshold. If deferring costs you a keystroke, the case for doing a ninety-second thing right now got weaker. The drift in practice runs the other way. Two minutes becomes “quick,” and “quick” becomes six.
One real cost stays on the tracking side, and it’s worth naming, because it’s the thing that should set your number. A deferred message has to be understood again when it comes back. That re-reading is the genuine expense, and it varies by message. A one-line request from a client you know is free to pick up again. A four-paragraph thread with two attachments and a question buried in the third paragraph is expensive to reconstruct. Defer that one and you pay the cost twice.
Which gives you a better tiebreaker than the stopwatch. On a borderline message, ask how much of what you now understand you’d have to rebuild on Thursday. The usual habit finishes the easy ones and defers the hard ones, which defers exactly the messages whose comprehension was expensive. This question turns you the other way.
The estimate is made after you have read it
Here’s the structural problem, and it sits under the rule rather than in your habits. You learn a message is a two-minute message by opening and understanding it. At that instant the reading is spent. What’s left is a comparison between finishing now and coming back later. Finishing now always wins it, because the cost you already paid is absent from both sides.
That would be fine if the estimate were good. People predicting how long their own work will take have been measured many times, and the error runs one way. In Exploring the Planning Fallacy, published in the Journal of Personality and Social Psychology in 1994, Roger Buehler, Dale Griffin and Michael Ross asked 37 students near the end of an honors thesis course to predict when they’d hand it in. They predicted 33.9 days on average and took 55.5. Fewer than a third finished inside their own best estimate.
A thesis runs on a different scale from an email, so carrying that ratio across would be dishonest. The authors’ shorter study is closer and smaller: 78 undergraduates predicting when they would finish a school project due within two weeks. Only 43.6 percent finished within the time they predicted. The ones who did finish had expected 6.0 days and reported 7.1. Modest, and in the same direction, and they were 83.5 percent confident on average.
The mechanism transfers cleanly, and the researchers found it by having people think aloud. “Participants appeared to construct narratives of how the future would unfold,” they write. “Participants rarely considered their own past experiences with similar tasks. When they did describe the past, they usually focused on previous occasions that justified their optimism.”
That’s an exact description of what happens when you glance at an email and think quick reply. You’re simulating the reply going well: you know the answer, you type four lines, you send. The last twenty times you thought that stay out of the simulation. Those were the times you opened the engagement letter to check a date, or realized halfway through that the honest answer needed a number you were missing, or wrote three versions because the client was annoyed.
The fix is plain and it works. Time it once. For one processing pass, start a stopwatch when you decide a message is a two-minute message, and stop it when the reply is sent. Most people find their two minutes is four or five. A few find one category where it’s fifteen. Once is enough. After that, your threshold comes from your own data rather than from a story you told yourself.
Answering is not the same as finishing
This break belongs to email alone, and it’s the expensive one.
An item in a paper in-tray, once handled, is gone. A reply is a message sent to a person who can reply to it. So in email the rule clears work part of the time and generates it the rest, and the generating is fast.
Somebody has measured this. Farshad Kooti, Kristina Lerman and three colleagues plotted how long replies take. Their data covered 1.3 million pairs of Yahoo Mail users, 2 million people exchanging 187 million messages. Nine replies in ten arrived inside a day. The single most common reply time in that whole distribution, the peak of the curve, was two minutes. Their paper, Evolution of Conversations in the Age of Email Overload, given at the 2015 World Wide Web conference, states it flatly: “the most likely reply time is just two minutes.” The halfway mark sits at 47 minutes.
Read that one precisely, because the mode is not the average. It doesn’t mean most replies come back in two minutes. The median is 47 minutes, and the slow tail drags the mean far higher. It means the single most common thing to happen after a message is sent is a reply about two minutes later. This is consumer mail, worldwide, from one provider, so treat it as the shape of email conversation rather than as a number for your firm. The shape is enough. The reply is the most likely next event, and it’s usually soon.
So the rule’s own arithmetic is incomplete. You compared two minutes of doing against the cost of tracking. The message that comes back sits outside that sum. So does reading it, deciding on it, and often a second short reply. A two-minute answer that opens a four-turn thread was the first installment of something larger.
The same study sizes a typical reply at a median of 43 words, with a mean of 153. Forty-three words is about a two-minute reply. So the rule covers the middle of your mailbox rather than its edges, which is why the total runs larger than anyone expects.
Speed is still worth giving. The delay side of the ledger belongs to batching email into set windows, the neighboring method and the neighboring failure. That’s where what it costs a client when your answer waits five hours, and what that does across a four-turn exchange, gets worked out. The point here is about volume rather than latency. A fast answer is a fine thing to give, and it counts as an answer rather than as an item removed.
One move genuinely closes the loop, and it’s a writing habit rather than a system. Answer the question that was asked and the obvious next one. State what happens next and who does it. The thread ends there. Doing that consistently returns more than any tool, and email habits worth keeping is where it’s worked through. Applied to the two-minute rule, it gives you a better test than the clock. Ask can I end this in two minutes, rather than can I answer this in two minutes.
