Summarize any long document or email thread in 60 seconds

One pattern, six ready-to-use lenses, and a reusable template. No terminal, no setup.

What you'll have when you're done

A working summary method you can run on any long document, email thread, contract, or meeting record. Not a flat recap you have to read anyway: a targeted answer to the specific question that matters for your decision. Most business documents can be handled in under 60 seconds from paste to usable output.

The report that has been sitting there

You have a 40-page market analysis from your strategy consultant. A 130-email procurement thread that got forwarded to you on Thursday afternoon. A vendor contract your lawyer sent back with track changes. A board deck from the company you are considering acquiring.

You know you need to read them. You have not, because reading them properly takes an hour you do not have.

The default move is to type "summarize this" and paste the document. The problem is that "summarize this" gets you a flat, accurate, completely inert recap. Everything technically correct. Nothing shaped around what you actually need to decide.

The fix is to specify your lens before you paste. A lens is not a format request ("use bullets"). It is an output-role request: you tell the model what you need to walk away knowing, and the whole summary reshapes around that goal. This is the same move as writing a strong brief in any prompt, applied specifically to the summarization task.

What you need first

• A Claude or ChatGPT account. If you are still setting up, your first 30 minutes with Claude covers it.
• A document, thread, or transcript you have been avoiding. The longer the better for your first test. Something you genuinely need to act on.
• A browser. Plain chat, nothing else. No plugin, no terminal.

That is it.

Step by step

The lens idea

Most people treat a summary like a compression job: take a long thing and make it shorter. The model is capable of a much more useful move: take a long thing and surface the specific subset that matters for your question.

A lens tells the model which subset to build around. "What are the risks?" gets you a different document than "What are the action items?" even if both are technically summaries of the same source. The model shapes the entire response around your role as the reader.

Six lenses, ready to use

Pick the one that matches what you need before you paste:

• "What are the decisions I need to make, and what does this document recommend?"
• "List every action item, who owns it, and any deadline."
• "What are the risks or red flags I should know about?"
• "Give me a one-paragraph TL;DR, then the three things that need my attention."
• "What changed since the last version, and what does that mean for me?"
• "What is this person or vendor actually asking me to do or approve?"

None of these are long. The value is not in the word count. The value is in naming the output role before the model starts reading.

By source type

Different documents have different structures. Here is a tailored ask for each:

Email or Slack thread: "Who wants what, what is unresolved, and what needs a reply from me before this moves forward?"

Report or deck: "What is the central argument, which numbers actually matter for my decision, and what is being asked of me?"

Contract: "List the obligations on each side, the key dates, the termination conditions, and any clause I should flag for a lawyer."

Meeting transcript: "What decisions were made, who owns each action item, and what questions were left open?"

These work on their own or stacked on top of one of the general lenses above. "Risks + contract" gets you the contract risks. "TL;DR + meeting transcript" gets you a three-bullet digest of the call.

The reusable pattern

Copy this once and use it every time:

[Paste the document, or upload the file]

You are helping me as a busy operator. [STATE YOUR LENS HERE].
Be specific: name the people, numbers, and dates when they appear.
Keep it under [X] bullets.

For uploading a PDF instead of pasting text, see how to get documents and images into AI. The pattern above works either way.

The go-deeper follow-up

After you get the summary, pick the single section that matters most and send a second prompt: "Go deeper on [that clause / that forecast / that open item]. What exactly does it say and what should I watch for?"

Broad lens first, then deep dive on the part that matters. This beats trying to get both in one overloaded prompt. Per Anthropic's and OpenAI's prompting guidance, one prompt doing five things trades off between them silently. You will not know which one it deprioritized until you are in the meeting.

How you'll know it's working

The output names specifics from your document. Not generic observations about contracts in general. Your actual party names, your actual deadline, the actual clause number. If the response sounds like it could have come from a model that never read the document, the lens is probably too vague or the paste did not go through.

You act on it. You forward the contract summary to your lawyer with the three flagged clauses already named. You reply to the thread with a clear answer because you now know what was unresolved. You walk into the meeting knowing which number matters. The summary reached the decision.

You catch something you would have skimmed past. A deadline buried in paragraph eleven. An obligation on your side you missed. A question in the thread nobody answered. The model reads carefully in ways you cannot when you are skimming under pressure.

When it breaks

Verification. Models can miss or misattribute items in a long document. For anything load-bearing (a contract commitment, a quoted figure, a deadline you are building a schedule around), confirm the detail against the original source before you act or forward it. Do not send the summary to your lawyer and say "Claude says the termination period is 30 days" without checking the clause yourself. Use the summary to know where to look; use the source to know what it says. How to verify AI output covers this step in full.

Privacy. Do not paste employee personal data, patient records, or attorney-client material into a consumer chat tool. The summary is not worth the exposure. If you are working with sensitive documents, check what is safe to share with AI tools first, or use your company's approved environment.

Very long documents. Most business documents (contracts, decks, reports, email threads) are well within what Claude handles in a single paste. For a genuinely massive file (hundreds of pages), upload it as a PDF rather than pasting the text, and ask about specific sections rather than relying on a single general summary. What a context window is and why it matters explains the underlying limit in plain language if you want the detail.

Flat output despite a lens. If the response still feels generic, add one concrete constraint. "Name the specific parties and dates every time they appear" forces grounding. "No hedging language" removes the model's instinct to hedge everything it is not fully certain about. One extra line is faster than re-explaining the whole lens.

Level up

Once the summarization habit is running, the natural next move is to stop re-explaining your business context every time you open a new conversation. Claude Projects let you upload your standard agreements, your org chart, your operating context, and your preferences once. Every conversation inside that project inherits it. You paste the document; the model already knows which risks matter to your business.

The six lenses above also work as standing instructions. Add your most-used lens to your project context: "When I share a contract, default to listing obligations on each side, key dates, and any clause I should flag for a lawyer." You do not have to retype it every time.

The document that has been sitting in your Downloads folder is a good first test. Paste it, pick the lens that matches why you have been avoiding it, and see what comes back.