AI Prompting Guide for Contract Work

10 Annotated Prompts with Working Examples

For practising advocates, solicitors, barristers and in-house counsel

The Golden Rule — Read Before You Start

Never paste identifiable client information into a free AI tool (ChatGPT free tier, Claude.ai free, etc.). Anonymise all facts before prompting. Replace client names with Party A / Party B, remove financial figures if not necessary, and strip any identifying detail.

This is not a stylistic preference. It is a professional conduct obligation under your jurisdiction’s data protection rules and your duty of confidentiality.

How to Use This Guide

Each prompt in this guide is formatted in the same way:

The prompt template, ready to copy and adapt
An annotation explaining why it is structured that way
A worked example — a realistic before/after showing the kind of output you can expect and, critically, what you would still need to do with it

These prompts are not magic. They are starting points. Every output requires your professional judgment before it can be used. That is not a disclaimer — it is the point. The value of AI in legal practice is not that it replaces your analysis. It is that it drafts the scaffolding so you can work faster and focus on what requires a lawyer.

Section 1 — Contract Review

Prompt 1: Risk Identification

Use this prompt to surface the key risks in a clause before you begin redlining or advising.

You are a commercial solicitor advising [Party A / the buyer / the seller] under
[English law / Indian law]. Review the following clause and identify the top 3
risks it creates for my client. Do not suggest remedies yet. Do not invent
cases. Flag any area where your analysis may be uncertain.

[Paste clause]

Why this works

Separating risk identification from remedy drafting gives you two distinct, reviewable outputs. Asking the AI not to invent cases is a necessary constraint — LLMs will confidently cite non-existent authorities if you leave that door open. The instruction to flag uncertainty forces the model to signal gaps rather than paper over them.

Worked example

Input clause (anonymised): “The Supplier shall not be liable for any indirect, special or consequential loss arising from this Agreement, however caused, including loss of profit, loss of revenue, or loss of data.”

Sample output (risks for a buyer under English law):

The exclusion of ‘loss of profit’ is unusually broad and, depending on the nature of the services, may exclude the primary head of loss the buyer would suffer in a breach scenario. This creates a risk that a successful breach claim may result in only nominal damages recovery.
The phrase ‘however caused’ extends the exclusion to negligence and potentially to deliberate breach, which English courts may scrutinise under the reasonableness test in UCTA 1977 (s.11). Note: the AI has not cited a specific case here — you should verify the applicable reasonableness analysis independently.
There is no carve-out for death or personal injury caused by negligence, which cannot be excluded under s.2(1) UCTA 1977. If this agreement touches on any physical services, the clause as drafted may be void to that extent. Uncertainty: this depends on the subject matter of the agreement, which is not visible from the clause alone.

What you would do next: Review these three flags, verify the UCTA analysis against your own sources, and then decide which risks are deal-breakers versus negotiating points. The AI has saved you the first pass — it has not done the legal work.

Prompt 2: Clause Comparison

Use this prompt when you receive a revised draft and need to understand what has changed and what it means commercially.

Compare the following two versions of a limitation of liability clause.
Identify:
  (a) what has changed,
  (b) which version favours the supplier, and
  (c) which changes carry the most commercial risk for the customer.

Governing law is English law.
Do not give legal advice — give me an analysis I can review.

Version 1: [paste]
Version 2: [paste]

Why this works

Comparison prompts are where AI genuinely saves time. The instruction “give me an analysis I can review” keeps the output in draft territory, which is where it belongs. It also signals to the model not to dress the output up as final advice. Specifying governing law is essential — the model will default to US law if you do not direct it.

Worked example

Version 1: “Supplier’s total liability shall not exceed the fees paid in the 12 months preceding the claim.”

Version 2: “Supplier’s total liability for all claims under or in connection with this Agreement shall not exceed the lower of: (a) the fees paid in the 3 months preceding the event giving rise to the claim; or (b) £10,000.”

