How to Install and Use Claude's Official PowerPoint Add-In: Step-by-Step Setup Guide

You Can Have Claude Building Slides in Under Ten Minutes

Most people spend more time reformatting a deck than they do thinking about what it should say. You’ve been there: adjusting font sizes, nudging text boxes, realigning bullet points that somehow drifted after a copy-paste. Claude’s official PowerPoint add-in doesn’t eliminate that work entirely, but it compresses the first draft from an hour to a few minutes — and the setup takes less time than a coffee break.

The install path is: PowerPoint → Home → Add-ins → search “Claude” → Add. Two models are available once you’re in: Opus 4.6 for complex tasks like full deck generation, and Sonnet 4.6 for quick edits and reformatting. A persistent instructions setting lets you encode your brand defaults — fonts, colors, speaker note format — so you don’t repeat yourself every session. That’s the core of the tool. Everything else is configuration and workflow.

This is a step-by-step setup guide. By the end you’ll have the add-in installed, the right model selected for your use case, and persistent instructions configured so Claude produces consistent output from the first prompt.

---

What You’re Actually Getting Here

Before you install anything, it helps to understand what this add-in is and isn’t.

It’s an official Anthropic product — not a third-party wrapper or a browser extension that scrapes your slides. It sits natively in PowerPoint’s sidebar and generates fully editable PowerPoint elements: real text boxes, real charts, real shapes. Not screenshots. Not images of slides.

That distinction matters because it means you can edit everything Claude produces. If a chart value is wrong, you click on it and type the correct number. If a headline is too long, you fix it directly. You’re not locked into regenerating the whole thing.

The add-in is still in beta as of this writing, which means you should expect occasional rough edges. But the core workflow — prompt, review, accept or reject — is stable enough for real use. If you want a deeper understanding of Claude’s capabilities and how the different model tiers compare, What is Claude and How to Use It for AI Agents covers the Opus/Sonnet distinction in detail — the same tradeoffs that apply in the PowerPoint add-in apply everywhere you’re choosing between capability and speed.

---

What You Need Before You Start

A qualifying Claude plan. The add-in requires a Claude Pro account ($20/month), or a Max, Team, or Enterprise plan. The free tier doesn’t have access. If you’re already on Pro for other Claude work, you’re covered.

Microsoft PowerPoint desktop or web. The add-in works on both. It does not work on iPad or Android — desktop and web only. If you’re on an older version of the desktop app, update it before you try to install; older builds sometimes can’t load Office Add-ins correctly.

A Microsoft account signed in to PowerPoint. The add-in installs through the Office Add-ins store, so your Microsoft account needs to be active in the app.

That’s the full prerequisites list. No API keys, no separate Anthropic account setup beyond your existing Claude subscription, no IT ticket required for personal use.

---

Installing the Add-In: Two Paths

Path 1: From Inside PowerPoint

1. Open PowerPoint (desktop or web).
2. Click Home in the ribbon.
3. Click Add-ins.
4. In the search field, type Claude.
5. The “Claude by Anthropic” add-in will appear. Click Add.
6. The Claude sidebar will open on the right side of the screen.
7. Click Login and authenticate with your Anthropic account credentials.

Now you have: The Claude sidebar active in PowerPoint, connected to your Claude account. You’ll see the model selector and a chat input at the bottom.

Path 2: Direct Link (If Path 1 Doesn’t Work)

If the add-in doesn’t appear in search — this happens occasionally with certain Microsoft account configurations — go directly to the Microsoft AppSource listing for “Claude by Anthropic with PowerPoint” and click Get it Now. Sign in with the same Microsoft account you use in PowerPoint. Restart PowerPoint. The add-in will appear in your sidebar.

Now you have: The same result as Path 1. Use whichever path works for your setup.

