Compare policies across your organisation
The problem
You've got a safeguarding policy, a complaints policy, a data protection policy, an HR handbook - and you suspect they don't all say the same thing. When someone asks about whistleblowing, do all the relevant documents agree on the process? You need to check for contradictions before an audit finds them.
The solution
Use an LLM to read all your policies and compare what they say about specific topics. The AI can identify where documents agree, where they contradict each other, and where there are gaps. You get a comparison table showing the differences, so you can decide which version should be the source of truth.
What you get
A structured comparison showing how each policy addresses specific topics (reporting procedures, timescales, responsibilities). Contradictions are highlighted with exact quotes from each document. You can also ask for gap analysis: what topics are missing from policies that should include them?
Before you start
- Your policies in digital format (PDF, Word, or text) - scanned PDFs need good OCR quality to work well
- A list of the topics you want to compare across documents
- A Claude or ChatGPT account (paid tiers handle longer documents better)
- Someone who knows what the policies should say to verify findings
When to use this
- You're preparing for an audit and want to check consistency
- You've just merged with another organisation and need to reconcile policies
- Your policies have been updated by different people over the years
- You're creating a new policy and want to check it aligns with existing ones
When not to use this
- You've only got one or two short policies - just read them
- Your policies are already managed in a proper document control system
- You need legally binding confirmation of compliance (AI can help you find issues, not certify compliance)
- Your policies are scanned PDFs with poor OCR quality - the AI will misread text and give unreliable results
Steps
- 1
Gather all relevant policies
Collect the policies you want to compare. This might be all policies, or just those that touch a specific area (safeguarding, data protection, HR). Make sure you have the current versions. Note the document name, version number, and last update date for each.
- 2
Define what you want to compare
List the specific topics or questions you want to check across policies. For example: 'What's the procedure for reporting concerns?', 'What are the timescales for responding to complaints?', 'Who is responsible for data protection?' Being specific gets better results than asking for general comparison.
- 3
Upload to NotebookLM or Claude
NotebookLM is particularly good for this: upload all your policies as sources, and it creates a research assistant that can answer questions about them. Claude Projects also works well. For either, upload the documents and give them clear names so you can tell which is which in the output.
- 4
Ask for topic-by-topic comparison
For each topic on your list, ask: "What does each policy say about [topic]? Please quote the relevant sections and note any differences or contradictions." The AI will pull the relevant sections from each document and compare them.
- 5
Request a comparison table
Ask for a summary table: policies as columns, topics as rows, with each cell showing what that policy says (or 'Not mentioned'). This makes gaps and contradictions easy to spot at a glance.
- 6
Verify findings against the original documents
For any contradictions or gaps the AI identifies, go back to the original documents to check. LLMs can misread or miss nuance. Your compliance lead or policy owner should review the findings before you act on them.
- 7
Create an action list(optional)
For each issue found, decide: which document needs updating, who owns that decision, and by when. Some contradictions are easy to fix (typos in timescales), others need proper review (different approaches to the same process).
Tools
Resources
At a glance
- Time to implement
- hours
- Setup cost
- free
- Ongoing cost
- free
- Cost trend
- stable
- Organisation size
- small, medium, large
- Target audience
- operations-manager, ceo-trustees, it-technical
Free tiers work for a few policies. For a full policy library, paid accounts (£18-20/month) handle the volume better. NotebookLM is free and good for this use case.