How to run an AI webinar that respects your audience's time. The preparation, roles, technology, and demo lessons AI Accelerator has learned running thousands of sessions.
Most people plan an AI webinar using a picture of how webinars used to run. Book a slot, share a link, talk over some slides, take a few questions, done. We see it differently, and after running thousands of these sessions we have the evidence to back it up.
A webinar is one of the most expensive things you can run, because the cost is not measured in your time. It is measured in everyone else's. Do the arithmetic. Put 100 people in a room and lose 15 minutes to a demo that will not load, and you have burned 25 hours of other people's working day. Turning up half-prepared does not just risk an awkward moment. It quietly undermines the whole reason you asked people to show up.
Everything here comes back to one idea: respect the time you are asking for, and prepare as though it matters. This is how AI Accelerator runs AI webinars, and what we have learned doing it, often the hard way.
A lot of people make assumptions about webinars based on how they used to work, and then they run an AI webinar the same way. It does not translate. AI changes what the audience wants to do with what you show them, and it changes where the risk sits.
When you demonstrate an AI tool, people do not just want to watch. They want to pause, try the same thing themselves, get it wrong, rewind, and try again. A pre-recorded, well-produced segment lets them do exactly that. A live, one-take demo does not, and it carries far more risk of falling over in front of everyone. The old instinct of "let's just do it live so it feels real" is precisely the instinct that undermines an AI webinar.
We have run thousands of these sessions. Clients regularly tell us they want more live demos, longer sessions, everyone following along step by step. We hear them out, and then we consistently prove that the approach in this guide produces a better result. We also see how often competitors get it wrong. The audience rarely knows why a session felt flat. They just quietly conclude that AI is not for them. That is the real cost of getting it wrong, and it is avoidable.
The webinar is the small, visible part of the work. The preparation is the part that decides whether it lands, and it is the part you must not rush. Before anyone joins the call, your content should be finished, not half-built, and your presenter should have done a full dry run end to end.
A dry run is not a read-through. It is the presenter working through the material out loud, in the order the audience will see it, with the actual technology in front of them. This is where you find the video that will not play, the link that has expired, and the section that runs eight minutes long. Rehearsal conditions rarely match the real thing, so rehearse until the mechanics are automatic and your attention is free for the people, not the buttons.
The single most useful thing we have learned: delivering a webinar and administering one are two different jobs, and one person cannot do both well once you get past a small group.
Presenting to 100 people while also watching the chat, spotting dropped audio, launching polls, managing the Q&A queue and keeping an eye on the clock is not multitasking. It is failing at several things at once. One person presents. One person produces, carrying the chat, the technical hiccups, the timings, and the person quietly typing "I can't hear anything". The producer is also your safety net: they surface good questions and step in when a live connection drops so the session never goes silent.
Wherever you can, hold a setup day before the session. Not an hour before. A day. Test the links, check everyone's configuration, iron out the software updates that always land at the worst moment, and confirm the platform behaves as you expect. Platforms update themselves and permissions change; the version you rehearsed on may not be the version you present on.
On the day, both presenter and producer should log in at least 15 minutes early. Check the link, the feeds, the cameras, the audio, and screen sharing. Fifteen quiet minutes buys you the confidence to begin on time and calmly.
Do not assume the machine in front of you can cope. The people who volunteer to run webinars are often on kit that was never meant for the job. A proper setup has a few non-negotiables:
A dedicated microphone, ideally a lapel or boom mic, not the built-in one. Sound is what people forgive least. A dedicated camera gives a sharper, more professional image than the laptop's own.
At least one extra screen, ideally two. One shows what the audience sees and the interaction around it; the other is reserved for the full-screen presentation. This lets the presenter know exactly what is going where, with no fumbling between windows mid-sentence.
Pre-load your live URLs. If you are opening any live pages during the session, fire them up in the browser before you start. Navigating somewhere fresh in front of an audience invites a slow load, a login prompt, or a dropped connection at the worst moment.