---

Choosing the Right Model

Once the sidebar is open, you’ll see a model selector near the top. Two options:

Opus 4.6 is the more capable model. Use it when you’re generating a full deck from scratch, restructuring a presentation’s narrative, or working from a complex source document like a multi-sheet Excel file or a dense PDF. It’s slower and uses more of your monthly usage allocation.

Sonnet 4.6 is faster and lighter. Use it for targeted edits: fixing a headline, reformatting a slide, adjusting copy length, translating text. If you’re doing iterative polish work — five small changes in a row — Sonnet 4.6 will feel noticeably more responsive and won’t burn through your credits as fast.

The practical rule: start a new deck with Opus 4.6, switch to Sonnet 4.6 for cleanup. You can change the model between prompts; you don’t have to commit to one for the whole session.

One thing to keep in mind: chat history does not persist between PowerPoint sessions. Every time you close and reopen PowerPoint, you start fresh. This means the model choice you made last session isn’t remembered — you’ll need to reselect it. If you’re thinking about how to manage token usage and model selection more deliberately across AI tools, the guide to saving tokens using Opus plan mode offers a useful mental model that transfers to the PowerPoint context.

---

Configuring Persistent Instructions (Do This Before Your First Prompt)

This is the step most people skip, and it’s the one that saves the most time over repeated use.

In the sidebar, click the Settings icon. You’ll find an Instructions field — a free-text box where you tell Claude how it should behave in every session. This is persistent: whatever you write here applies to every prompt you send, automatically, without you having to repeat it.

Useful things to put here:

• Always use [your brand font] for all text
• Always use [hex color] for headings and [hex color] for backgrounds
• Add speaker notes to every slide
• Keep bullet points to a maximum of three per slide
• Use a corporate, direct tone — no marketing language

If you work in a specific industry or have a house style, encode it here. Claude will follow these instructions on every generation without you mentioning them in the prompt.

Now you have: A Claude instance that knows your defaults. The first time you generate a deck, it will already be working toward your brand standards rather than making arbitrary choices.

This same principle — encode your intent precisely once, generate repeatedly — shows up across AI tooling. When you’re building applications rather than documents, Remy takes a comparable approach: you write an annotated markdown spec, and it compiles a complete full-stack TypeScript application (backend, database, auth, deployment) from that spec. The abstraction level shifts, but the underlying pattern is identical to persistent instructions: the spec is the source of truth, and the generated output is derived from it.

---

Loading Your Template (The Most Important Pro Tip)

Claude automatically reads your slide master layouts, fonts, and color schemes from whatever file is open. It achieves roughly 90% first-pass accuracy when a template is pre-loaded — meaning the generated slides will match your visual structure most of the time without additional prompting.

The pro tip: always load your template before you start prompting. Open your branded PowerPoint template file, then open the Claude sidebar, then start your session. Claude uses whatever is already loaded as its reference point. If you start with a blank file and then ask for a “corporate blue” deck, Claude will make something up. If you start with your actual template, Claude will follow it.

This single habit is the difference between output that needs heavy reformatting and output that’s close enough to use. For teams that want to take this further — enforcing consistent visual language across all AI-generated content, not just slides — the approach in How to Use Google Stitch’s Design.md File with Claude Code for Consistent UI is directly applicable: a single design reference file that any AI tool reads before generating output.

---

Your First Generation: A Practical Walkthrough

With the add-in installed, the model selected, persistent instructions set, and your template loaded, here’s how a typical first generation goes.

Type a prompt like: Build a four-slide deck about our Q1 sales performance. Corporate tone. One key metric per slide.

If you have Ask before edits enabled (recommended for new users — it’s in the bottom-left of the sidebar), Claude will show you what it’s about to do before making changes. Review it, then click Allow once to proceed. You can also click Allow always to skip this confirmation step in future, but hold off on that until you’ve seen a few generations and trust the output.

Claude will generate the slides. On a full deck generation with Opus 4.6, expect this to take a few minutes. Sonnet 4.6 on a single-slide edit is much faster.

