Phase 3 · Claude · Level 1 · Foundations
Guided walkthrough: interrogate a policy from start to finish
By the end, you'll be able to…
- Take one real document from paste to a checked, usable answer, following each step
- Interrogate a document with pointed questions and verify the answers against the source
- Turn what you learn into a short output you can share or keep
Why it matters
Reading about Claude's strengths is one thing; watching a real task done start to finish is another. This walkthrough takes the Fernway remote-working policy, summarises it, interrogates the details, catches where a confident answer could mislead, and ends with something you could actually send: the exact flow you'll reach for again and again.
What we're doing
Imagine you've just joined Fernway. Priya has sent round the hybrid-working policy and asked everyone to "read it and check your own arrangement fits." It's several pages. Rather than skim it and hope, you'll use Claude to understand it quickly, check the parts that affect you, and come away with a short note of what you actually need to do.
We'll follow the whole task in seven steps. Work at your own pace; there's no rush, and you can stop and pick it up later. Everything here uses the Fernway remote-working policy, so you can do it alongside the reading with safe sample material.
Step 1: Start a fresh chat
Open claude.ai and click New chat so you're on a clean, blank conversation. A fresh chat keeps this task tidy and separate from anything else. If it's a task you know you'll return to over days, this is also the moment you might start a Project instead, but for a single sitting, a plain new chat is right.
Step 2: Give Claude the document
Hand Claude the policy. Two easy options: attach the file with the paperclip or plus icon, or paste the text straight into the message box. Either way Claude can now read the whole thing; you don't need to break it into chunks.
Step 3: Ask for a structured summary
In the same message, tell Claude what you want and how to shape it. Label the document clearly so it knows what to work on.
You're helping a brand-new Fernway employee get up to speed. Read the policy below and give me: (1) a four-sentence plain-English summary, then (2) a short list titled "What this means for me day to day". Keep it jargon-free. DOCUMENT: [paste the remote-working policy]
Why this works: It sets who the answer is for, names an exact output shape, and fences off the document with a label, so the first reply is already organised the way you need it.
Read the reply against your own sense of the document. It should mention the hybrid model, the minimum office days, the core hours, and how remote arrangements are agreed. If anything looks off, that's the next step's job.
Step 4: Interrogate the details
This is the step most people skip, and it's the most valuable. Before you rely on the summary, probe the parts that actually affect you. Ask pointed questions and demand the exact wording.
Using only the policy, answer these, and quote the exact line for each: What's the minimum number of office days a week? What are the core hours when I must be contactable? Who do I need to agree a fully remote arrangement with? Am I allowed to use my own personal laptop for work?
Why this works: Asking for the exact lines turns Claude's answer into something you can verify in seconds against the original, and asking several precise questions is more reliable than one broad 'tell me everything'.
Now do the checking. For each answer, find the quoted line in the original policy. This takes a minute and it's the habit that keeps you safe: a summary can drift, but a quote you've confirmed with your own eyes cannot.
Step 5: Deliberately test a tricky question
Here's where you learn to catch a confident answer that could mislead. Ask Claude something the policy doesn't actually spell out, and see whether it invents a tidy answer or admits the gap.
Does this policy say whether Fernway reimburses home broadband costs, and whether I can claim for a desk chair? Answer using only the policy, quote the relevant line, and if either isn't stated, say "not stated".
Why this works: A question the policy doesn't clearly answer reveals whether Claude will guess. Adding 'if it isn't stated, say so' gives it permission to admit the gap instead of filling it with something plausible.
The policy says general utility and broadband costs are not routinely reimbursed, but it says nothing specific about a desk chair beyond "reasonable, pre-agreed costs". A good answer quotes the broadband line and marks the chair "not stated". If Claude instead confidently asserts a chair policy that isn't there, you've just seen exactly the kind of slip you're learning to catch, and why the "quote the line" habit matters.
Step 6: Turn it into something usable
Now shape what you've learned into a short output you'd actually keep or send.
Based only on what the policy actually says, write me a short personal checklist titled "My hybrid-working set-up" covering the things I need to confirm or arrange (office days, core hours, who signs off remote work, equipment). Keep each item to one line. Mark anything the policy doesn't cover as "check with my manager".
Why this works: Asking for a specific, self-contained format (a short personal checklist) gives you something to act on rather than a wall of prose. It will often open as an editable side panel you can tidy.
Because this is something substantial, Claude will often open it in a side panel, an artifact, rather than burying it in the chat. It sits to one side where you can read and edit it.
Step 7: Refine and take it away
The first checklist is a draft. Nudge it until it's right: "Put the must-do items first." "Add a line about keeping my shared calendar updated; that's in the policy." Each change updates the artifact in place. When it looks right, use the panel's copy or download control to take it into your notes, an email to yourself, or wherever you keep your to-dos.
That's the whole task: a multi-page policy in, a checked understanding and a personal checklist out, in about the time it would have taken just to read the thing once.
Try it now
Common mistakes
- Trusting the summary without checking. Always confirm a couple of exact quotes against the original before you rely on it.
- Accepting the first checklist. It's a draft; reorder it, trim it, add what's missing.
- Copying the artifact back into the chat to edit it. Edit it in the side panel, or just ask Claude to change it; it updates there.
- Doing it all in one giant prompt. Summarise, interrogate, then build. Small steps give better results and are easier to fix.
- Over-trusting a confident answer to your tricky question. As Step 5 shows, Claude can answer a question the document doesn't actually address, and it sounds no less sure when it does. Treat any answer you can't tie to a quoted line as unconfirmed. The confidence is not the evidence.
Keeping current
The flow (give it the document, summarise, interrogate, check, build, refine) is durable and will outlast any particular version of Claude. What changes is the detail: how artifacts look, what file types attach, how much a chat can hold. When the screen differs from this walkthrough, check the Claude release notes and help centre. Accurate as of 13 July 2026.