Sample output:

(a) What has changed:

The reference period has reduced from 12 months to 3 months
A hard cap of £10,000 has been introduced
The lower of the two figures now applies, compounding the restriction

(b) Which version favours the supplier: Version 2 substantially favours the supplier. In most SaaS or professional services agreements, three months of fees is a materially lower figure than twelve, and the addition of a hard cap means that even high-value relationships are capped at a nominal sum.

(c) Highest commercial risk for the customer: The ‘lower of’ construct is the most commercially significant change. It means the cap defaults to the lower figure in all cases, not just where the hard cap would otherwise apply. This change should be flagged for negotiation.

What you would do next: Advise your client on whether the cap is proportionate to the risk profile of the engagement. A £10,000 cap may be completely inadequate if the supplier is handling, say, critical payroll processing.

Prompt 3: Plain English Summary for Client

Use this prompt to produce a first draft of a client-facing explanation of a contract clause — particularly useful before a client meeting or in a covering email.

Summarise the following contract clause in plain English for a non-lawyer
client. Maximum 100 words. Do not use legal jargon. Do not give legal
advice or conclusions — describe what the clause does in neutral terms.

[Paste clause]

Why this works

Client communication drafts are one of AI’s strongest use cases. The word limit forces the model to be useful rather than comprehensive. The instruction not to give legal conclusions is important — you want a description, not a recommendation, so that you can add your own advice around it.

Worked example

Input clause: “Either party may terminate this Agreement immediately upon written notice if the other party commits a material breach of its obligations under this Agreement and fails to remedy that breach within 30 days of receiving written notice specifying the breach.”

Sample output (96 words):

“This clause allows either side to end the contract early if the other side seriously fails to meet their obligations under it. Before terminating, the party who wants to end the contract must first send a written notice explaining what the problem is. The other side then has 30 days to fix it. If they do not fix it within that time, the contract can be ended immediately, also in writing. Either party can use this process — it is not one-sided.”

What you would do next: Review the draft, adjust the language to your firm’s tone, and add your own advice about whether this termination right is adequate for your client’s situation.

Section 2 — Contract Drafting

Prompt 4: First Draft of a Clause

Use this prompt to generate a first draft of a standard commercial clause from a clear set of requirements.

Draft a confidentiality clause for a service agreement governed by
[English / Indian] law. The clause should:
  (a) cover both parties,
  (b) exclude information already in the public domain,
  (c) have a 3-year post-termination survival period, and
  (d) be written in plain English.

Draft only — do not advise on whether this is appropriate for any
specific situation.

Why this works

Specificity in the brief produces a usable draft. Without the constraints in (a)–(d), the model produces a generic clause that could be for any agreement in any jurisdiction. The final instruction keeps the model in drafting mode and prevents it from appending unsolicited legal commentary that you will have to strip out.

Worked example

Sample output (English law):

Confidentiality

Each party (the “Receiving Party”) agrees to keep confidential all information disclosed by the other party (the “Disclosing Party”) in connection with this Agreement that is marked as confidential or that would reasonably be understood to be confidential given its nature or the circumstances of disclosure (“Confidential Information”).

This obligation does not apply to information that: (a) is or becomes publicly available through no act or omission of the Receiving Party; (b) was already known to the Receiving Party at the time of disclosure; or (c) is disclosed with the Disclosing Party’s prior written consent.

The obligations in this clause survive termination or expiry of this Agreement for a period of three years.

What you would do next: Review the exclusions. Do they match what your client needs? Is there a carve-out required for regulatory disclosure? Does the clause need teeth — i.e., a specific remedy or injunction provision?

Prompt 5: Redline Suggestions

Use this prompt to generate negotiating amendments when you are acting for a party who has received a draft from the other side.

I am acting for the licensee in a software licence agreement. The following
clause was proposed by the licensor. Suggest 3 amendments that would better
protect the licensee's position under English law. Present each amendment as:
  (i)   the issue,
  (ii)  the proposed revised wording, and
  (iii) the reason for the change.
Do not invent cases.

[Paste clause]

Why this works

The structured output format (i/ii/iii) makes the response directly usable in a negotiation note or client email. Without this structure, AI tends to produce flowing prose that you would have to reformat. Specifying the client’s position — licensee — ensures the amendments are directional, not neutral.

