Will Clients Know You Used AI? Set Expectations First

A client might know you used AI because you tell them, because the workflow records it, or because they ask. This article has no test data showing how often clients can identify AI-assisted work from the finished text.

Correction, September 22, 2026: The earlier client email, reply, and successful resolution were fabricated. Claims about what clients can usually detect have been removed. The example wording below is explicitly hypothetical.

Do not confuse quality with proof of authorship

Generic wording, missing details, invented facts, and a mismatched tone are reasons to revise a draft. They do not by themselves prove that AI wrote it. Human work can have the same defects.

Likewise, polished editing does not establish that AI was absent. This guide makes no claim that a detector can conclusively identify or exclude AI use in every case. Do not build a client relationship around trying to evade detection.

Set the working agreement first

Clarify which tools, tasks, and types of data are permitted. A client may allow brainstorming with public information while restricting uploads of internal documents. They may require disclosure, review, or a particular approved service.

Check the actual contract and applicable professional or legal requirements. There is no universal disclosure rule established by this article. If those obligations are uncertain or consequential, obtain appropriate advice before proceeding.

Permission to use a tool also does not authorize invented experience, fabricated sources, or claims that a person performed work they did not perform.

Describe the process accurately

A hypothetical message before a project could say:

I propose using an AI tool to organize non-confidential notes and draft outline options. I will verify factual claims and review the final work. Please confirm whether this fits your policy and which materials or tasks must stay outside the tool.

Use that wording only if it describes your intended process. If you use AI more extensively, say so. Do not label a generated draft “just spell-checking” because that sounds easier to explain.

The freelance AI guide separates draft preparation from delivery and includes prompts that avoid inventing scope.

If an agreement was breached

First establish what happened: the tool, material, access, and action involved. Stop the affected workflow, preserve the information needed to assess it, and follow the relevant agreement and incident process. An inaccurate paragraph and an unauthorized data upload may require different responses.

Be truthful about the facts you know and the limits of your investigation. Do not promise that deleting a chat reverses every consequence, or that rewriting a paragraph resolves a confidentiality problem.

Deliver work you can account for

Before sending, compare the output with the brief, verify material claims, inspect citations, and remove invented examples presented as fact. Keep the review proportionate to the consequences.

At PickyFox, that includes disclosing AI-generated content and using the publication’s byline. An anonymous owner does not require a fictional human biography.

For the task boundary, use what to delegate to AI. For the wider process, see the AI workflow checklist.