Gains:
- Ability to summarize a long contract in a structured way around parties, obligations and key dates
- Ability to write a reusable review prompt that includes role, output format and citation requirement
- Ability to apply verification discipline by linking the AI summary back to the source article
Tens of pages of contracts land on a legal or compliance professional's desk every day: a service contract, a lease contract, a framework supply agreement. It takes time to read them cover to cover; But not reading is a risk. This is where artificial intelligence (AI) provides the most concrete, visible benefit in law: turning a long text into a structured summary in seconds. However, in this unit, we will underline one thing from the beginning: AI summarizes, you decide. The AI-generated brief is not a substitute for the lawyer reading the text; It functions as a map that speeds up reading and directs your attention to the right items.
Let's clarify a few terms. A prompt is a written instruction you give to the AI; A good summary starts with a good prompt. Hallucination is when the AI produces information (a date, an amount, an item) that is not actually in the text as if it were present; This is the most dangerous risk in law. Obligation is the action that a party is obliged to perform in accordance with the contract. Key dates are time points that have legal consequences, such as the duration of the contract, its termination, renewal or the date on which a notification must be made. There is also the concept of grounding: the discipline of basing everything the AI says on the relevant article number and quote. This is the heart of this unit.
Anatomy of the Structured Summary
A blank “summarize this agreement” instruction will give you a messy paragraph. However, the summary useful for legal review is structured around certain axes. A good contract summary answers these questions: Who are the parties? What is the subject of the contract? Who has which obligations? What are the prices and payment terms? How long is the period and how does it end? What needs to be done on which dates? Which substances are high risk? What is the applicable law and the competent court?
Here is the step by step process:
- Define the role and output format. Tell the AI to act like a senior contract specialist and output the output in a titled table/list.
- Give the text in full. The summary is only as good as the text it sees; A missing page means an incomplete summary.
- Make citation required. Ask for the item number and a brief quote for each determination.
- Require ambiguities to be marked. AI should be directed to say “not found in text” rather than making up information that is not in the text.
- Verify. Check each critical statement in the summary by going back to the source article through the quote.
Your role: a senior contract reviewer.Task: Summarize the following contract in a structured format.Output format (table):| Title | Summary | Item number | Short quote | Headings: Parties, Subject, Price/Payment, Duration, Termination, Liability/Compensation, Confidentiality, Applicable law/Authority, Key dates. Rules:- For each line, add the relevant article number and a quote of maximum 15 words.- For information you cannot find in the text, write "Not found in the text"; NEVER guess or make up.- Add comments; report only what is in the text.<agreement>[agreement text here]</agreement>
Tip: The instruction, "For information you can't find in the text, don't guess, write 'not found'" is the single most powerful hallucination-reducing sentence. When you give AI permission to fill in gaps, it is less inclined to make up gaps.
Requiring key dates as a separate output is worth its weight in contract tracking. Missing the termination notice period could mean being stuck with an automatically renewed contract for another year.
From the above contract, extract ALL dates and durations that have legal consequences. For each: event, date/duration, what article it comes from, what needs to be done. Example: “Notice of termination — 30 days before expiration — Clause 8.2 — written notice required.” Leave the relative periods ("60 days after signature") as is; Don't make assumptions to convert it to an exact date.
Weak Prompt / Strong Prompt
Weak prompt:Summarize this contract.
The result: a loose paragraph that doesn't tell you where to look, isn't linked to the source, and you can't verify what information is actually in the text. At worst, the AI can fill in the blanks with plausible-sounding but false statements and you won't even notice.
Powerful prompt: [senior expert role + titled table format + article number and citation requirement + "not found" rule + separate key date extraction + "add comment" limit]
The result: a verifiable summary that honestly flags missing information, with every line tied to the source article. The difference is whether the output can be legally trusted or not.
Different Types of Abstracts
Different summaries are required for different purposes from the same contract. You present a three-sentence executive summary to a manager; risk-focused brief to a negotiating team; metadata in the form of imprint for an archive record.