Worked example

Input clause: “The Licensor may at any time, on 7 days’ written notice, modify or withdraw any feature of the Software without liability to the Licensee.”

Sample output — Amendment 1:

(i) Issue: Seven days is insufficient notice for a licensee to adapt workflows or source alternatives, particularly in an enterprise context.

(ii) Proposed wording: “The Licensor may modify or withdraw any material feature of the Software on not less than 90 days’ prior written notice.”

(iii) Reason: 90 days is a more commercially reasonable period. The addition of ‘material’ limits the obligation to significant changes, which is a compromise position the licensor may accept.

Sample output — Amendment 2:

(i) Issue: The ‘without liability’ carve-out is absolute and may exclude compensation even where a material feature withdrawal causes demonstrable loss to the licensee.

(ii) Proposed wording: “…provided that, where such modification or withdrawal materially reduces the functionality available to the Licensee, the Licensor shall offer a pro rata reduction in fees for the affected period.”

(iii) Reason: Ties relief to a commercial outcome rather than legal liability, which is often more palatable in negotiation.

What you would do next: Apply your judgment about which amendments are commercially realistic given the relative bargaining positions of your client and the licensor.

Section 3 — Dispute Analysis

Prompt 6: Issues Identification

Use this prompt to generate a first issues map at the start of a dispute matter.

The following is a summary of a contract dispute. Identify the top 5 legal
issues a court applying English / Indian common law principles would likely
need to resolve. Present as a numbered list. Do not give a view on outcome.
Do not invent cases.

[Paste factual summary — anonymised]

Why this works

Issues maps are excellent AI output. They give you a framework for your own analysis without the model speculating on outcome. The instruction not to give a view on outcome stops the model from producing confident predictions that could mislead a client if passed on unchecked.

Worked example

Anonymised facts: Party A (supplier) entered a services agreement with Party B (customer) for software implementation. Party A delivered a system that Party B claims did not meet the agreed specification. Party B refused to pay the final invoice of £180,000. Party A contends the specification was agreed in writing and the system met it. Party B relies on verbal assurances allegedly made by Party A’s sales team before contract execution.

Sample output:

Whether the written specification constitutes the entire agreement and whether any pre-contractual oral representations are incorporated into or collateral to it (entire agreement clause analysis; parol evidence rule).
Whether, if the specification is paramount, Party A’s deliverable met the contractual standard — which turns on interpretation of the specification’s terms and may require expert evidence.
Whether any pre-contractual representations by Party A’s sales team give rise to a claim under the Misrepresentation Act 1967, independent of the contract.
Whether Party B’s refusal to pay the final milestone constitutes a breach of contract and, if so, whether it is repudiatory.
Quantum: how the £180,000 unpaid invoice relates to the value of any defective performance, and whether Party B has a cross-claim that extinguishes or reduces Party A’s claim.

What you would do next: Expand each issue into a research note. The AI has structured your starting point — it has not done the research.

Prompt 7: Breach Analysis

Use this prompt to structure a preliminary analysis of whether particular conduct might constitute a repudiatory breach.

Under English contract law, analyse whether the following facts disclose
a potential repudiatory breach. Structure your answer as:
  (a) the applicable legal test,
  (b) the facts that point toward repudiation,
  (c) the facts that point against, and
  (d) the uncertainty in the analysis.
Do not give a conclusion. Do not cite cases unless you are certain they exist.

[Paste anonymised facts]

Why this works

Asking for both sides of the argument is one of the most valuable things you can do with AI. It surfaces counterarguments before the other side does. Instructing the model not to give a conclusion is deliberate — you are using it to generate material for your own analysis, not to outsource the analysis itself. ‘Do not cite cases unless certain’ is a necessary friction that reduces hallucination risk.

Worked example

Anonymised facts: Party A (contractor) was engaged to complete a building project by 31 March. On 1 March, Party A’s project manager informed Party B (employer) by email that the project would ‘definitely not be finished before June’. Party B wishes to terminate immediately.

