

How to Make Conference Talk Slides with AI (TED-Style in 2026)
TED-style and conference talk slides have the opposite constraints of business decks: 1-3 words per slide maximum, one dominant visual, 40-80 slides for a 15-minute talk (many of which are near-blank transition slides). In 2026, AI tools are only 50% helpful here β they excel at generating the image-brief for each slide and suggesting a narrative arc, but they cannot replace the speaker's hand-picked hero image, the personal story, or the rhythmic build. This guide walks through the "slide-per-beat" model used by top TED speakers, the AI prompts that generate the image brief and narrative structure, and the three human-only craft moments in every great conference talk. If your goal is a 12-minute talk, expect AI to save you 4 hours of structure work but not replace the 8 hours of rehearsal and visual curation.
If you have ever sat through a conference talk where the speaker read bullet points off a 12-slide deck, you already know what a TED-style talk is not. A real conference keynote is choreographed β one idea per beat, one image per beat, and silence where most presenters would cram a sentence. Getting AI to help you build that kind of deck requires a completely different prompting approach than business slides.
The Slide-Per-Beat Model
Most business decks follow a "slide-per-section" model: three bullets about revenue, three bullets about churn, three bullets about roadmap. Conference talks do not work that way. The underlying unit is not a section; it is a beat β a single emotional or logical move the audience experiences.
A beat can be:
- A question you pose to the audience ("What if we are wrong about sleep?")
- A stark fact that reframes the problem ("42 percent.")
- A photo that makes the abstract concrete (a single fishing boat on an empty sea)
- A pause β literally a black slide β before the reveal
- A chart stripped to one line and one number
In the slide-per-beat model, your 15-minute talk is not 10 slides. It is 40-80 slides, and at least a third of them contain three or fewer words. Chris Anderson, who curates TED, has written publicly that the best speakers use slides the way filmmakers use cuts: not to display information but to control pacing.
This is why asking an AI "make me TED-style conference slides on climate change" produces something that looks correct but feels flat. The AI generates one slide per idea. A real talk generates five slides per idea, each revealing a little more.
The AI Prompts That Help
AI is excellent at three specific tasks in conference-talk prep: drafting the narrative arc, generating a shot-by-shot image brief, and producing alternative opening hooks so you can test which one lands. Here are the paste-ready prompts our users have converged on in 2026.
Prompt 1: The Image Brief Generator
You are a visual director for a TED-style conference talk. My topic is [TOPIC]. My core argument in one sentence is [ARGUMENT]. The talk is [N] minutes long. Generate a shot list of 50 slides where: - Each slide has ONE dominant visual idea - Text on slide is 0-3 words maximum (write "[SILENT]" if no text) - Every 4-6 slides is a "breath" slide (black, or single word) - Describe each visual as if briefing a photographer: subject, framing, lighting, mood Output as a numbered table: slide number | on-screen text | visual brief | the single beat this slide serves.
This prompt works because it forces the model away from its default habit of packing information onto slides. The "breath" rule alone fixes 80 percent of the problem.
Prompt 2: The Narrative Arc Builder
Act as a talk coach who has prepped speakers for TED, SXSW, and Web Summit. My topic is [TOPIC]. My audience is [AUDIENCE]. My one takeaway sentence is [TAKEAWAY]. Build a 7-beat narrative arc: 1. Opening hook (a question, fact, or image that earns 20 seconds of attention) 2. Personal stakes (why does the speaker care β specific moment) 3. The status-quo belief the audience holds 4. The break (evidence the belief is wrong or incomplete) 5. The new frame (the actual insight) 6. The demonstration (one concrete example that proves the frame) 7. The call (what the audience should do, think, or feel next) For each beat, give me: the beat's job, a sample line of spoken script (1-2 sentences), and the corresponding slide visual.
Prompt 3: Opening-Hook Alternatives
Give me 8 different opening 15-second hooks for a talk on [TOPIC]. Include at least one of each: a surprising statistic, a personal confession, a provocative question, a sensory description, a historical parallel, a misdirection, a single-word opener, and a silence-based opener (describe the stage action). For each, note the risk and the payoff.
Run this three times. The eighth or twelfth variation is usually where something unexpected comes out β that is the one worth rehearsing.
For a fuller workflow that turns written source material into a spoken-ready deck, see our guide on how researchers turn papers into conference presentations with AI.
The 3 Human-Only Craft Moments
Here is where AI stops being useful and your job starts. These three moments decide whether the talk lands or dies, and no model in 2026 can do them for you.
1. Hero Image Selection
Every great conference talk has two or three images that the audience remembers a year later. Brene Brown's arena. Simon Sinek's circle. Hans Rosling's bubble chart. These are not stock images, and they are not AI-generated renders. They are either real photographs with personal weight (often the speaker's own) or diagrams the speaker has iterated on for weeks.
AI image generation is tempting here and almost always wrong. Generated hero images have the uncanny flatness of consensus β the algorithm returns the average of a million reference images, and average is the enemy of memorable. Spend the hours finding or making the real thing.
2. The Personal Story
Every TED talk analysis study published since 2019 reaches the same conclusion: the speaker's personal anecdote is the highest-retention moment of the talk. Audiences forget statistics within minutes. They remember the story about the speaker's grandmother, the failed experiment, the phone call at 2 a.m.
AI can draft a generic story. It cannot pick the right one from your life, and it cannot tell you which detail (the color of the kitchen tile, the word your mother used) is the one that makes strangers lean forward. This is the irreducible human part.
