How Long Should a Keynote Speaker Speak?

So apparently, a keynote speaker should speak for 20 to 45 minutes at a technology conference, and 30 to 60 minutes at a corporate event. That number comes from the organizer of said events and not from anything about how attention actually works. It usually ends up being whatever slot you got with no real clear explanation of how it got there, so I figured it’s time for some science.

I checked every conference talk of mine ever recorded. There are ~75 of them. From these talks, the median length is 30 minutes, the shortest is a 4 minute lightning talk at dotJS, and the longest is 92 minutes. Not one of them is 18 minutes long, which is kind of weird because everyone says a keynote should be 18 minutes. But like, why 18? Where does that number even come from?

TL;DR

• Keynotes run 20 to 45 minutes at developer conferences. 30 minutes is the median across my 75 recorded talks, and 80% of them land between 20 and 45.
• The claim that audiences tune out after 10 minutes has been checked twice in peer review and didn’t hold up both times.
• Attention does drift more as a talk goes on, but is recoverable with the right delivery.
• TED’s 18 minute limit was a scheduling fix, not a neuroscience result.
• 30 minutes appears to be the global default for a conference talk and 5 minutes is the standard lightning slot, from London to Singapore to Melbourne.
• Even Apple’s WWDC keynote has fallen from 138 minutes in 2019 to 76 in 2026.
• Your slot is a ceiling, not a floor. Finishing 3 minutes early is a gift. Running over is an unforgivable stage sin.

What is a keynote speaker?

A keynote speaker is the person who gives the talk that sets the theme of an event, on the main stage, with no competing sessions scheduled against it. Everyone at the event is in that room. That’s what separates a keynote from every other slot on the schedule.

The word is borrowed from music. Before a group sings without instruments, somebody sounds the key note: the single note that tells everyone else which key they’re in. English has used “keynote” for that note since the 1770s and for a speech since the late 1800s when American political conventions started opening with one.

I love this nomenclature because it’s exactly what the job is: you’re putting the room in the same key. Everything after you then has something to land against. A track talk can assume the 40 people who chose it already care. A keynote gets to assume nothing except that everyone is in the room.

The metaphor kinda falls apart eventually though because a literal key note lasts just a second and gets out of the way, but a keynote lasts usually much longer so it has to also have substance.

Compared to other formats, here’s how a typical keynote looks:

Format	What it is	Typical length
Keynote	Sets the theme for the whole event. Main stage, nothing scheduled against it	20 to 45 min
Plenary	Any session the whole event attends. Every keynote is plenary, not every plenary is a keynote	30 to 60 min
Track talk	One topic in one track, running against other rooms	25 to 50 min
Lightning talk	One idea against a hard timer, usually inside a shared block	5 to 20 min
Workshop	Hands on, people bring laptops and build something	3 hours to 2 days

How long a keynote runs at different events

Let’s explore how long a keynote runs through two industries: professional speakers/corporate events and developer conferences since that’s my thing.

Event	Usual keynote slot	Who sets it
Developer or technology conference	20 to 45 min	The call for speakers
Corporate or leadership event	45 to 60 min	The events team or the bureau
TED and TED style events	5 to 18 min	A house rule
Internal all hands	15 to 30 min	Whoever owns the calendar

Speaker bureaus describe the corporate standard as 45 to 60 minutes, often delivered as 35 to 40 minutes of talk plus moderated questions inside a longer slot. Developer conferences run shorter. They run more tracks and turn rooms over faster.

So a TED speaker gets 18 minutes and a corporate keynoter gets an hour. Both of them will tell you the science backs their number. I went looking for that science and found that at the bottom of it was a paper from 1978 about note taking. It’s James Hartley and Ivor Davies reviewing other people’s note taking studies, so there’s no new data in it at all. The studies it leaned on did see students write less as a lecture went on, but in the last 10 minutes and not the first. One of them traced that to the lecturer wrapping up and simply saying less. Hartley himself even pointed out that note taking isn’t a reliable sign of attention in the first place.

Even the longest keynote in tech is shrinking

Apple publishes the running time of every WWDC keynote: the one keynote series anybody can measure over a decade.

Year	WWDC keynote runtime
2019	138 minutes
2021	107 minutes
2023	126 minutes
2024	104 minutes
2025	92 minutes
2026	76 minutes

Notice how the line wobbles: 2023 went back up by 22 minutes, probably overcorrecting for the pandemic. Across 7 years though, the most produced keynote in the industry with an unlimited budget and nothing scheduled against it has cut itself roughly in half. Wild.

The 10-minute attention span is a myth

You may have heard that audiences check out after 10 or 15 minutes. It’s in slide decks about slide decks! Some universities have cut lectures to 15 minutes because of it. Unfortunately, that’s all bullshit.