Sample output:

(a) Legal test: A repudiatory breach is conduct that indicates an intention to abandon or refuse to perform contractual obligations — either expressly or by necessary implication. An anticipatory repudiation can arise before the date for performance if one party evinces a clear intention not to perform when performance falls due.

(b) Facts pointing toward repudiation: The project manager’s statement is unequivocal — the project will not be complete by the contractual date. If the individual had authority to bind the contractor, this is a clear statement of non-performance by the due date.

(c) Facts pointing against repudiation: The statement was made by a project manager, not a director or authorised representative. It may reflect a progress concern rather than a settled intention to breach. It is also 30 days before the completion date — there may be an argument that performance by 31 March remains possible.

(d) Uncertainty: The key uncertainty is whether the project manager’s statement constitutes a sufficiently clear and unequivocal renunciation, or whether it leaves open the possibility of performance. This is a factual and authority-dependent question that requires legal research.

What you would do next: Advise Party B whether to accept the repudiation (with the risks that entails if the breach analysis is wrong) or to wait and see whether performance is actually refused on 31 March.

Prompt 8: Quantum Note

Use this prompt to produce a structured first-cut analysis of heads of loss in a breach of contract claim.

Draft a short note setting out the heads of loss a claimant might claim
in a breach of contract claim under English law, based on the following
facts. Include: expectation loss, reliance loss, and any consequential
loss heads that are arguable. Note where any head is likely to face a
causation or remoteness challenge. Do not invent figures.

[Paste anonymised facts]

Why this works

Quantum analysis is often left late — sometimes until after a letter of claim has gone. This prompt forces an early structured look at what a claim is actually worth, which is essential for advising a client on whether to litigate, negotiate, or walk away. ‘Do not invent figures’ is critical — the model should flag heads of loss, not fabricate financial calculations.

Worked example

Anonymised facts: Party B engaged Party A to design and build a bespoke e-commerce platform for £250,000. Party A delivered the platform four months late and with material defects. Party B lost an estimated £120,000 in sales during the delay period (based on Party B’s own revenue projections) and spent £18,000 on a third-party firm to remedy the defects.

Sample output:

Expectation loss: Party B bargained for a functioning platform delivered on time. The expectation measure seeks to put Party B in the position it would have been in had the contract been performed. This is the primary measure.

Reliance loss: The £18,000 spent on third-party remediation is a direct out-of-pocket loss causally linked to the breach. This should be recoverable as reliance loss, though Party B may need to establish that the sum was reasonable.

Consequential loss (loss of profits / revenue): The £120,000 in claimed lost revenue is the most contested head. Two challenges arise: (a) causation — Party B must establish that but for the late delivery, those sales would have been made; and (b) remoteness — the lost revenue head must have been within the reasonable contemplation of the parties at the time of contracting. Whether the platform was critical to revenue generation, and whether Party A knew this, is the key factual question.

What you would do next: Advise Party B on the strength of the quantum case before any litigation decision is made. A claim where the main head (£120,000) is highly uncertain on causation looks different from one where it is robust.

Section 4 — Due Diligence

Prompt 9: Contract Summary for DD Schedule

Use this prompt to generate a first-draft summary of a contract for inclusion in a due diligence report.

Summarise the following contract for inclusion in a due diligence report.
Cover: parties, governing law, term, key commercial obligations, termination
rights, change of control provisions, and any unusual or high-risk clauses.
Present as bullet points under each heading. Flag anything that requires
legal review before signing off.

[Paste contract or excerpt — anonymised]

Why this works

Due diligence summaries are time-intensive and AI handles them well at a first-draft level. The structured headings ensure coverage across the key areas any DD review would expect. The ‘flag for legal review’ instruction ensures the output does not become a substitute for actual review — it drives the model to surface issues rather than paper over them.

Worked example

Context: Reviewing a supplier contract as part of an M&A due diligence exercise.