Go light on your slides. At AI Accelerator we are moving away from Keynote, PowerPoint and Google Slides towards HTML. It is lighter, more robust, quicker to render, and plays far better with the AI tools we build in. A presentation becomes a few kilobytes to share rather than a few gigabytes.
Audio is notoriously difficult to share cleanly over Zoom, Teams or Google Meet. If you are demonstrating a voice feature, or a video has its own soundtrack, getting the platform to share that audio reliably is fiddly. Where you can, avoid it. It usually causes more problems than it solves.
There is a better way to show something like voice mode. Screen capture it, then use production software to add subtitles to the recording. Colour the subtitles differently, or position them in different places on screen, to show who is speaking: one style for the AI, another for the person. The whole conversation then plays out visually on everyone's screen, and people can follow exactly what is happening without needing to hear a word of the audio.
One setting worth getting right: encourage everyone who is not speaking to mute, but encourage emoji reactions and visual cues freely. That feedback is how the room talks back to you when it cannot talk.
This is where we differ most sharply from conventional advice, and it matters especially for AI demos. Be very careful with live demos. From the audience's point of view there is almost no difference between you pressing the button and a recording of you pressing it. What differs is risk. The moment you go live, far more can go wrong, and if it does, in front of 200 people, you already know the maths.
So capture the interaction as a screen recording and play the video instead. Nobody loses anything, and you lose the single biggest source of on-the-day disaster. This is not cutting corners: live demos fail because the cognitive load of narrating, clicking, watching the room and troubleshooting at once is enormous, and rehearsal conditions never match the live environment. A recorded demo is polished, flows without bugs, and gives every viewer the same clean experience.
Keep clips to ~20 seconds. Anything longer and you sit waiting for it to catch up while the audience grows impatient. Cut whatever you record into short sections so the session keeps moving.
Don't fear a longer hero video (5–10 min) when the content earns it. Play it, and answer questions live in the chat while it runs. People get instant answers without interrupting the flow, and can rewind later. This works especially well for AI, where people want to pause, try it, then play on.
If you do want people to try something live, keep it to one simple, open-ended action. Not "press this, then press that, then press the other". Give them a single thing to do and room to explore it: "take a picture of this and drop it into your AI", or "put this one prompt into your AI and see what it does". One instruction, then space to play.
The moment you ask a live audience to follow a multi-stage workflow with you, it breaks down. Someone lands on a different screen, someone else is a step behind, and people start to feel like they are falling behind and disengage. Anything multi-stage belongs in a video they watch in their own time, where they can pause, rewind, and go at their own pace.
On the call, show them where that video lives and what the finished workflow looks like, but do not march the whole room through it together. It simply does not work.
A passive audience drifts; an audience that has to do something stays. Open with an interaction. Under 15 people, ask everyone to introduce themselves and say where they are on their AI journey. Over 15, that no longer scales, so switch to a poll or a question everyone answers. The mechanism matters less than forcing a small act of engagement in the first few minutes.
Keep it going. Industry benchmarks suggest polls are the most effective engagement tool, and that a fresh poll or question every 15 to 30 minutes acts as a reset button for attention. The same benchmarks find attendees watch only around 42 minutes of a 60-minute webinar, and that most people say 30 to 45 minutes is ideal. Attention is finite. Spend it well and top it up often.
We generally hold Q&A for the back end, so questions do not fragment the delivery. But this is where two people pays off again: one can answer questions in the chat while the other keeps delivering. Questions get handled in the moment, the flow is never broken, and nobody feels sent round in circles waiting for an answer.
Resist the urge to give everything away. A good AI webinar shows people there is depth behind what they are seeing and leaves them wanting to find it. If you have ten examples, show two or three. Make it clear the others exist, and let those who want to go deeper do so on their own terms. Showing all ten does not look thorough. It exhausts the audience and removes any reason to explore. Depth is more compelling as a promise than as a firehose.
Fill roughly 75% of the time you have booked, and no more. If you have an hour, build for 45 minutes and leave the rest as headroom.
People love getting 15 minutes back. They do not love being two minutes late for their next meeting. An early finish feels like a gift; an overrun feels like a theft, and it is the last thing people remember about your session.
