How to Make Presentation Slides with AI: Tools, Workflow, and Prompts

You know what you want to present, but the slides are still blank. AI can draft the outline, the slide text, the speaker notes, and even the layout.

In short, choose your tool by the format you will finally deliver in.

• Delivering a PowerPoint file: Copilot built into PowerPoint, or @Presentations in the ChatGPT desktop app
• A prototype to show on the web: Gamma, which drafts slides from a topic
• Co-editing in Google Slides: Gemini (Google Slides)
• Content you developed in a conversation with Claude: Claude Slides (available on every plan, including Free)
• Create inside ChatGPT and edit together with your team: ChatGPT Slides (from October 16, 2026, rolling out to Pro, Business, and Enterprise users with Space access)

Whichever you choose, finish "what to say" and "how to show it" as separate stages. You can do both in the same app, but mistakes in the content and broken layouts are fixed in different ways. Below, we cover two approaches, tool comparisons, practical steps, prompts, and pitfalls.

1. What part of slide-making can AI automate?

"AI slides" actually spans three stages. Which one you want to hand off changes which tool you pick.

• ① Structure (outline): what to say, in what order, in how many slides. The skeleton of the talk.
• ② Script / body: the wording on each slide, the bullets, and the speaker notes (what you'll actually say).
• ③ Design: layout, color, fonts, diagrams/charts, images. The visual finish.

What AI produces is a draft of these three stages. Review them stage by stage, too: does the structure serve your goal, do the figures in the text match the source material, and is the design readable on the recipient's screen? How much revision you need depends on the use. A short internal update and a client proposal where every piece of evidence is checked against its source need different amounts of review, even at the same slide count.

2. Two approaches: all-in-one vs. division of labor

So how do you run those three stages? There are two broad routes. Think of them as differing in where you draw the line between "draft" and "finish" from the previous section.

Choose according to whether you need a quick prototype or want to work through the reasoning and evidence first. For a short update, start with a draft from approach A; for an important proposal, reviewing the script first can make the work easier. You can feed a script from B into an A-type tool, or move from conversation to slides within one app, as with Claude Slides. Section 4 shows a workflow using separate tools as one example.

3. Comparing the major tools

Here are the representative AI slide tools, lined up by strength and exit. What to decide by is whether you will ultimately deliver in PowerPoint, share on the web, or co-edit in Google Slides.

* Features, pricing, and support are vendor-published values as of 2026, not results from building the same deck in each tool. Pricing and availability change easily, so check each official source before adopting.

When choosing a tool, also look at whether you can keep your own copies of the script and deliverables. Save the original text, supporting sources, and exported files together so that moving to another tool does not mean rebuilding the explanation from scratch. Consider both the editing experience and the formats you can take away afterward.

4. A practical workflow: a finished deck in 5 steps

From conversation to slides in Claude

Anthropic's official help describes Claude Slides as a way to build presentations from notes, reports, or conversations, edit slides directly, present within Claude, and export to PowerPoint or PDF. Decks can also go straight to Google Slides as editable files. On October 8, 2026 it came out of beta and became available on every plan, including Free. For Enterprise, it turns on by default on October 15, 2026, and until then an owner has to enable it. "Cloud code execution and file creation" must also be enabled under Capabilities (personal settings for Free, Pro, and Max; organization settings for Team and Enterprise).

On the same October 8, 2026, Claude Motion also joined in beta to bring motion to your slides (Team and Enterprise only). You can turn a quarterly report into a 30-second explainer video or add motion to the charts in your slides, and export the result as MP4. It does not use a video generation model. Instead, Claude writes code that moves text, charts, shapes, and images, so you can adjust the wording, numbers, and timing later. No AI-generated footage of people is included (Build live dashboards and animate explainers with Claude, Anthropic, October 8, 2026).

For how to build a Claude Motion, example requests, and which plans include it, see What Are Claude Dashboards and Motion?.

For example, after developing your content in a conversation, ask: "Turn this proposal into a ten-minute presentation for executives. Separate the conclusion, supporting evidence, and decisions I need them to make. Use only figures from this source material." Check the order of the points and compare the figures with the source material, then adjust the amount of information on each slide. Finally, open the actual file you will deliver and make sure fonts, line breaks, and charts have not broken. Creating everything in one app does not remove this final step.

For a proposal or procedure that people will read as a document, you could choose Claude Docs instead of slides. First decide whether you need something to present on screen or a document people can refer back to.

The following five steps cover a workflow where you finalize the structure and script before handing the design to another tool. If you stay in one app, you can still check structure, script, appearance, facts, and delivery in that order.

