“Can I use AI?” sounds like a yes-or-no question. In a real client project, it covers several different decisions. Can you use a tool to organize your notes? Can you enter client data? Can you generate a first draft? Who checks the result, and what should the client know about the process?

Ask these questions before work begins. A brief conversation can establish what is permitted, what must stay out of external tools, how outputs will be reviewed, and whether specific uses require disclosure. Put the answers in writing. If the client already has a policy, ask for it and discuss any language that leaves the work ambiguous.

Our broader guide to AI boundaries for independent professionals covers the overall governance decision. This article gives you a practical conversation to have with one client about one engagement.

Why assumptions are unreliable

Workers and clients may attach different meanings to “minor use,” “assistance,” or “fully human work.” A 2026 CHI study by Hwang and colleagues examined interviews and surveys involving freelancers and clients on freelance platforms. In that study, workers often expected clients to infer AI use from the output, while clients were less confident they could do so and tended to favor more proactive disclosure. Participants also interpreted client policies differently, particularly when those policies allowed some kinds of use but left the boundary vague.

That research describes its participants and method; it does not establish what every client expects. It does give a strong reason to ask rather than guess. A silent client may welcome AI assistance, forbid particular tools, or care mainly about data and review. You cannot reliably infer which from the absence of a rule.

The conversation also has a trust benefit. You are showing that you understand the difference between using a tool responsibly and handing an unchecked output to a buyer. Make the discussion specific to the work, not an abstract debate about AI.

Question 1: Which tasks are in scope?

Walk through the actual workflow. You might propose AI support for brainstorming, outlining, code suggestions, transcription, research triage, or editing. The client may treat some of these differently. A planning memo based on public material has a different risk profile from a model that processes customer records or generates a final recommendation.

Try: “I sometimes use AI tools to speed up early synthesis and drafting. For this project, which parts of that process are you comfortable with, and which should be handled without them?” Follow with examples from the brief. If the client says “minor tasks only,” ask what that means here. Is rephrasing a sentence minor? Is summarizing an interview minor? Is proposing analysis code minor? Concrete examples prevent a vague permission from becoming a later dispute.

Do not present AI use as a requirement. A client can reasonably have constraints that make a tool inappropriate. If their preference changes the time or method needed, discuss the effect on scope and price before accepting the work.

Question 2: What information may enter a tool?

Separate public material from confidential, personal, regulated, licensed, or unpublished information. Ask whether the client permits any project information in third-party tools, whether approved tools or accounts are specified, and whether information must be anonymized. Do not assume removing a name makes a dataset safe to share; context can still identify a person or reveal a trade secret.

A useful question is: “What project information may I enter into an external AI service, if any? Are there approved tools, settings, or data-handling requirements I should follow?” If the answer is unclear, keep client material out until it is resolved. Confirm who can authorize an exception.

The NIST Generative AI Profile, published in July 2024, offers a framework for identifying and managing generative-AI risks. It is not a client contract and does not itself grant permission to share data. Its value here is to prompt a deliberate review of privacy, security, reliability, and human oversight in the workflow you propose.

Question 3: How will the work be reviewed?

Tell the client what you will personally check. Depending on the service, that may include factual claims, calculations, code behavior, citations, accessibility, legal or brand requirements, and whether the output actually answers the brief. The review should match the consequence of an error. A brainstorming list can tolerate rough edges; a client-facing analysis or production system cannot.

Ask: “What level of validation do you expect for AI-assisted work, and are there parts you want a named human reviewer to approve before delivery?” Clarify whether the client will review interim work, who owns final approval, and what evidence of checking would be useful. Avoid implying that an AI system can certify its own output.

If the work requires specialist judgment outside your competence, using a tool does not close that gap. The right move may be a qualified collaborator, a narrower deliverable, or a referral.

Question 4: When and how should use be disclosed?

Some clients want an upfront description of the workflow. Others want disclosure only when a tool materially contributes to a deliverable or processes specific data. Ask what would be useful to them and make the arrangement operational.

For example: “Would you like me to note AI assistance in the project plan, flag it in individual deliverables, or ask before specific uses? I can document the tasks, tools, and review steps we agree on.” This is more informative than a blanket “AI was used” label that says nothing about where judgment entered the process.

If a contract, policy, platform rule, or professional obligation specifies disclosure, follow it. If two instructions conflict, pause and resolve the conflict before proceeding. Do not treat the client’s casual comment as permission to ignore a written term.

Put the agreement into a short project note

After the conversation, send a concise summary: permitted tasks, prohibited tasks, approved tools or accounts, information that must remain outside tools, review responsibilities, disclosure format, and who can approve a change. Invite corrections. Attach or reference the client’s policy and relevant contract language rather than rewriting them from memory.

A plain note might say: “For this engagement, AI may be used for outline options and copy editing using public or approved project material. Client customer data and unpublished financial information will not be entered into third-party AI tools. I will verify factual claims and review all final copy. I will identify material AI assistance in the delivery note. Any additional use requires written agreement.” The terms are an example, not a default policy for every project.

Revisit the note when the work changes. A later request to analyze customer interviews, connect a new tool, or deliver automated outputs may require a fresh decision. Permission for one task does not automatically cover another. Keep the conversation proportionate: a short check before a new use is easier than repairing trust afterward.

If the answer remains unclear

When the client cannot provide a policy or an authorized answer, choose a workflow that does not depend on the disputed use while you seek clarification. Explain any effect on timing or cost. If the contract is material or the data is sensitive, qualified legal or security advice may be appropriate.

The goal is an agreement both parties can actually follow. SmartBid can help you prepare for an opportunity and track the pursuit decision; AI permissions for client work must come from the client relationship and governing terms. Clear questions at the start give your professional judgment room to do its job.