Practical Tips and Considerations When Using AI

Artificial intelligence (“AI”) tools have rapidly become part of everyday business operations. Many people now use AI to assist with digesting information from lawyers, drafting contracts, summarizing agreements, reviewing policies, conducting preliminary legal research, and generating correspondence. While AI-generated drafting or reviews raise important legal and practical concerns, these tools may still serve useful purposes in certain situations if used carefully and appropriately.

This article outlines some practical tips and considerations when using AI tools in connection with contracts, legal documents, or legal concepts.

For an overview of the risks of relying on AI, please see our related article, “Risks of Using AI or Template Agreements”, and for a broader overview of AI-related considerations, please see “Using Artificial Intelligence (AI): Important Things To Know”.

 

1. Provide Context

The quality and usefulness of AI-generated content is heavily influenced by the instructions and information provided by the user. Broad, vague or incomplete prompts often produce generic or incomplete answers, while clear instructions and relevant context are more likely to produce useful and tailored results.

Moreover, AI tools can only work with the information available to them. Even a detailed prompt may omit facts, documents, legal requirements, or business considerations that the user does not realize are relevant.

This can be particularly important in legal matters, where the significance of a fact may not be obvious and one agreement may need to be considered together with other contracts, corporate documents, legislation, or regulatory requirements.

Practical Tip #1: Provide sufficient relevant context regarding the transaction, parties, objectives, jurisdiction, and intended audience, but do not provide confidential, privileged or personal information unless you have first determined that it is appropriate to do so.

Practical Tip #2: Do not assume that a complete-sounding answer means all relevant issues have been considered. Ask what additional facts, documents, assumptions, or legal considerations could affect the response.

Practical Tip #3: Ask the AI to identify the legislation or authorities on which its response is based. Current legal requirements should be verified using authoritative and up-to-date sources.

 

Less Effective (vague) More Effective (specific)
“Draft a services agreement.” “Draft a services agreement governed by Ontario law for a software consulting company providing implementation services to a commercial client. Include intellectual property, confidentiality, limitation of liability, and payment provisions.”
“Summarize this contract.” “Summarize this contract for a business executive and identify key obligations, termination rights, payment provisions, indemnities, and unusual risks.”
“Explain this clause.” “Explain this clause in plain language and identify any legal or commercial risks it may create for the customer.”

 

2. Do not assume AI has all of the relevant information

One of the limitations of using AI for legal or business matters is that the user may not know which facts are legally significant. An AI system can respond to the information it receives, but important issues may never be considered if the user does not know to mention them.

Practical Tip #4: Treat AI as a tool for assisting with an identified task rather than as a substitute for identifying the scope of the problem itself. If significant rights, obligations, money or liability are involved, professional advice may be appropriate even where the AI-generated answer appears complete.

 

3. Avoiding Leading Prompts and Confirmation Bias

The quality of an AI-generated response is often influenced by how the question is asked. Prompts that favour a particular conclusion, contractual position, or legal interpretation may result in responses that disproportionately support that outcome. By contrast, prompts that are framed in a neutral and open-ended manner are more likely to generate a balanced analysis, identify competing considerations, and highlight potential risks or uncertainties.

This is particularly important in legal contexts, where users may unintentionally frame a prompt in a way that seeks validation of an existing view rather than an objective assessment.

Practical Tip #5: When drafting prompts, avoid language that assumes a legal conclusion or seeks confirmation of a conclusion (e.g., “explain why this clause is enforceable” or “confirm that this agreement contains no material risks”). Instead, frame prompts in a manner that encourages analysis rather than confirmation by asking for supporting and opposing considerations, and explaining what additional facts or context could affect the analysis.

Less Effective (leading) More Effective (Open-Ended)
“Explain why our standard limitation of liability clause is clearly enforceable.” “Review the following limitation of liability clause and identify arguments that support and challenge its enforceability under Ontario law. Highlight any uncertainties and additional facts that may be relevant.”
“Draft a non-solicitation clause that will prevent former employees from working with our customers and will be fully enforceable.” “Draft a non-solicitation clause for an Ontario employment agreement that seeks to protect customer relationships while maximizing the likelihood of enforceability. Identify any legal risks or drafting considerations.”
“This vendor agreement appears favourable to our company. Confirm that it contains no significant legal risks.” “Review this vendor agreement and identify material legal, operational, and commercial risks, including both favourable and unfavourable provisions from our perspective.”
“Explain why the employee is clearly an independent contractor.” “Based on the facts below, identify factors that support and undermine characterization as an independent contractor and explain the associated legal risks.”