Sample output (abbreviated):

Parties: Supplier: TechCo Limited (England and Wales). Customer: AcquireCo Limited (England and Wales). [DD flag: verify whether TechCo is the correct entity for the group.]
Governing law: English law. Jurisdiction: non-exclusive English courts.
Term: Initial term of 36 months from 1 January 2023. Auto-renews annually unless 90 days’ notice given.
Key commercial obligations: Supplier to provide software access, 99.5% uptime SLA, and annual support package. Customer to pay monthly licence fee and reasonable co-operation obligations.
Termination: For cause (30 days’ notice to cure); no termination for convenience. [DD flag: buyer inherits this contract — no exit route without breach.]
Change of control: Clause 14.3 — supplier may terminate on 30 days’ notice in the event of a change of control of the customer. [DD flag: this is a critical risk for the proposed acquisition — requires negotiation or waiver before completion.]

What you would do next: Review flagged items against the deal structure. The change of control clause, in particular, requires immediate attention in any M&A context.

Prompt 10: Red Flag Identification

Use this prompt to surface unusual or high-risk clauses in a contract before or during due diligence.

Review the following contract and identify any clauses that would typically
be flagged in due diligence as unusual, high-risk, or requiring negotiation.
For each clause:
  (i)   state the clause,
  (ii)  explain why it is a concern, and
  (iii) suggest what a buyer's solicitor should request.
Do not invent cases. Note your uncertainty where it exists.

[Paste contract or excerpt — anonymised]

Why this works

Red flag prompts are most useful as checklist generators. They surface issues for the lawyer to consider, not conclusions to rely on. The structured format (i/ii/iii) makes the output directly transferable to a DD issues list. Asking the model to note uncertainty is important here because AI sometimes flags standard clauses as unusual — the instruction to acknowledge uncertainty forces a degree of calibration.

Worked example

Sample output — Flag 1:

(i) Clause: Clause 11 — Intellectual Property. ‘All intellectual property created by either party in connection with this Agreement shall vest in the Supplier.’

(ii) Why it is a concern: This clause purports to transfer IP created by the customer to the supplier. In a development or co-creation context, this means any bespoke work the customer commissions vests in the supplier, not the customer. This is unusual and commercially high-risk.

(iii) What the buyer’s solicitor should request: Amendment to vest IP created by the customer (or using the customer’s materials or know-how) in the customer, with a licence back to the supplier for the purposes of the agreement only.

Sample output — Flag 2:

(i) Clause: Clause 19 — Governing Law. ‘This Agreement is governed by the laws of the State of Delaware.’

(ii) Why it is a concern: Both parties appear to be English-registered entities. A Delaware governing law clause is unusual in this context and may indicate that standard terms have been used without customisation. It should be reviewed against the dispute resolution clause.

(iii) What the buyer’s solicitor should request: Amendment to English governing law and exclusive English court jurisdiction, absent a specific reason for Delaware law.

What you would do next: Cross-check the AI’s flags against your own reading of the contract. AI is good at pattern-matching to commonly flagged clause types — it is less reliable on context-specific issues that require understanding of the deal.

Before You Use Any of These Prompts

Professional obligations — checklist

Anonymise all client data before prompting. No names, entities, specific figures, or identifying details.
Verify every case reference independently. Do not rely on AI citations. LLMs invent authorities.
Treat all output as a first draft requiring professional review. AI does not have access to your client’s full facts, your firm’s risk appetite, or your professional judgment.
Check the governing law in your prompt. AI defaults to US law if you do not specify. Always name the applicable jurisdiction.
Never send AI output to a client or court without reading it in full. You are responsible for the content of anything sent on your file.
Check your firm’s AI policy. Some firms require specific tools, platforms, or approval for AI-assisted work. Know what applies to you.

A Final Note

AI will not make you a better lawyer. It will make a good lawyer faster.

The value of these prompts is not in the outputs they produce. It is in the thinking they free you up to do. When AI handles the first pass on risk identification, clause comparison, or quantum mapping, your attention goes where it creates the most value — in the judgment call, the client relationship, and the strategic advice that no model can replicate.

Use these tools deliberately. Keep your professional obligations front of mind. And check your output before it leaves your desk.