When it’s done, you’ll see the slides in your deck. Everything is editable — click any text, any chart value, any shape and modify it directly. If Claude hallucinated a chart value (a known limitation in this beta), just click the number and type the correct one. The chart is a native PowerPoint element, not an image.

To target a specific slide, click on it in the slide panel before typing your prompt. The sidebar will automatically add Slide 3 selected to the prompt context, so Claude knows which slide you’re referring to.

---

When Things Go Wrong

The add-in doesn’t appear after install. Restart PowerPoint completely. If it still doesn’t appear, try the direct AppSource link (Path 2 above). Occasionally a full sign-out and sign-back-in to your Microsoft account resolves it.

Claude is editing the wrong slide. Make sure you’ve clicked on the target slide in the slide panel before prompting. The Slide X selected context indicator in the sidebar confirms which slide is active.

Chart values look wrong. This is a known beta limitation. Claude processes PowerPoint files to markdown internally, which means it can’t actually see your slides visually — it infers content from the file structure. Chart values are particularly prone to hallucination. The fix is always the same: click the chart element and correct the value manually. The chart is fully editable.

The file won’t load. PowerPoint files have a 30MB cap in this add-in. Large presentations with many high-resolution images will hit this limit. The workaround is to split the deck into sections, work on each section separately, then merge them afterward.

The output doesn’t match your template. You either didn’t load the template before prompting, or the template has complex custom layouts (multi-step chevron processes, custom master slides with unusual geometry) that Claude’s markdown-based processing can’t fully reconstruct. Simplify the template or add more specific instructions about colors and fonts in your persistent instructions.

The add-in isn’t available on your device. iPad and Android are not supported. Desktop (Windows or Mac) and PowerPoint web only.

---

Where to Take This Further

Once the basic setup is working, a few capabilities are worth exploring in order of usefulness.

PDF and Excel input. Click the + icon in the sidebar to attach files. You can upload a PDF report or an Excel workbook and ask Claude to turn it into a slide deck. This is the highest-leverage use case for most business users — you have data, you need slides, and the manual translation step is now optional.

Website URL input. The add-in has built-in web search. Paste a URL into the prompt and ask Claude to generate a pitch deck based on the site. It will pull the content and attempt to match the site’s color scheme. Useful for competitive research decks or client-facing materials where you want to mirror a brand’s visual language.

Translation. Translate this entire deck to French works as a single command. The add-in will update all text across all slides. Useful for teams operating in multiple languages.

Speaker notes at scale. Add speaker notes in English to each slide generates notes for the full deck in one pass. The notes are contextually aware — they’re based on the actual slide content, not generic filler.

Skills and connectors. The sidebar has a Skills section and a Connectors section for data pull integrations. If your organization uses external data sources you want to pull into presentations, this is where those connections live.

For teams building more complex AI workflows around content generation — where the presentation is one output among many — MindStudio offers a different layer: 200+ models, 1,000+ integrations, and a visual builder for chaining agents and workflows, so the slide generation step can be part of a larger automated pipeline rather than a standalone task.

The persistent instructions setting is where most of the long-term value lives. The more precisely you encode your defaults — fonts, colors, tone, structure — the less correction work you do on every generation. Treat it like a system prompt for your presentation workflow: write it once, refine it over a few sessions, and let it do the repetitive specification work for you.

If you’re thinking about how AI tools fit into larger content workflows, the skill-based content automation approach is a useful frame for thinking about where manual work can be systematically replaced. The same logic that makes persistent instructions valuable in the PowerPoint add-in — define the repeatable pattern once, execute it on demand — applies to content systems at scale.

The add-in is in beta. Some edges are rough. But the core loop — load template, set instructions, prompt, review, accept — is solid enough to replace a meaningful chunk of manual slide work today. Install it, configure your persistent instructions before your first real prompt, load your template, and use Opus 4.6 for generation and Sonnet 4.6 for edits. That’s the setup that produces consistent output from session one.