Karen Wilson and James Korn went looking for the evidence in 2007 and published what they found in Teaching of Psychology. They reviewed note-taking studies, direct observation of students, self-reports and physiological measures. Their conclusion was that the research behind the 10 to 15 minute claim provides little support for it and that several of the most frequently cited sources were secondary sources and personal observations that later writers cited as data.

Neil Bradbury checked again in 2016 in Advances in Physiology Education, under the title “Attention span during lectures: 8 seconds, 10 minutes, or more?” The finding was blunt: “the available primary data do not support the concept of a 10- to 15-min attention limit.”

The citation trail is the fun part: let’s watch a number turn into a fact.

Source	What it actually was	What it became
Johnstone and Percival, 1976	90 lectures observed, 87% by a single observer, with no stated definition of an attention lapse	“Attention breaks at 10 to 18 minutes”
Hartley and Davies, 1978	A critical review of note taking, reporting that the quantity of notes falls over a lecture	“Attention declines after 10 minutes”
Brain Rules, 2008	A trade book, popularizing the above	“The brain has a 10-minute rule”

It’s kinda wild: the source everyone cites for a rapid decline in attention barely discusses attention at all! A finding about how much people write down became a “finding” about whether they were listening! Then John Medina’s book Brain Rules turned that into a number conference organizers now schedule around.

Bradbury goes further: across the whole literature, the biggest source of variation in attention was the difference between one teacher and another, not the difference between formats.

Meaning, you are the secret sauce that holds attention and if you can’t, it is 100% undoubtedly a skill issue.

What attention actually does

Throwing out the 10-minute cliff doesn’t get you a room that stays rapt for an hour btw. Attention wanders. The research that actually measures the wandering is way more useful than the myth ever was: Diane Bunce, Elizabeth Flens and Kelly Neiles ran the cleanest measurement in 2010, in the Journal of Chemical Education. They gave chemistry students clickers and asked them to report attention lapses as they happened, during lectures and during other teaching approaches. Three things in there matter if you’re the one on stage.

1. The lapses students reported most often were shorter than 1 minute from silly things like glancing at a clock, not from leaving for 10 minutes. Some lapses were reported within the first 30 seconds.
2. Attention alternated between engaged and not engaged in cycles that got shorter as the session went on. (Attention was harder to pay the longer the presentation went).
3. Clicker questions and demonstrations produced significantly lower reported attention decline than straight lecture. The effect carried into the lecture that immediately followed.

Point 3 is kinda ridiculous: varying the shit you do in your talk buys you back attention later and also rekindles attention for the next speaker. Huge takeaway!

Evan Risko and colleagues in 2012 and Jonathan Farley and colleagues in 2013 both measured mind wandering across a lecture and watched it climb with time. Memory for the material also dropped right along with it, so it’s pretty solid that the longer you go, the worse it gets. Despite this, nobody has ever found a cliff at minute 10. Study after study lands back on the same variable: the speaker.

This also lines up with the segmenting effect in multimedia learning. Günter Rey and colleagues pooled 56 investigations and 88 pairwise comparisons in Educational Psychology Review in 2019 and found that breaking material into meaningful segments produces small to medium gains in both retention and transfer. Transfer is important because it measures whether people can actually use what they learned.

Segment the talk if it’s long, to retain attention and transfer.

Where TED’s 18 minutes actually came from

The 18-minute rule is the most quoted number in public speaking. It always gets introduced like it’s neuroscience even though it never was lol. By the time Chris Anderson took over TED in 2001, a 15-minute limit was already the norm. Anderson has said that in his first year he found “15 was often interpreted as 20 or 25”. The 18 came in as an attempt to be more precise. Techinch traced the history and 1843 magazine covered the same ground.

Anderson, though he himself never once claims it’s science, calls 18 minutes long enough to be taken seriously and short enough to hold attention. He likes the limit because it forces speakers to cut. In his words it “has a clarifying effect. It brings discipline” and honestly, he’s right about that part. A limit forces you to cut. Thing is, that works at literally any number you pick: the 18 has nothing to do with it.

What 75 of my own talks say

I’ve got the running time of every talk of mine that got recorded, straight from YouTube. They’re all listed on my speaking page. 5 of those recordings run past 90 minutes. 4 are full conference livestreams. The fifth one I’ll come back to. Drop all 5 and we’re left with 75.

