Compare Documents Side by Side in ChatGPT

You have two versions of a contract. Or two vendor proposals. Or an old policy and a new one. You could read them line by line, but that is slow and easy to skim past the difference that matters. A well framed comparison in ChatGPT can turn hours of reading into a short table you can review in minutes.

The pattern is the same every time. Upload the files, name them clearly, and describe the columns you want. If you skip the columns, ChatGPT will invent its own and you will end up with a comparison of the wrong things. Column names are the whole game here.

What a side by side comparison really is

You are asking ChatGPT to read the documents, find the sections that talk about the same thing, and line them up in a table. Each row is a topic. Each column is one of the documents. The last column is usually a short note about the difference. That final column is where the real value lives.

When comparison beats reading in full

• Two versions of a contract before you sign the newer one
• Two vendor proposals with similar features but different terms
• Old and new versions of a policy you need to explain to a team
• Two candidate resumes for a role with a shared checklist
• Two research papers that reach different conclusions on the same topic

Prompt for two contract versions

Files: contract_v1.pdf and contract_v2.pdf.
Task: compare on these topics: payment terms, termination, liability cap, IP ownership, and governing law.
Return a table with columns: topic, v1, v2, change summary.
Quote the exact clause for each version.
If a topic is missing from a version, write "not present".

Why this prompt works

The topic list forces the comparison to focus on what matters. The quoting rule makes the table verifiable. The not present rule stops ChatGPT from inventing a match that does not really exist. Together these keep the table honest and short.

Prompt for two vendor proposals

Files: vendor_a.pdf and vendor_b.pdf.
Task: compare on price, delivery time, support hours, cancellation terms, and security posture.
Return a table: topic, vendor a, vendor b, notes.
Add a final row called overall risk with a short reasoned note based only on the compared topics.
Do not include a recommendation. I will decide.

Prompt for old and new policy

Files: policy_old.pdf and policy_new.pdf.
Task: list every change that affects an employee doing everyday work.
Return a table: topic, old policy, new policy, impact on employees.
Skip changes that are only editorial such as fixed typos.
Quote or closely paraphrase the exact wording so I can find each change in the source.

What to include in a comparison request

• The exact filenames so the columns are labeled correctly
• The topics or sections you care about, in the order you want them
• A rule to quote or closely paraphrase the source
• How to mark topics that are missing from one document
• A ban on invented recommendations if you want to decide yourself

How to refine when a row looks wrong

Do not redo the whole table. Point at the row that looks off and ask for the quoted text again with the page or section reference. That will usually surface either the missing evidence or the mistake. Row level corrections keep the rest of the table stable.

Common mistakes

• Asking for a comparison without naming the topics or columns
• Letting ChatGPT recommend a winner when you needed the facts
• Trusting a row with no quoted source
• Ignoring not present markers because they feel incomplete
• Mixing three or more documents into one table without a clear label per column

How to check the table

Pick two rows and open both documents to the referenced sections. If the quoted text matches, the row is trustworthy. If one quote is off, ask for a redo with stricter rules and rebuild the table before you make any decision on it.

Takeaway

A comparison is only as good as the columns. Name the topics, force the quotes, and the table becomes a short brief you can act on.