Smallest first is a ranking
Strip the rule down and it’s a scheduling policy. Of everything in front of you, do the shortest items first, always.
That’s the policy you’d pick to reduce the number of things waiting, and it’s very good at that. It ranks on length alone. The length of a message tells you how long it is, and what it decides is a separate fact.
In practice the correlation runs backwards. The shortest replies in a professional mailbox are confirmations, acknowledgments, yes-that-works, forwarding a file, and finding a time. The messages that decide the quarter are the ones where a client is unhappy, a scope has moved, a number is wrong, or somebody needs a judgment from you. Those take longer than two minutes, every time. So a rule that always runs the short queue first spends your first and best half hour on the least consequential mail you received.
You feel this at 6:00 in the evening, as the particular flatness of a day with forty completed items and nothing finished. The work was real. It just happened in the wrong order.
Ranking by what a message is worth, and when it’s actually due, is a different job from ranking by size. It’s genuinely hard, and paying closer attention leaves it hard. Important versus urgent is that distinction in full.
The correction inside the method is small, and it matters. Run the two-minute pass at the end of the session rather than at the start. Do the two or three that decide the day first. Then spend whatever’s left on the short queue. Same rule, same items, and the day comes out the other way up.
The arithmetic on a full mailbox
The rule runs uncapped. Whatever fraction of arrivals qualifies as short, you’re committed to all of them, every day, ahead of everything else.
Do it on your own numbers. If ninety things arrive and thirty of them are genuinely two-minute answers, that’s an hour. It’s an hour spent inside the processing pass, ahead of the work that comes after it. If your measured two minutes is really four, it’s two hours. The cost of deciding, on a mailbox that size, is a separate charge on top. That arithmetic is set out in why the inbox zero method rarely lasts, the other method in this set that fails on volume.
This compounds rather than merely adds, because the short queue grows with your mail. The Yahoo study found that as load rises, people reply to a shrinking fraction: about 25 percent of a day’s arrivals at low load, dropping to under 5 percent at around 100 messages a day. That’s capacity running out, measured across millions of mailboxes. The set of things eligible for the rule scales with what other people send you. Your hours stay where they are.
Which is why the place the rule gets recommended most is the place it works worst. Catch-up advice for coming back from time away routinely includes a two-minute rule for quick actions, alongside triaging by sender and clearing newsletters. LinkedIn’s guidance on handling an inbox after a break sets it out in exactly that order, checked September 6, 2026. Against six hundred messages, “do all the short ones” costs several hours of the morning you needed most. By definition, it spends them on the least important mail in the pile. What a re-entry plan should actually hold is covered in the inbox zero piece linked above. The version where the catching up is done for you is a catch-up digest after time away.
Where the two minutes genuinely pay
The last four sections could read as an argument against the rule. The rule is right in a definable set of cases, and being able to name them is most of what makes it usable.
When the reply ends the thread. A confirmation, an acknowledgment, a yes with a date in it. The thread closes there, so the two minutes really were the whole cost.
When you’re the blocker. Somebody is waiting on your answer to move, and the answer is short. Here the two minutes buy someone else’s hours, which is the best return available in a mailbox.
When the message will be expensive to understand again. You’ve just read a complicated thread and you know what it needs. If the response is anywhere near the threshold, finish it. The re-reading is the real cost, and deferring makes you pay it twice.
When it’s a decision rather than a composition. Approving, declining, picking one of two options. These are quick, they’re frequently what somebody is waiting on, and they stay the size they started.
Inside a pass, never outside one. This is Allen’s own scope and the most valuable line in the original. The rule belongs to the processing session. Applied to a message that surfaces while you’re working on something else, “it will only take two minutes” is how a block of real work most often dies. The two minutes leave out the cost of getting back to where you were. There’s real measurement of what that return costs, and it lives in when to check email.
A version of the rule that holds up
Keep the rule. Change six things about how you run it.
- Time it once, honestly. One pass with a stopwatch, from decision to sent. Set your threshold from that number rather than from the book’s. If your real figure is four minutes, your two-minute rule has been a four-minute rule for years, and that explains where the mornings went.
- Set the threshold from your tracking cost. If deferring is one keystroke and the message is simple to re-read, the break-even is well under two minutes. If the thread is complex and you’ve just understood it, the break-even is much higher. That’s the same principle Allen states, applied to a system that arrived after his book.
- Make “can I end it” the test. Two minutes that closes a thread is a finished item. Two minutes that opens one is the first of four. If the short answer will bring a question back, write the longer answer that settles it, or defer the whole thing to when you can.
- Move the short queue to the end of the pass. Handle what decides the day first. The rest of the rule survives intact, and your best half hour stays yours.
- Cap it. Give the short queue a budget, ten or fifteen minutes, and stop when the budget is spent. Whatever’s left is short tomorrow too. An uncapped rule aimed at a queue other people fill has exactly one failure mode, and you’ve already met it.
- Exclude by category rather than by size. Some messages stay ineligible however short the answer looks: a price, a date you’re committing to, a scope change, a client who’s unhappy. Two minutes is enough time to type those answers and too little to get them right. The rework lands on your week rather than on the rule that caused it.