Measure	Value
Talks measured	75
Median	30 minutes
Mean	29 minutes
Middle half	23 to 34 minutes
Shortest	4 minutes, a lightning talk at dotJS
Longest conference talk	92 minutes, at THAT Conference 2023
Between 20 and 45 minutes	60 of 75, or 80%
Between 25 and 35 minutes	34 of 75, or 45%
Exactly 18 minutes	0

I didn’t choose any of these timeframes btw, conferences did. Every one of those numbers is somebody’s scheduling decision, some of whom probably researched how long it should be. The Dots Connect, my entire story, at THAT Conference 2023 is the longest talk I’ve ever given at 92 minutes long and (take my word for it) people paid attention the whole time. It’s cheap/crude N(1) science, but in a small way it starts to prove that a talk can absolutely hold a room for an hour and a half. That one is a story though, told to people who chose to sit down for it.

How to choose a length in a talk proposal

Newer speakers usually meet this question for the first time inside a call for speakers, where the conference lists the formats it will accept. Real examples from 2026 calls:

Conference	Where	Session options
DevOpsDays	London, Singapore, Melbourne, Berlin, Chicago	30 minute talk, 5 minute Ignite talk with 20 auto-advancing slides
Jfokus	Stockholm	45 minute sessions, 15 minute lightning talks
PyTorch Conference Europe	Paris	25 minute sessions, 10 minute lightning talks
DevBcn	Barcelona	50 minute talks including questions, 20 minute lightning talks
FOSDEM devrooms	Brussels	10, 25 or 50 minutes, with 25 commonly set as 20 talk plus 5 questions
PyCon India	Bengaluru	40 minute featured slots, 15 minute opening and closing addresses
Momentum	Cincinnati	45 minute sessions, 5 minute lightning talks

Line them up and the numbers start repeating. 30 minutes for a main stage talk, 5 for a lightning slot. DevOpsDays runs that same 30 and 5 in London, Singapore, Melbourne, Berlin and Chicago. That’s an industry convention, not a local taste.

So here’s what I’d tell someone submitting for the first time:

1. Pick the shortest slot your idea genuinely fits. A tight 25 minute talk is accepted more often than a loose 50 minute one because a program committee cares about certainty.

2. Read whether questions are inside the slot. DevBcn says “including Q&A” in its call. FOSDEM devrooms commonly publish 25 minutes as 20 of talk plus 5 of questions. Some conferences have abolished questions entirely: Write the Docs tells speakers there will be no Q&A and asks them to stand at the front of the stage during the following break instead. DevOpsDays Singapore says plainly that it does not facilitate Q&A as part of the talks. The difference between these cases is 5 to 8 minutes of script. That’s the difference between finishing and getting cut off.

3. Submit the talk you can give at that length, not the talk you wish you had time for. The most common failure I see in new speakers is a 45 minute idea crushed into a 25 minute slot and delivered at speed. The good part gets cut because it happened to come last. :(

What makes a talk good at any length

These principles help deliver a quality talk at any length and I’ve found them to be valuable in my own practice.

1. Communicate one idea

If someone is asked in the hallway afterward what your talk was about, they get one sentence. Decide now what that sentence is. Cut everything that doesn’t serve it: that’s what a time limit is for.

2. Open and close on purpose

The two best-established findings in memory research are primacy (beginning) and recency (end): in a list, people recall the beginning and the end better than the middle. Bennet Murdock mapped the curve in 1962. Murray Glanzer and Anita Cunitz showed in 1966 that the two halves come apart under different conditions. That’s why the effect is still in every textbook.

We should be careful though not to overinterpret things because this is exactly the kind of leap that produced the 10-minute myth in the first place: those experiments used word lists in a lab, not talks. The one study I can find that tested recall by position inside a real lecture was Giles and colleagues in Medical Education in 1982. They found something close to the opposite: recall was worst for material in the first 15 minutes and best for material presented between minutes 15 and 30. The room needed time to warm up. This makes the case for good Emcees (MCs?).

It also makes the case to open strong because that’s when you earn the right to be listened to. Don’t bury your best material (the lede) in minute 2 on the strength of a word-list experiment. Don’t spend the opening on your name, your employer and your social handles, and don’t spend the ending on “that’s all I have, any questions?” because this is prime memory real estate man.

3. Tell a story

Greg Stephens, Lauren Silbert and Uri Hasson scanned speakers and listeners in 2010 and published the result in PNAS. While a person told a story, the listener’s brain activity tracked the speaker’s. The coupling disappeared when communication failed. The strength of the coupling correlated with how much of the story the listener could recall afterward. In some regions the listener’s brain ran ahead of the speaker, anticipating what came next. The greater that anticipation, the better the understanding.

Melanie Green and Timothy Brock had shown a behavioral version of this in 2000 in the Journal of Personality and Social Psychology: the more transported a person is into a narrative, the more their beliefs move toward it.

That’s a mechanical argument for narrative over a list of features. A story gives a listener something to predict. Prediction is what the scanner caught.

