Summarizing a Contract With Gemini Without Missing the Catch
"Summarize this contract" is close to the least useful prompt you can give Gemini for an actual legal document. It's not that Gemini can't summarize a contract, it's that a general summary optimizes for readability, which means it smooths over exactly the asymmetric, one-sided clause that a careful review exists to catch. A summary that reads as balanced and reasonable can be technically accurate about every clause it mentions while still failing at the one job that matters: flagging the clause that's going to hurt your client. If you're new to Gemini itself, the Complete Beginner's Guide to Gemini covers the basics this article assumes.
Why a general summary hides the catch
A contract summary written for readability tends to compress each clause to roughly the same length and give each one roughly the same weight, regardless of how consequential it actually is. An indemnification clause that shifts unlimited liability onto your client gets the same one-sentence treatment as a boilerplate notices clause about where to send mail. Both are technically "covered" in the summary. Only one of them should have made you stop reading and call the client.
The fix isn't a better one-shot prompt. It's asking Gemini to look for specific categories of risk by name, the same categories an experienced reviewer scans for automatically, instead of asking for an even-handed overview.
Clause categories worth naming explicitly
Indemnification
Who pays if a third party sues. Watch for one-sided or uncapped indemnification obligations.
Limitation of liability
Whether liability is capped, and whether the cap is mutual or applies to only one party.
Termination
What triggers early termination, and whether the exit terms favor one side disproportionately.
Auto-renewal
Whether the contract renews automatically and how much notice is required to opt out.
Non-compete / exclusivity
Scope and duration of any restriction on future business or hiring.
Governing law and venue
Which jurisdiction applies, and whether that favors the other party's home turf.
A prompt that names these directly gets a fundamentally different, more useful response than a generic request:
Read this contract and identify every clause related to indemnification, limitation of liability, termination rights, automatic renewal, non-compete or exclusivity terms, and governing law. For each one, quote the specific language and tell me whether it appears to favor one party over the other, and why.
”That prompt does two things a general summary doesn't. It forces Gemini to actually locate and quote the relevant language, rather than paraphrase from memory of the document, and it asks explicitly for an assessment of which party a clause favors, which surfaces exactly the asymmetry a general summary tends to flatten.
Watching it happen on an actual clause
Here's a hypothetical clause, the kind of thing that shows up in a mid-size vendor services agreement, not a real contract:
Illustrative clause only, not a real agreement
"This Agreement shall automatically renew for successive one-year terms unless either party provides written notice of non-renewal at least ninety (90) days prior to the end of the then-current term. Vendor's liability under this Agreement, whether in contract, tort, or otherwise, shall not exceed the fees paid by Client in the twelve (12) months preceding the claim, except that this limitation shall not apply to Vendor's indemnification obligations under Section 9."
There are two catches buried in that one clause, and they work against each other in a way a generic summary is likely to miss entirely.
A generic summarization prompt run on the clause above, illustrated
That summary isn't false, exactly. It mentions the renewal and it mentions the cap. It's also missing both catches: it treats a ninety-day notice window as unremarkable, when a lot of clients expect the more common thirty or sixty days and could easily miss a deadline that opens three months earlier than they'd assume, and it describes the liability cap as protective without mentioning the carve-out that exempts indemnification claims from it entirely, which matters because indemnification is often where the largest exposure actually sits. A cap that doesn't apply to your biggest risk isn't much of a cap.
The clause-category prompt run on the same clause, illustrated
The second version catches both problems because the prompt asked two things the first one didn't: quote the actual language (which surfaces the ninety-day figure instead of letting "automatically renews" pass as boilerplate), and assess who a clause favors and why (which forces a comparison against a normal baseline, and forces the carve-out in the liability clause to be read against the cap it's supposedly limiting). A summary written for readability has no reason to make either of those comparisons on its own.
A second pass worth doing separately
Once the flagged clauses are in front of you, a second, narrower prompt is worth running specifically on the indemnification and liability language, since those tend to carry the most financial exposure and the most ambiguous drafting:
Focus only on the indemnification and limitation of liability sections. Is the liability cap, if any, mutual or one-sided? Does the indemnification obligation extend to the other party's own negligence? Are there any carve-outs from the liability cap, and what do they cover?
”Running this as a separate, focused pass rather than folding it into the first request tends to produce a more careful answer, the same reason a human reviewer reads the liability section twice rather than once alongside everything else.
What this still doesn't replace
This is not legal advice, and it's not a substitute for reading the contract yourself
Gemini can misidentify which party a clause favors, miss a cross-reference to a defined term buried in an earlier section, or misread a clause that depends on context from an exhibit or schedule it wasn't given. Every flagged clause needs a human legal read before it informs an actual decision, and nothing here should be represented to a client as the review itself.
There's also a practical limitation worth naming: contracts frequently define terms in one section and use them in another, and a defined term can quietly change the meaning of a clause many pages away. Gemini can miss that connection if the document is long or the definitions section isn't clearly linked in the text it's working from. Asking it to specifically trace defined terms used in a flagged clause back to their definitions is worth doing as its own step for anything genuinely high-stakes:
In the indemnification clause you flagged, find every defined term (capitalized terms like "Losses," "Claims," or "Affiliate") and quote how each one is defined elsewhere in the document. Tell me if any definition is broader or narrower than the term would normally suggest on its own.
”Confidentiality is a separate question worth asking first
Before uploading a client's contract into any AI tool, confirm what your firm's policy actually allows, and what protections apply to the specific product tier you're using. This is worth resolving before the workflow above, not after, since the answer determines whether any of this is appropriate to do with a given document at all.
The actual value here
Used this way, Gemini isn't replacing the review, it's changing what the first pass costs. Locating and quoting every indemnification, liability, termination, and renewal clause in a 40-page agreement by hand takes real time even for an experienced reviewer. Having those clauses already pulled out, quoted, and flagged for asymmetry before you start your own read means the human time goes toward judgment, not toward the first scan. That's a meaningfully different value proposition than "AI reviews your contract," and it's the one that's actually true.