Every one of those is small. Together they turn a permission into a policy. That’s the difference between a rule that carries you through a heavy Tuesday and one that quietly eats it.
When the small things are handled before you sit down
Everything above assumes one arrangement. A mixed pile arrives. You open items to find out what they are. Then you personally do whichever ones turn out to be short. The rule is a reasonable local response to that arrangement. Point is an email app over the mailbox you already have, and it changes the arrangement.
The ordering problem goes first. Every thread is weighed on arrival, on two separate measures, how much it matters and how late it is. Each one carries a plain summary you can read before you open anything. So your feed leads with the few messages that decide the day. The short queue sits there looking like the short queue, rather than like whatever happened to land most recently. That’s rule four, done for you. How email triage works is the mechanism.
The break-even moves, in the direction Allen’s own reasoning points. Replies come back drafted in your voice. A good answer is one quick edit from sent. A message that would have cost two minutes of composing costs you a read and a send. Drafts wait for you by default. The words are prepared, and the decision to send stays with you until you raise the setting for that kind of work yourself. That’s replies in your voice, and the default is described in why a draft waits by default.
Deferral gets a clock rather than a folder, so the message you left at 9:40 comes back at the hour you picked. And the category that generates more short replies than any other goes quiet. Agreeing a meeting time is handled end to end, invitation included, rather than costing you four separate two-minute answers across two days. That’s scheduling without the email chain.
You choose, per kind of work, how far any of that goes on its own. The range runs from suggest-only, through review, to fully handled. Every kind starts at review. So the drafting and the preparing happen, and the sending waits on you until you raise that particular dial. A plain log records what was done, and undo covers Point’s own actions. The benefits page lists the lot.
Two limits, since this is a page about a method. What counts as a two-minute job in your practice stays your call, and so does the exclusion list in step six. And a fast reply stays a reply, whoever typed it and whatever software helped. The person on the other end can still write back. If you get thirty messages a day and the rule has served you for a year, it’s serving you, and this page leaves it alone.
Common questions
What is the two-minute rule for email?
It comes from David Allen’s Getting Things Done. While you’re processing your in-tray, if the next action on an item takes less than two minutes, do it right then rather than filing or scheduling it. Allen’s stated reason is efficiency. Around two minutes is where tracking an item starts to cost more than just handling it. In email, that means answering short messages during your processing pass instead of deferring them.
Is two minutes the right threshold?
It’s a starting point, and Allen says so in the book. He describes it as a guideline. Stretch it to five or ten minutes with a long processing window, or cut it to thirty seconds when you want to move fast. The better way to set it is from your own numbers. Time yourself for one pass, and price what deferring actually costs you now that setting a message aside is a keystroke rather than a paper note.
Why does the two-minute rule not work for me?
Usually one of four reasons, and every one of them is structural. Your real two minutes is four or five, because you make the estimate after reading the message and before you hit the parts beyond typing. The short queue is unbounded, so it grows with your mail while your morning stays the length it was. Sorting by size puts your least consequential messages first. And a reply often brings another message back, because the person on the other end can answer it.
Should I answer email as it arrives if it only takes two minutes?
That’s the version of the rule most likely to cost you, and the book asks for something narrower. Allen scopes the rule to processing new input, and warns against becoming a slave to two-minute actions. Applied to a message that surfaces while you’re in the middle of other work, the two minutes leave out what it costs to get back to what you were doing.
Does the two-minute rule work on a backlog?
Poorly, and it gets recommended exactly there. Against several hundred messages, doing every short one is hours of work chosen by length rather than importance. Those are the hours of your first morning back. Triage by who sent it and what it decides, handle those, and treat the short queue as something to sample rather than complete.
How is this different from the inbox zero problem?
They fail on different things. Inbox zero comes apart at its target, a count that can only ever be met or missed, and why the inbox zero method rarely lasts makes that case. The two-minute rule fails on estimation and ordering. Its threshold takes a stopwatch to measure, and it’s applied to a queue sorted by the one attribute that carries no information about value. The two often run together, which is how a pass meant to take twenty minutes takes ninety.
The short version
The two-minute rule is a break-even calculation wearing a stopwatch. Allen’s own reason for the number is that two minutes is where tracking starts to cost more than doing. So the correct threshold depends on what tracking costs you. In a modern mailbox, deferring is a keystroke, so that break-even has dropped. The drift in practice has taken it higher.
Three things then go wrong in email specifically. You make the estimate after reading the message, when the reading is already spent and every remaining incentive says finish it. Estimates of your own completion times run short in the studies that have measured them. The queue is sorted by length, the one attribute in a mailbox that stays silent about what a message is worth. And answering opens a thread as often as it ends one, because the most likely thing to follow a sent message is a reply, most commonly about two minutes later.
Keep the rule inside your processing pass. Run it at the end rather than at the start. Cap it. Exclude anything with a price, a date, or an unhappy client in it. And change the test from can I do this in two minutes to can I end this in two minutes. That version survives a heavy week, and it’s the one the book was describing all along.