In this example, settle the conclusion and evidence in STEPs 1→2, then refine the presentation in STEP 3. That way, when a problem turns up after the design stage, you can tell whether it lies in the content or in the layout and fix it accordingly. If turning a point into a diagram shows that the explanation falls short, go back to STEP 2 and add to the script.

5. Copy-paste prompts

Here are prompts that work for STEPs 1 and 2. The more you hand over "audience, goal, time, tone," the closer the output gets to usable.

① Build the outline (structure)

# Create a presentation structure
You are an experienced presentation designer. Build a slide structure under these conditions.

- Theme: [e.g. internal proposal for a new service]
- Audience: [e.g. 5 executives, not AI-savvy]
- Goal: [e.g. get budget approval]
- Time: [e.g. 10 minutes, about 10 slides]
- Tone: [e.g. concise, logical]

Output: per slide, a table of "title / the one thing to convey / key points (up to 3)."
At the end, add 3 likely objections from the audience, with answers.

② Flesh out one slide + speaker notes

# Flesh out a slide
Finish "Slide 3" above.
- Body: up to 3 bullets, each under ~10 words (to be shown, not read)
- If you include a number, mark it [needs source] so I can verify later
- Separately, write "speaker notes (60 seconds)" in spoken style, ready to read aloud

③ Reformat for a design tool

# Convert into a format to paste into Gamma etc.
Reformat this structure so it's easy to paste into an AI slide tool.
- Separate each slide as Markdown: "heading -> bullets"
- One message per slide. Minimal text
- Where something could be a diagram, note "diagram here (comparison/process/figures)"

By flagging "where something could be a diagram" as in ③, you give the design tool cues for replacing prose with diagrams. Content that has a shape, such as comparisons, steps, and trends, works better as a diagram than as bullets.

6. Six tips for slides that land

Use these checks to refine an AI draft, adapting them to slides you will present aloud or material people will read on their own.

7. Pitfalls (broken layouts, bloat, fabricated data)

For all the convenience, AI slides have common pitfalls. Knowing them in advance can reduce rework.

• Layouts break on .pptx export: web-based generators (like Gamma) tend to shift margins and fonts when exported to PowerPoint. If your final deliverable must be PowerPoint, choose a tool that creates a PowerPoint file from the start (Copilot, ChatGPT's @Presentations, etc.) — that's the safe move.
• The information density does not fit the purpose: if the slides are crowded, narrow the points, use a diagram, or move detail into speaker notes. When setting a deadline, set aside time for this editing and for checking the display, separately from the time generation takes.
• Plausible fabrication (hallucination): AI sometimes naturally invents stats, quotes, and sources. When a specific figure or "per X Corp's survey" appears, always go to the original. If someone points out an error on the day, the whole deck comes under suspicion.
• Handling confidential information: pasting unreleased figures or customer data into a cloud AI means sending it externally. Read your company's rules and each service's data handling policy first. If concerned, have it build only the structure with names and figures redacted.
• Tool dependence: service features and terms can change. Keep your script as text in your own hands, and build in a way that isn't bound to one specific tool.

Splitting the pre-delivery check into four parts: sources, information density, display, and sharing scope makes it harder to miss anything. Check figures and quotes one by one against the source material. Review the amount of text on each slide according to whether it is for presenting or for handing out. Check the display by opening the deck in the actual delivery format (.pptx, PDF, or a share link). Finally, confirm that the information may be included and who will receive it.

8. Recommendations by use case

When in doubt, draft just a few slides from material you already have, then open them in the format you will actually deliver. Checking that the recipient can read them properly first reduces the risk of reworking the entire deck later.

Summary

When making slides with AI, consider the final review and handoff as well as the convenience of the creation screen. Keep these points in mind.

• What AI can take on is the "draft" of structure, script, and design. The finish (cutting, verifying) is the human's job.
• Check content and appearance separately. You can divide the work between tools or use one app such as Claude Slides. Keep your own copy of the script.
• Choose by the exit. Delivering in PowerPoint → Copilot or ChatGPT's @Presentations; web-shared prototype → Gamma; Google co-editing → Gemini; content developed with Claude → Claude Slides.
• The more you give "audience, goal, time, tone" in the prompt, the higher the quality of the first draft.
• Review four things before delivery: original sources / information density / appearance in the delivery format / sharing scope.

Start with a few slides from material you already have, then present them aloud to check whether the conclusion comes across. Once the workflow includes checking against the source material and opening the actual delivery file, you can reuse it for your next deck.

FAQ