4. Change the mode every few minutes

This is the Bunce finding turned into stagecraft: move between explaining, showing, running code, asking the room something, and telling a story. Each change buys attention back, including for the stretch that follows it.

I write live code on stage for this reason. Nobody is really there for the code, but it’s like stimulating: they stay awake because they can see it might not work.

5. Put pictures on your slides

People usually deliver this one as a joke about PowerPoint, but the actual research is narrower: Richard Mayer, Julie Heiser and Steve Lonn tested this directly in 2001. People watched a narrated animation about lightning. One group saw no on-screen text, one saw a short summary line, and one saw the full text of the narration on screen. On the transfer test, the group with no on-screen text beat both text groups. Adding the same words a second time, through the eyes, made people understand less.

On what to put there instead, a study of 105 continuing medical education lectures scored 1,179 evaluations against the slides used. Speakers who used more image-based slides scored significantly higher. Text density, the number of words per slide, predicted nothing at all.

To be fair, redundancy isn’t always harmful and there are conditions where matching text and narration does help, including for people listening in a second language. What’s reliably bad is the specific thing speakers do: putting a paragraph on a slide and reading a slightly different version of it aloud. Now the room has to choose between you and the wall. Very very bad.

6. Rehearse against a clock

Slide count tells you nothing. One slide can hold you for 4 minutes. Words are a better guide. There are two numbers here that people mix up. Presentation coaching lands on 140 to 160 words per minute. Real conversation is much faster than that: Jiahong Yuan, Mark Liberman and Christopher Cieri measured 2,438 telephone conversations and got an average of 196 words per minute. So presenting well means deliberately talking slower than you normally talk. You have to rehearse that because when you’re on stage your adrenaline kicks in and you end up speaking like way faster than usual and it can become somewhat unpleasant to listen to.

Slot	Words at 140 to 160 wpm	What I would actually write
18 min	2,520 to 2,880	2,300
20 min	2,800 to 3,200	2,500
30 min	4,200 to 4,800	3,800
45 min	6,300 to 7,200	5,700

In your talk, it’s probably best to plan for fewer words than the arithmetic allows because pauses, demos, laughter, a joke that lands and a joke that doesn’t all consume time that no word count shows.

Finish early

After going to many, many conferences, I’ve seen that nearly every conference runs late by the afternoon. If you hand 3 minutes back to the organizer, they’ll love you because they usually have to fix the schedule anyway and nobody in an audience has ever once resented a talk that ended early, except maybe the MC who has to panic run to the stage and say “another round of applause wooooooo”.

Going over time is far worse since it eats the next speaker’s slot, the break or the questions. It ultimately stresses out the organizers, MCs, next speakers and really everyone involved. We’ve already seen how attention drifts more the longer the talk goes on as well: it’s just really not good for anyone.

Still, it’s so bad that some conferences plan for people to go over: PyCon India’s organizer handbook instructs volunteers to hold up cards at 15 minutes, 10 minutes, 5 minutes and time is up, and says those cards “are to be adhered”. Somebody has been asked to make those cards because speakers keep needing them. Rehearse so that you never see the 5.

What I still don’t know

I can tell you what the attention research doesn’t support. I can’t tell you the optimal length of a keynote. As far as I can find, nobody has run that experiment. There’s no trial of the same talk delivered at 18, 30 and 45 minutes that measures what an audience remembers a week later. I’d love to run or fund this experiment with someone who’s keen, reach out if you are.

The literature I’ve cited here is also mostly about classrooms and about learning. A keynote isn’t a lecture. It has different goals, a self-selected and well-caffeinated audience and a success condition closer to “people talked about it at lunch” than “people passed the test”.

While it would be cool to know, it’s probably best to treat the numbers in this post as descriptions of practice and the research as a defense against one specific bad idea: that there’s a biological clock in the room forcing you to be 18 minutes long or some sh

Takeaways

• Speak for the slot you were given. At a developer conference that’s usually 20 to 45 minutes, and at a corporate event 45 to 60. Confirm whether questions are inside it.
• Stop quoting the 10-minute attention span. It failed peer review twice and it started life as a study of note taking.
• 18 minutes is a scheduling convention with a good rationale, not a scientific result.
• Segment the talk and change modes often. That’s what actually holds a room, and the effect outlasts the change.
• Rehearse against a clock, plan for fewer words than the math allows, and finish early.

The length of your keynote is the least interesting decision you’ll make about it. Pick one idea, put the room in the same key, and get off the stage on time.

If you’re programming an event and would rather have this delivered than read, here’s what I speak about and how to book me. If you’d rather have your team taught than your audience addressed, I run workshops. If any of this was interesting or useful, please share it with someone who might benefit.

ok bye