New to AI4 min readFreelance & Solopreneur AI

Solo AI rules for client work

Record six practical rules for each client: inputs, authorization, commitments, review, specialist help and incidents. Make the next tool decision from that record.

What you should be able to do

Write down the exact client-authorized tool, account, task and data path. Recheck the decision when any of them changes, and name the incident route before something goes wrong.

Saved only in this browser.
In this article

A freelancer needs a usable answer to “may this tool process this client’s material?” before the material enters it. Your own preference is only part of the answer: the contract, client instructions, other people’s rights and the actual account configuration also matter.

Write a short project card that records those decisions. A standing default is useful, but review it for each client rather than assuming one client’s permission covers the next.

Six rules, with a named decision for each

RuleWhat to record
InputsCategories kept out; minimum material permitted for the task; no credentials
AuthorizationExact product, account, purpose, data class and dated client approval where required
CommitmentsWho approves price, scope, deadlines and changes
Review and disclosureWhat must be checked, by whom, and what the client or publishing context requires you to disclose
Specialist helpThe point where the work exceeds your competence or authority, and who can help
Incident responseWho to contact and what to do if data or an unapproved output crosses a boundary

The card should let you decide a real task, not merely announce that you “use AI responsibly.” The NIST AI Risk Management Framework is useful background for documenting context and responsibility; your card is a small operational record, not a claim of framework compliance.

A completed permission decision

This is a fictional teaching example.

A client permits a copywriter to shorten already public service-page text in a named company account. The client’s unpublished pricing annex remains confidential. The copywriter records:

“Allowed: edit wording from the public page in the authorized account. Check every service claim against the page. Excluded: the private pricing annex, customer examples and new integrations. I approve the final scope and any delivery promise before sending.”

When the tool offers to connect the client’s CRM, the earlier decision does not cover that new data path. The copywriter stops before connecting and asks the authorized client contact. The useful control is the boundary written in advance, not a disclaimer added to the resulting copy.

Public service text passes through a named authorized account and pre-send checks, while a proposed CRM connection branches to a new permission decision.
The existing permission covers the public-page editing task. A new CRM connection creates a new data path and needs a fresh client decision. Open the full-size diagram.

Check the account, not the sales pitch

A paid plan or an enterprise product description does not establish that the account you opened meets the client’s conditions. Check training, retention, access, connected services and the applicable agreement. ChatGPT’s data controls and Temporary Chat documentation describe different controls; neither constitutes client authorization or a promise that submitted data was never processed.

For personal information, limiting the input is one requirement, not the entire legal analysis. The ICO’s minimisation guidance explains keeping data relevant and necessary. The EDPB’s controller and processor guidance addresses roles under the GDPR. Obtain qualified advice where the contract or data-processing role is unclear rather than asking a chatbot to approve the arrangement.

Keep commercial promises tied to authority

AI can format a proposal from terms you have chosen. It should not silently add a discount, deadline, warranty or extra revision round. Compare the draft against your approved scope and price before release. Use the proposal workflow for a fuller example and creative credits when publication requires a disclosure decision.

You can authorize a bounded automated step where the client agreement and your tested controls support it. The card should state that authority precisely, including what triggers a stop; it need not declare every formatting action permanently human-only.

Give incidents a first action

If an accidental disclosure occurs, stop the affected use and record the tool, account, time and data category without copying the sensitive material into another service. Follow the client’s incident process promptly. For exposed credentials, use the authorized revocation or rotation route. Preserve relevant evidence while the responsible people decide next steps; deletion is not a complete undo.

Use the six-rule client card with the client-secrets stop check. Keep it short enough to consult when the next unfamiliar task arrives.

Read next

Continue through the same learning path with the next practical articles.