Building in that buffer also gives you somewhere to absorb the small delays that always creep in, such as a longer Q&A, a technical hiccup, or a good question worth following, without blowing the clock. Plan short, and treat finishing on time as finishing early.
Something will go wrong, and usually it is at the attendee's end. They cannot see the screen, their mic will not work, their speakers are silent, and no one on your side can fix it for them. This is the producer's moment. While the presenter keeps going, the producer quietly helps that person; often the fix is as simple as logging out and back in.
If it is your connection that stumbles, do not panic and do not apologise your way through it. Take your time, steady yourself, and move on. Composure reads as competence. Flustered scrambling reads as an amateur.
Where you present matters more than people think. Wherever possible, run webinars from a professional studio environment rather than from home, keep background noise to an absolute minimum, and book the space so you are not interrupted. There is a lot of competition running sessions from a spare bedroom; it might feel authentic, but it rarely looks the part and it undercuts the professionalism of everything else you do.
On scripting: some things work well scripted, especially in the presenter's own voice. But if the presenter genuinely knows the material, they will not need a script and it will come across better for it. If you use an autocue, test it and set it up beforehand, and always have a fallback for when it fails. Finally, record and transcribe the session wherever you can. It gives you a reusable asset, a reference for attendees, and raw material for whatever comes next.
At AI Accelerator we charge a premium for webinars, and we are comfortable saying why. We give people a high-quality experience: a professional studio, proper pre-production, good lighting and audio, and actual practitioners on camera rather than trainers reading someone else's material. That is not the cheapest way to run an AI webinar, and it is not the right choice for every situation. But it is the reason our sessions consistently draw hundreds, and sometimes thousands, of attendees who stay to the end.
If you are choosing an organisation to run your AI webinar, do your homework. Ask how they actually run it. Where do they present from? Who is on camera, a practitioner or a hired trainer? How do they handle demos, audio, and questions? The answers tell you far more about the experience your audience will get than the price does. And if you are running the webinar yourself, take note of everything in this guide, and do not rush it.
"Can you just jump on tomorrow and show us some AI stuff?" is a completely different proposition from a webinar built around content, exposition, and polished delivery. The casual version has its place. The mistake is not being clear about which one people are getting. Tell your audience what the experience will be, rather than surprising them with something half-baked. Set the expectation, meet it, and keep the quality high, and you will consistently run AI webinars that fill up and stay full.
None of this is complicated. It is disciplined. Prepare properly and do not rush it. Split the presenter and producer roles. Test the day before and log in early. Get the kit right. Favour recorded video over live gambles, in short clips. Keep any live hands-on task to one simple, open-ended action, and leave multi-stage workflows to video. Make people participate, and leave them wanting more. Fill 75% of the slot and give people their time back. Respect the time you are asking for, and the AI webinar will respect you back.
How long should an AI webinar be?
Aim to fill about 75% of the slot you have booked. Attention benchmarks show attendees watch only around 42 minutes of a 60-minute session and most prefer 30 to 45 minutes, so building for 45 minutes in an hour keeps energy high and gives people time back.
Should you run live AI demos or use pre-recorded video?
Favour pre-recorded video for AI demos. The audience cannot tell the difference between a live click and a recording, but live demos carry far more risk of failing, and a clean recorded clip lets people pause, try it themselves, and rewind. Keep clips to around 20 seconds each.
How many people do you need to run an AI webinar?
At least two. One person presents while a second person produces, handling the chat, polls, timings, and any attendee whose audio or screen is not working. Beyond a small group, one person cannot deliver and administer at the same time.
How do you keep an AI webinar audience engaged?
Open with an interaction, use a poll or question roughly every 15 to 30 minutes, keep any hands-on task to a single open-ended action, and leave people wanting more by showing two or three examples rather than all ten.
Why do AI webinars need more preparation than normal webinars?
Because AI demos are fragile and audiences want to try things themselves. A dry run, pre-recorded demos, pre-loaded links, and a proper studio setup remove the moments most likely to fail live, which protects both the experience and your audience's time.