Small impactBeginnerTask

Compare two versions of a contract and see what actually changed

When a client or counterparty sends back a marked-up agreement (or worse, a clean one), find every change and what it means for you.

Instead of reading two versions side by side and hoping you catch the one sentence that moved, you give both versions to an AI tool and ask for a change table: what changed, where, and why it might matter. You still read the changes that matter. You just stop spending your attention finding them.

Time
Saves 10 to 30 min
About 35 min down to 10 min per document pair
Who does it today
Partners and firm owners, usually at the end of the day
Practice area
Firm operations

Illustrative planning estimate, not a measured result. Assumes a 6 to 12 page engagement letter, lease, or operating agreement where the other side returned a clean copy with no tracked changes. Traditional time is a careful side-by-side read. AI time is uploading both, reading the change table, and checking the flagged clauses against the source.

Traditional vs AI-native

Traditional

  1. 1Print or tile both versions
  2. 2Read paragraph by paragraph
  3. 3Mark differences
  4. 4Decide which ones matter

AI-native

  1. 1Upload both versions
  2. 2AI builds a change table
  3. 3AI flags the ones that matter
  4. 4You read those clauses in the source

Where you sit

You decide what to accept, push back on, or ask about, and you read every flagged clause in the original. The tool finds the changes. It does not give legal advice.

The traditional way

  1. Open the version you sent and the version that came back.
  2. Read them side by side, or run a word-processor compare if both are editable files.
  3. Mark each difference, including the small ones: a 'shall' that became 'may', a changed number of days.
  4. Decide which changes you accept, reject, or need to ask about.

Where it hurts

  • Clean copies returned as PDFs defeat the built-in compare tools.
  • Small wording changes with big consequences are the easiest to miss.
  • The careful read gets pushed to later, and later is when things get signed.

The AI-native way

  1. 1You

    Give it both versions and your side

    Upload the original and the returned version, and say whose side you are on and what you care about (fees, scope, liability, termination, payment timing). Context changes what counts as significant.

    Tools: Claude, ChatGPT, Gemini, or Microsoft Copilot

  2. 2AI

    Build the change table

    The model lists every difference with the section, the old text, the new text, and a one-line plain-English read of the effect. It separates substantive changes from formatting and renumbering.

  3. 3AI

    Flag what matters to you

    Changes that touch the things you named are marked for attention, with a suggested question or response for each. It does not tell you whether to sign.

  4. 4You

    Read the flagged clauses yourself

    Open the source documents at the sections flagged and read them in full. The table is a map, and the documents are the territory.

What you need

  • Both versions (Word, PDF, or even a clear scan)
  • A sentence on whose side you are on and what you care about
  • Any general-purpose AI tool that accepts file uploads

Copy the prompt

Two-version change table

I am attaching two versions of the same agreement. VERSION A is what we sent. VERSION B is what came back.
We are the [accounting firm / client / tenant]. We care most about: fees, scope of services, limitation of liability, termination, and payment terms.

1. List every difference between A and B in a table: section, text in A, text in B, plain-English effect.
2. Separate substantive changes from formatting, numbering, and typo fixes.
3. Mark any change that affects the things we care about with FLAG, and suggest one question or response for each.
4. List anything in A that is missing entirely from B.

Rules:
- Quote the exact text from each version. Do not paraphrase inside the quote.
- If you are not sure two passages correspond, say so.
- Do not advise whether to sign.

Example

Fictional data.

Engagement letter, sent vs returned (excerpt)

VERSION A §4: "Fees are due within 15 days of invoice."
VERSION B §4: "Fees are due within 45 days of invoice."
VERSION A §7: "Either party may terminate on 30 days' written notice."
VERSION B §7: "Either party may terminate on 30 days' written notice. Upon termination, Firm shall deliver all workpapers to Client."
VERSION A §9: limitation of liability paragraph
VERSION B: no §9

Change table (excerpt)

FLAG  §4  Payment terms  15 days became 45 days. Triples your collection window.
            Ask: is 30 days acceptable?
FLAG  §7  Termination     Adds delivery of "all workpapers" on termination.
            Under AICPA ET 1.400.200 you generally don't have to hand over working papers
            unless a contract requires it. This clause would be that contract. Ask counsel how to word this.
FLAG  §9  Missing         Limitation of liability removed entirely.
      §2  Formatting      Renumbered subsections, no change in meaning.
Missing from B: all of §9.

Guardrails

  • Treat the change table as a finding aid. Read every flagged clause in the original before responding.
  • This is not legal advice. Anything that changes liability, indemnification, or ownership of work product goes to counsel.
  • Remove or anonymize client identifiers before uploading to a consumer AI tool, or use an enterprise tool under a data agreement.

What can go wrong

Scanned PDFs are read imperfectly, so a real change is missed or a false one appears.

Ask for exact quotes from both versions for every row, and check the flagged rows against the source.

Demo

Demo video coming soon

Until then, the steps and the example above are the walkthrough.

Want to implement this in your firm?

Bring it to the weekly call. Members work through builds like this one together, with accountants who are already doing it.

Keep browsing