Turn a Long Policy PDF into a One-Page FAQ with Claude
Most employees never read the policy. They ask a colleague, guess, or email HR the same question for the tenth time. A one-page FAQ fixes most of that, and Claude can draft one from a long policy PDF in about two minutes. The part that makes the draft safe to use is not the summary itself. It is two rules in the prompt: every answer cites its section, and anything the policy doesn't cover is labelled instead of guessed.
Here is the short video, then the full prompt and the review steps.
What you need
- The policy as a PDF (or Word file). The one in the video is a fictional 15-page hybrid working, travel and expenses policy.
- Claude on a plan your company has approved for internal documents. Only upload a document to a tool your company allows for that kind of data.
- Ten minutes for the review afterwards. The draft takes two minutes; checking it takes longer, and that is fine.
The prompt
Start a new chat, attach the PDF, and paste this:
Write a one-page FAQ for employees from the attached policy, grouped under short headings. Reply in the chat, no artifact.
1. List the 15 questions employees are most likely to ask, in plain English.
2. Answer each in two sentences or less.
3. After each answer, cite the section number, like (s. 8.4).
4. If the policy doesn't answer something, write "Not covered in this policy". Don't guess.
5. End with 3 questions employees will ask that this policy doesn't answer.
Each line has a job:
- "Questions employees are most likely to ask" makes Claude write from the reader's side ("Can I work from abroad?") instead of mirroring the policy's own section titles.
- "Two sentences or less" keeps it to one page. A long answer usually means the policy itself is unclear there.
- "Cite the section number" is the rule that matters most. It turns every answer into something you can check in seconds.
- "Not covered in this policy. Don't guess." gives the model a permitted way to say no. Without it, a model tends to fill gaps with something that sounds reasonable.
- "3 questions this policy doesn't answer" is a free policy review. In the video it flagged part-time office days, reasonable adjustments and expenses for an electric car, none of which the fictional policy addresses.
Check it before you share it
Treat the FAQ as a strong first draft. Before it goes to employees:
- Check every answer against its cited section. Open the PDF next to the FAQ and read the cited clause for each answer. This is quick precisely because of the citations.
- Check numbers and deadlines twice. Amounts, day limits and notice periods are where a small error costs the most.
- Decide what to do with the gaps. Either answer them in the FAQ after asking the policy owner, or note them for the next policy review. Don't let the FAQ invent policy.
- Ask the policy owner to sign it off. The FAQ will be read far more often than the policy, so it should carry the same approval.
Variations that work
- Benefits guide or employee handbook: the same prompt works; raise the number of questions to 20 for a long handbook.
- New starters: add "Write for someone in their first week" to the first line.
- Managers: ask for "the 10 questions managers will be asked by their team, with the section to point them to".
- Other languages: add "Then translate the FAQ into Arabic, keeping the section numbers" for a bilingual version, and have a native speaker check it.
Why this belongs in AI training
This is the kind of task I use in hands-on team sessions, because it teaches the habits that make AI output safe at work: give the model the source, make it cite, give it permission to say "not covered", and review before you send. Teams that learn those four habits on their own documents get value from ChatGPT or Claude in the first week.
If you want your team trained this way on your own documents, see remote AI training for teams or the corporate AI training page.