How to Summarize a PDF on Your Mac Without Uploading It

On most recent Macs, the quickest private option is already built in: select the text in Preview, Control-click, choose Show Writing Tools, then Summary. Apple runs that on your Mac when it can, but says some requests go to its Private Cloud Compute servers. If you want the file to never leave your Mac at all, use an AI model that runs locally. Below is how each route works, where your text actually goes, and what to do when the PDF is a scan.

Option 1: Apple Intelligence Writing Tools (Built In, Free)

Writing Tools is part of Apple Intelligence. Here's what Apple says you need:¹

• A Mac with Apple silicon (M1 or later)
• macOS Sequoia 15.1 or later
• About 7 GB of free storage for the on-device models
• Apple Intelligence turned on in System Settings > Apple Intelligence & Siri
• Your Mac's language and Siri's language set to the same supported language

Apple Intelligence doesn't currently work on Macs bought in China mainland, or when your Apple Account region is set to China mainland.¹

How to summarize a PDF passage in Preview

1. Open the PDF in Preview.
2. Choose Tools > Text Selection, then drag over the passage you want summarized.²
3. Control-click the selected text and choose Show Writing Tools.
4. Choose Summary, Key Points, List, or Table.
5. Because a PDF is read-only, the result appears in the Writing Tools window instead of replacing your text. Click Copy to keep it.³

Apple says Writing Tools works in most places you write, including third-party apps and websites, so the same steps work on text in Mail, Safari, and elsewhere.³It summarizes what you select, so on a long document you decide which pages it sees.

Where your text goes

Apple's privacy page says that when possible, Apple Intelligence models run entirely on your device. When a task needs more computing power than the Mac can provide, Apple sends only the data relevant to the request to Private Cloud Compute, its own Apple silicon servers, and says that data is not stored or made accessible to Apple. Apple's own example is Writing Tools: your Mac may send an email to Private Cloud Compute to proofread it.⁴

You don't pick which happens. Apple Intelligence decides whether a request can be handled on the device. If you want to see what went to Apple's servers, go to System Settings > Privacy & Security, open Apple Intelligence Report, choose a time period, and click Export Activity.⁵

The ChatGPT extension is a different thing

Private Cloud Compute is Apple's. ChatGPT is not. If you set up the ChatGPT extension (System Settings > Apple Intelligence & Siri, then under Extensions, ChatGPT > Set Up), Siri and Writing Tools can hand some requests to OpenAI. When they do, your request and any attachments go to OpenAI, along with limited data such as your time zone, country, device type, and language.⁶

Apple says you're always asked before any files are sent to ChatGPT. If you use it without signing in, OpenAI must not use your request to train its models. If you sign in with a ChatGPT account, your ChatGPT account settings and OpenAI's policies apply.⁶

Option 2: Run a Local AI Model Yourself (Free, More Setup)

A local model runs on your Mac's own chip, so the document stays on your Mac. Two free apps make this fairly approachable:

• Ollama. Its desktop app for macOS lets you drag text files and PDFs into the chat. For large documents you can raise the context length in settings, which uses more memory.⁷ Our Ollama setup guide walks through installing it and choosing a first model.
• LM Studio. You can attach .pdf, .docx, or .txt files to a chat. If a document fits in the model's context, LM Studio adds all of it. If it's very long, LM Studio uses retrieval (often called RAG), which pulls out the passages that look relevant to your question. Its docs suggest mentioning the terms you expect to find in the document.⁸

The trade-offs are real. You download a model that takes several gigabytes, and a few settings matter. With retrieval, the model may only see excerpts, so a request to “summarize the whole thing” can quietly skip pages. Other chat apps built on local models vary, and some need an extra plugin or step before they can read a PDF at all.

Option 3: Web Summarizers and Cloud Chatbots

“Uploading” means a copy of your file goes to the company's servers, gets processed there, and is kept for as long as their policy and your settings allow. For plenty of documents that's fine. The big providers publish their controls, and it's worth knowing them:

• ChatGPT. You can turn off “Improve the model for everyone” under Settings > Data Controls.⁹ Temporary Chats aren't used to improve models and are deleted from OpenAI's systems after 30 days.¹⁰
• Claude. Anthropic's consumer plans (Free, Pro, Max) have a Model Improvement privacy setting.¹¹ Deleted conversations are removed from its back-end storage within 30 days; if you allow model improvement, de-identified data may be kept for up to 5 years.¹²
• Gemini. The Keep Activity setting controls whether your chats and files are saved. Even with it off, chats are kept with your account for 72 hours. A subset of chats is reviewed by people and kept for up to three years, and Google asks you not to enter confidential information you wouldn't want a reviewer to see.¹³
• Adobe Acrobat AI Assistant. Adobe says AI Assistant analyzes your document on Adobe servers and caches it for up to 12 hours, and that your content isn't used to train the models behind the feature.¹⁴
• Free “PDF summarizer” websites. Policies vary a lot. If a site doesn't say clearly how long it keeps files and who processes them, don't give it anything private.