3. The Rhythmic Build
Rhythm in a talk is the pattern of slide-change, silence, pace-up, pace-down. You cannot design it on a screen. You design it by standing up, speaking the talk out loud, and noting where you trip over your own words or where the slide changes feel mechanical. Then you cut.
Most first-draft decks have 20 percent too many slides. The rehearsal is where you find which ones to kill. Budget eight hours of out-loud rehearsal for a 12-minute talk β that ratio has held across decades of conference coaching.
Slide Count by Talk Length
Here is the rough slide count guideline our users report works well across conference formats. These numbers assume the slide-per-beat model, not the slide-per-section model.
| Talk length | Total slides | Text-heavy slides | Image-only slides | Near-blank slides |
|---|---|---|---|---|
| 5 min (Ignite / Lightning) | 20 | 2-3 | 12-14 | 3-5 |
| 10 min | 35-45 | 4-6 | 20-28 | 7-10 |
| 15 min (classic TED) | 50-70 | 6-8 | 30-40 | 12-18 |
| 18 min (full TED) | 60-80 | 8-10 | 35-45 | 15-20 |
| 30 min (keynote) | 80-110 | 12-15 | 50-65 | 20-30 |
| 45 min (extended keynote) | 100-140 | 18-22 | 65-85 | 25-35 |
If your auto-generated deck from a general-purpose AI tool has 12 slides for a 15-minute talk, you have a business deck, not a conference talk. Go back and expand.
Conference Talk Mistakes (AI Makes These)
AI tools default to business-deck conventions, which produces predictable failure modes when you try to generate conference slides. Watch for these.
- Bullet-point contamination. The model defaults to 3-5 bullets per slide because that is what 90 percent of its training data looks like. You have to explicitly forbid bullets in the prompt, repeatedly.
- Title-case everything. Conference slides rarely have titles. The AI will insist on "Introduction," "Background," "Conclusion" slide headers. Strip them all.
- No breath slides. Models are trained to maximize information density. They will not voluntarily give you a black slide or a single-word slide. You have to ask for them by name ("breath slides," "pause slides," "transition slides") and specify the ratio.
- Uniform pacing. AI-generated decks feel metronomic β every slide gets the same weight. Real talks have acceleration and deceleration. You fix this in rehearsal, not in the generator.
- Generic opening. "Hi, today I will talk about..." is the default. No conference speaker opens that way. Always run the opening-hook prompt separately and hand-pick.
- Over-designed visuals. Gradients, icon grids, 3D charts. Conference visuals are usually one photograph, one chart, one word. The AI wants to decorate; you want to strip.
- Missing the callback. Great talks plant an image or phrase in minute 2 and return to it in minute 14. AI rarely builds these callbacks. You have to design them by hand and then ask the model to weave them in.
For teams also producing a recorded version of the talk, the tooling stack is different from the live deck. See our roundup of the best AI tools for creating presentation videos in 2026 for the video-specific workflow.
Frequently Asked Questions
Can AI really generate a TED-quality talk?
No, and anyone claiming otherwise has not watched enough TED talks. AI can produce a competent first draft of the structure and a shot list of image briefs, which saves roughly 4 hours of scaffolding work on a 15-minute talk. The remaining craft β story selection, hero images, rhythm, delivery β still takes 8 to 15 hours of human work. The honest framing is: AI handles the scaffold, humans handle the soul.
How many words should a conference slide have?
Zero to three, most of the time. The ceiling for a text slide is about seven words, and you should have fewer than ten of those in a 15-minute talk. If a slide has a full sentence, the audience reads instead of listening β which means they disengage from the speaker. This is why TED's internal coaching guide treats the slide as a visual aid, not a teleprompter.
What presentation tool is best for TED-style decks?
Any tool that lets you build 50-plus slides quickly and change visuals without fighting templates. Keynote and Google Slides work. PowerPoint works if you disable its bullet-point defaults. AI-native tools like 2Slides work well when you prompt them explicitly for the slide-per-beat model rather than the default business format.
How long does it take to prepare a 15-minute conference talk?
The benchmark across coaching programs is 30 to 60 hours for a first-time speaker, 15 to 25 hours for an experienced one. AI shortens the structural work meaningfully β probably 4 to 6 hours saved β but does not reduce the rehearsal budget, which is where the talk actually becomes good.
Should I use AI-generated images as hero visuals?
Generally no, with two exceptions. If you need a conceptual diagram that cannot be photographed (data flow, imaginary creature, speculative future), AI imagery is fine. For everything else β especially anything the audience might have personal feeling about β use real photographs, ideally your own. Generated images have a distinctive flatness that audiences now recognize and quietly discount.
The Takeaway
Conference talks are the highest-craft deck format in existence, which is exactly why the AI-does-everything pitch breaks down here. Business decks reward information density, consistency, and speed β all things language models are superb at. Conference talks reward silence, specificity, and personal truth β all things language models approximate but cannot actually produce. Treating AI as a structural collaborator rather than a ghostwriter is the move.
The practical workflow for 2026 is this: use the three prompts above to generate a 50-slide scaffold and a seven-beat arc in 20 minutes. Then spend the next 15 hours doing the human work β picking the hero image that makes you wince a little when you see it, writing the personal anecdote you almost do not want to tell, and rehearsing out loud until the slide changes feel like breathing. That combination, done honestly, produces talks audiences remember. Anything less produces another forgettable slide-reader with nicer graphics.
Generate your slide-per-beat deck in 2 minutes β try 2Slides free.
About 2Slides
Create stunning AI-powered presentations in seconds. Transform your ideas into professional slides with 2slides AI Agent.
Try For Free