Summary type
To whom
Content
length
Executive summary
Decision maker/management
Topic, price, main risk, advice
3-5 sentences
Operational summary
Contracted team
Obligations, dates, actions
1 page
Risk-focused summary
law/negotiation
High risk substances, with citations
1-2 pages
Imprint (metadata)
Archive/CLM system
Side, date, type, value, duration
Table/fields
From the operational summary below, produce a 4-sentence executive summary for senior management: (1) contract what and with whom, (2) financial size and duration, (3) single most important risk, (4) clear recommendation (“available for signature”/“negotiate this item”). Use simple, jargon-free language.
Three Mini Cases
Case 1 — Missed renewal. A logistics company's warehouse lease agreement contained a clause stating that "if written notice of termination is not given 90 days before expiration, it will be renewed for another 3 years". This date was overlooked among the pile of contracts, and the notification was remembered on the 71st day. Key date extraction with AI in a pilot study scanned 140 contracts in the same portfolio and tabulated 11 critical renewal dates; The team realized that 3 of these would arrive within 60 days and initiated negotiations in a timely manner. The cost of a single missed notification was a $1.4 million annual commitment for that warehouse.
Case 2 — Fabricated amount. A compliance expert said "summarize this contract" and saw the phrase "penalty clause: 20% of the contract price" in the summary he received. However, the penalty rate was not written at all in the text; The AI had made up a “typical” rate. The expert almost put it in the report because he didn't want a quote. The team then added the rule "item number and citation are mandatory, write 'not found' if you cannot find it" to all summary prompts; Over the next 50 examinations, the fabrication detection rate dropped to zero.
Case 3 — Saving time. A legal team would review 200 supplier contracts, averaging 32 pages, for annual review. They did the initial screening with a structured AI digest, reserving only the 46 contracts flagged as “high risk” for full manual review. Total review time decreased from an estimated 240 hours to 90 hours; Moreover, the attention of the lawyers conducting the review was focused on the really risky files. The critical point: no contract was closed as “clean” based solely on the AI summary; The summary was used as a triage tool.
Common mistakes
- Relying on the quote-free summary. A summary that is not linked to the source cannot be verified; Each determination must bear the item number and citation.
- Giving incomplete text. The summary would be misleading if a page was skipped or part of the scanned PDF could not be read; Make sure the text is complete and readable.
- Let AI fill in the blanks. If there is no instruction to "make up what you can't find", the AI can fill in the blank with typical values.
- Substituting the summary for reading. For high-risk contracts, the summary is only a guide; Critical articles should still be read in full.
- Skipping key dates. If termination, renewal and notification periods are not separately extracted, they will be lost in the summary.
- Giving confidential data to an uncontrolled vehicle. Tool selection and masking (see Units 9 and 11) should not be overlooked in contracts involving client/personal data.
In summary
Contract summarization is the task where AI produces value most quickly in law: turning a long text into a source-linked and structured map. But its strength comes with discipline. Link each detection to the source with item number and citation, have the AI say “not found” instead of making up missing information, and use the summary as a triage tool — the final reading and decision always remains with you. AI summarizes; The competent professional decides what the contract means and whether it will be signed.
Application task
Select a contract you have (or a sample). (1) Summary with structured table prompt; Verify that each line has the item number and quote. (2) Run a separate key date extraction and note the 3 closest dates on a calendar. (3) Produce a 4-sentence executive summary from the summary. (4) Verify 5 randomly selected findings by going back to the source article; If there is a mismatch, reinforce your prompt with the "not found" rule.
checklist
- [ ] Does each finding in the summary include an item number and citation?
- [ ] Has the AI been instructed to "make up what you can't find, write 'not found'"?
- [ ] Have key dates and notice periods been issued separately?
- [ ] Is the text provided complete and readable (no missing/scanned pages)?
- [ ] Has the appropriate summary type (executive/operational/risk) been selected?
- [ ] Have the critical findings been verified from the source material?
- [ ] Have the tools and masking been checked for confidential/personal data?