 

4. Verify the Output

AI-generated responses can be inaccurate even when they are written confidently and appear authoritative. An AI tool may misunderstand the facts provided, overlook an important issue, misstate a legal principle, or generate an incorrect or nonexistent case, statute, quotation or source.

When accuracy matters, do not rely on the apparent confidence or detail of the response as evidence that it is correct.

Practical Tip #6: Independently verify important facts, legal authorities, calculations, dates, quotations and other information before relying on or acting on an AI-generated response. Where the response identifies a case, statute, regulation or other source, review the original source rather than relying solely on the AI’s summary of it.

 

5. Protect Confidential and Sensitive Information

Users should exercise caution before entering confidential, proprietary, or legally sensitive information into AI platforms. Information may, depending on the platform, account type and settings, be used for training future AI models, product improvement or other secondary purposes.

In some circumstances, sharing confidential information through a third-party platform may also create privilege or confidentiality concerns. For these reasons, sensitive information should not be disclosed unnecessarily.

Practical Tip #7: De-identify information where possible (for example, by using placeholders for names or identifying details), and remove confidential or sensitive information from documents before uploading them unless disclosure is necessary and appropriate.

Practical Tip #8: Before using an AI platform, review its privacy terms, data-retention practices and available security/privacy settings.

 

6. Use AI Iteratively

The first response generated by an AI tool does not need to be the final response. Better results often come from reviewing the initial output, identifying gaps or weaknesses, and asking targeted follow-up questions.

For example, users can ask the AI to explain its assumptions, identify information that may be missing, consider alternative interpretations, or review its own response for potential errors or overlooked issues.

Practical Tip #9: Treat the initial response as a starting point. Ask follow-up questions that test the analysis, challenge assumptions, and identify areas where additional information or review may be required.

 

7. Know When Human Review is Necessary

AI can be useful for preliminary drafting, summarizing information, identifying issues, or organizing thoughts. However, AI-generated content may appear complete or authoritative without identifying important legal, factual or strategic considerations, making reliance on AI alone particularly risky where a matter involves important legal rights, financial exposure, ongoing obligations, regulatory requirements or potential liability.

Practical Tip #10: Consider obtaining professional review where the outcome is important, the legal issues are complex, the consequences of an error could be significant, or you are unsure whether the AI has identified all relevant issues.

 

Final Thoughts

Although these tools can improve efficiency, reduce administrative burden, and help users quickly identify issues within lengthy documents, AI should be viewed as an assistive technology rather than a substitute for legal advice.

Even where detailed prompts are used, AI-generated agreements may still contain legal, factual, or drafting errors. AI-generated content should not be assumed to have been independently or reliably verified, even where the response includes sources, citations or confident legal conclusions. Those sources should themselves be reviewed before the output is relied upon.

Accordingly, AI-generated drafts should generally be treated as preliminary working documents rather than finalized agreements, particularly where the transaction involves significant financial exposure, ongoing obligations, regulatory considerations, or potential liability.

 

If you have questions about using AI or AI tools, or would like to discuss your legal needs, our team at CARREL+Partners LLP is here to help guide you every step of the way. Get in touch today to discuss your situation.

Contact Us
May we leave you a voicemail at this number?

 

Disclaimer:

This publication is for general information purposes and is not to be taken as legal advice. The information within is current only to the date of publishing. The use of AI in legal drafting is an evolving area, and the risks associated with AI-generated or template-based agreements will vary depending on the specific tool, the nature of the agreement, and the applicable jurisdiction. Legal requirements may differ across provinces, including Ontario. Businesses and individuals should consult qualified legal counsel to assess their specific situation. If you have any questions regarding article content, please contact the author(s) directly.