Guide

Can Apple Intelligence Summarize a PDF?

Partially, yes. Apple Intelligence includes Writing Tools, and Writing Tools can summarize any text you can select — including the text inside a PDF. If your document has a real text layer and your device supports Apple Intelligence, you can get a native summary in a handful of taps, without installing anything.

The short answer: partially, with Writing Tools

The boundary matters as much as the capability. Writing Tools summarizes selected text, not documents. That one design decision explains almost everything about where the feature works and where it doesn't, so this guide covers both halves honestly: the exact steps that work today, and the situations where the native route stops — scanned files, long documents, multiple files, and anything you need to verify against a source.

Short version: Writing Tools gives you a quick, one-shot digest of one selectable-text PDF at a time. It does not read scans, cite pages, answer questions about the document, or work across more than one file. For those jobs you need an app built on Apple's on-device foundation models — covered further down.

Step by step: summarizing a PDF natively with Writing Tools

Writing Tools appears anywhere the system lets you select text, and that includes Apple's own PDF viewers. Make sure Apple Intelligence is turned on in Settings first (Settings › Apple Intelligence & Siri).

On iPhone or iPad

1. Open the PDF in the Files app — or in Books, or as an attachment preview in Mail.
2. Touch and hold a word to start a selection, then tap Select All, or drag the handles across just the section you care about.
3. In the popup menu, tap Writing Tools. If you don't see it, tap the arrow to page through the menu options.
4. Tap Summary. Key Points, List, and Table sit alongside it if a different shape suits the material better.
5. Copy or share the result. Note that it is not saved into the PDF — it lives only in that panel until you copy it out.

On a Mac

1. Open the PDF in Preview.
2. Press Command-A to select all the text, or drag-select the passage you want.
3. Right-click the selection and choose Writing Tools › Summarize. The same commands live under the Edit menu.
4. The summary appears in a panel; copy it from there.

The same selection trick works in Safari's built-in PDF view and in Quick Look. The rule of thumb: if you can highlight the words with a cursor, Writing Tools can compress them.

Where the native route stops

These are structural limits of a selection-based tool, not bugs, and no settings toggle changes them.

Scanned PDFs have nothing to select

A scanned PDF is a stack of photographs of pages. There is no text layer, so Select All selects nothing, and Writing Tools never gets input. Depending on the app and platform, Live Text can sometimes recognize the page you are currently looking at — Preview on the Mac does this — but it works one page at a time and saves nothing back into the file. For a 40-page scanned contract, that is not a workflow.

Long documents compress into very small summaries

Writing Tools produces a one-shot digest: a compact paragraph or a short list of key points, no matter how much text you selected. There is no way to steer it ("focus on the termination clauses"), no way to ask for more depth on one section, and no follow-up. For a dense report, you get the altitude view and nothing else.

One file at a time, and no questions

Writing Tools only sees the selection in front of it. It cannot compare two PDFs, summarize a folder, or pull an answer from whichever of your documents happens to contain it. And it only rewrites — it cannot answer "what does clause 7 say about renewal?" even when the answer is inside the text you selected.

No citations — and no way to check

The summary arrives as unattributed prose. Nothing tells you which page a claim came from, so verifying it means re-reading the document — the job you were trying to avoid. For anything you would act on (a contract, a medical report, a filing), a summary you can't trace back to a page is a starting point, not an answer.

How third-party apps use Apple's on-device foundation models to go further

With iOS 26, iPadOS 26, and macOS 26, Apple opened the on-device foundation model behind Apple Intelligence to third-party developers through the Foundation Models framework. That means an app can now run the same class of on-device model that powers Writing Tools — but wrap it in the document machinery that Writing Tools deliberately leaves out.

I develop on this framework, so here is the honest shape of it: the model is only one piece. To do real document work, an app has to build everything around the model itself — extracting text from dozens of formats, running OCR on scanned pages, splitting long documents into indexed chunks, retrieving the right chunks for a question, and checking the model's answer against the source before showing it. The framework provides none of that; it provides the model. What separates document apps is the pipeline around it. Our guide to what on-device AI can actually do covers where those local models are genuinely strong and where they still fall short.

Summarizing and questioning PDFs with a document app — including scans, on-device

OpenIntelligence is my implementation of that pipeline: an open-source, 29-step retrieval engine built on Apple's foundation models. It addresses each of the four native limits directly:

• Scans: Vision OCR runs across every page at import, so a scanned PDF becomes as searchable and summarizable as a born-digital one.
• Long documents: files are indexed locally (SQLite full-text search plus Metal-accelerated retrieval), so you can summarize a section, then drill into it with follow-up questions instead of settling for one paragraph.
• Multiple files: questions run across your whole library — PDFs, Office and iWork documents, spreadsheets, notes, even recordings — not one selection.
• Verification: every answer carries tappable citations back to the exact passage, and when your documents don't support an answer, the app says so and abstains rather than guessing.

Everything — retrieval, verification, answering — runs on the device. It works in airplane mode, needs no account and no API keys, and its App Store privacy label is "Data Not Collected." The same engine also handles chatting with a PDF entirely offline.

Requirements, stated plainly: OpenIntelligence needs iOS, iPadOS, or macOS 26 or later on Apple Intelligence-capable hardware — an iPhone 15 Pro or later, or an iPad or Mac with an M1 chip or later. It is free to download on the App Store; a one-time Lifetime purchase or a Pro subscription unlocks the full feature set.

Requirements: which iPhones, iPads, and Macs can do any of this

If your device is older than these lines, neither route is available: both depend on the Neural Engine capacity that Apple Intelligence hardware provides. There is no software workaround on unsupported devices — an honest "no" is the correct answer there.

Related questions

Can Apple Intelligence OCR a scanned PDF?

Not as a document operation. Live Text — the system's built-in recognition — reads text in photos, screenshots, and the camera view, and in Preview on the Mac it can recognize the scanned page currently on screen. But there is no native way to batch-convert a multi-page scan into searchable text, and nothing is written back into the file. OpenIntelligence handles this case by running Vision OCR over every page during import.

Can Apple Intelligence translate a PDF?

Yes, natively — selection at a time. Select text in the PDF and choose Translate from the same popup menu that holds Writing Tools, or paste passages into the Translate app. For a document with a text layer, this works well; you just work in passages rather than whole files.

Can my iPhone read a PDF aloud?

Yes, natively, and no AI is required. Turn on Speak Selection or Speak Screen under Settings › Accessibility › Spoken Content, then select the PDF's text (or swipe down with two fingers from the top of the screen) to have it read aloud. This also works only when the PDF has a text layer — scans stay silent.

What about lecture recordings and voice memos?

Different input, same principle: the transcription and summarization can both happen on the device. That workflow — including where the native tools stop — is covered in our guide to transcribing and summarizing recordings without the cloud.