Work, school, and business accounts often come with different terms. If the PDF belongs to a client or employer, check their rules and the provider's data controls for your plan before uploading.

Scanned PDFs: Why You Can't Select Anything

A PDF made from a scanner or a phone photo is often just pictures of pages, with no text layer underneath. Every tool above, including Writing Tools, needs actual text. If you choose Tools > Text Selection in Preview and can't highlight words, that's usually the reason.

Your Mac has two built-in ways to get the text out:

1. Embed Text in Preview. Open the PDF, choose File > Export, and select Embed Text. Preview recognizes the text and embeds it so it can be selected in any PDF viewer. The option only appears if the document hasn't had text recognized before.¹⁵
2. Live Text on an image. Open a photo or image of the page in Preview, hold the pointer over the text, drag to select it, then Control-click and choose Copy Text.¹⁶ Apple says Live Text uses on-device machine learning.¹⁷ It isn't available in every region or language.¹⁶

Which Should You Use?

• Not sensitive (a product manual, a public report): use whatever is convenient. A cloud chatbot is often the fastest way to get a decent summary of a long file.
• Sensitive, once in a while (your lease, one medical bill): Writing Tools is a reasonable choice if your Mac qualifies, knowing some requests may use Apple's Private Cloud Compute. Decline any ChatGPT prompt. If you need the text to stay on your Mac, use a local model.
• Sensitive, every week (client files, contracts at work): use a local model. Ollama or LM Studio if you don't mind some setup, or a dedicated app that handles the text extraction and questions for you. Check your employer's or clients' rules either way.

Tips for Long PDFs, Whatever Tool You Use

• Go section by section. Summarize the payment terms, then the cancellation section, then the fine print. One summary of 40 pages loses the details you care about.
• Ask targeted questions. “What are the deadlines?” “What am I agreeing to?” “What happens if I cancel?” “Which fees are mentioned?” These get sharper answers than “summarize this.”
• Ask where it says that. Requesting the exact sentence makes it quick to find the spot in the original.
• Check every figure, date, and name against the PDF. AI tools can misread a table or drop a “not.” The summary is a map, not the document.
• Treat it as preparation, not advice. For a contract or tax notice with real consequences, use the summary to work out what to ask a professional.

Where Hey Eduardo Fits

Hey Eduardo is a Mac menu-bar app we make for this kind of reading. Highlight text in any app and press ⌥⇧S for a summary: a one-sentence gist, a few short points, and the figures, dates, and names pulled out separately. ⌥⇧A asks a question about the highlighted text, ⌥⇧E rewrites dense wording in plain language, and ⌥⇧C lets you drag a box over a scanned page so Apple's on-device text recognition can read it. For a whole file, choose Ask About a Document… from the menu bar, pick the PDF, and go straight to your question. The “What should I watch out for?” preset lists what a document commits you to, what expires, and any silent-agreement clauses. It reports only what the text says; it isn't legal, tax, or financial advice.

Local mode is the default. The model runs on your Mac, nothing you capture is sent to us or any AI company, and it works offline after setup. Cloud mode is opt-in, uses your own Anthropic, OpenAI, or OpenRouter API key, and sends captured text (never images) to that provider. A Local/Cloud badge is always visible.

The honest limit: on the local model, each question reads a window of a few thousand words. When a document is longer, Eduardo tells you instead of answering about pages it never read, so for a long contract you'll still work section by section. If a PDF has no text layer, it says so and points you to ⌥⇧C. It needs an Apple Silicon Mac (M1 to M5) with macOS 15 or later, and it's a one-time purchase from $49 with a 14-day money-back guarantee. The guide shows every shortcut, and you can download it here.

Related: Our complete guide to on-device AI, the Ollama setup guide for Mac, and Hey Eduardo vs Apple Intelligence.