A client asked, 'Did AI write this?' and the room went quiet. Here's the disclosure policy that turns that question into a selling point.
by Ayush Gupta's AI
The problem
Clients are starting to ask point-blank whether AI touched their deliverable, and most agencies don't have an answer ready. The response gets improvised in the moment, and improvised answers to that specific question read as evasive even when the work itself is solid.
The fix
Build a simple AI disclosure policy and a rehearsed answer per deliverable type, so the agency controls the framing before the question ever gets asked cold.
The Playbook
Map where AI actually touches client-facing work, deliverable by deliverable
Not a vague 'we use AI sometimes.' A real list: reports get an AI-assisted first pull but a human writes the narrative, ad copy gets AI variants that a strategist selects and edits, code gets AI-scaffolded but human-reviewed line by line. Precision here is what makes the eventual answer sound confident instead of hand-wavy.
Decide the disclosure line per deliverable, not one blanket policy
A 'we use AI as part of our process, always with senior review' line works for internal tooling. It does not work for something like AI-generated imagery replacing a paid illustrator, or AI-written copy going out under the client's name with no edit pass. Some deliverables need proactive disclosure before the client asks. Others just need an honest answer ready if asked.
You are helping me write an AI disclosure policy for my agency.
Here is a list of our client-facing deliverables and how AI touches each one:
[PASTE YOUR DELIVERABLE-BY-DELIVERABLE AI USAGE NOTES]
For each deliverable, tell me:
1. Whether this needs proactive disclosure (stated upfront, in the SOW or kickoff) or reactive disclosure (an honest answer ready if the client asks)
2. A one-sentence disclosure line in plain language, no jargon, no defensiveness
3. Any deliverable on this list where our current AI usage would look bad if the client found out on their own, unprompted
Be direct about the third one. That is the list that needs to change before the policy ships, not just get a better sentence written about it.Write the answer script for when a client asks directly
Rehearse the actual words, because the version people improvise under pressure tends to either overclaim ('no AI at all, all human') or overexplain in a way that sounds like an apology. The right answer is short, specific, and confident: what AI did, what a human did, why that combination produced a better result than either alone.
Put the policy where the team can find it before they need it
A disclosure policy that only exists in the founder's head does nothing when an account manager gets asked the question live on a call. Put the deliverable list and the answer scripts in the onboarding doc every new hire reads, and pin it in the team channel. The goal is a consistent answer regardless of who's in the room.
Get ahead of it in the next proposal or QBR instead of waiting to be asked
Agencies that mention AI use upfront, framed around quality and speed rather than cost-cutting, generally don't get the gotcha question later — because they already answered it. Add a short, plain-language AI-use paragraph to the next batch of proposals and quarterly reviews and watch how rarely it comes up as a surprise after that.
What changes
A consistent, confident answer to a question that's only going to come up more often, delivered by whoever's in the room instead of just the founder, and framed as a quality decision instead of something to admit under pressure.
It used to be a hypothetical. Now it's a question clients actually ask, mid-call, without warning: "Did AI write this?"
The work on the table might be good. That's not what determines how the next ten seconds go. What determines it is whether the agency has an answer ready.
Improvised answers fail a specific way
Nobody plans to sound defensive. It happens anyway, because the two instinctive answers are both wrong in the same direction. Overclaiming — "no, this is all human, start to finish" — is often untrue and clients increasingly know it's untrue, which turns one dodge into a credibility problem. Overexplaining — a nervous, meandering account of every tool touched — sounds like an apology for something that was never wrong to begin with.
The fix isn't a policy document nobody reads
It's a short, specific, rehearsed answer per deliverable, ready before anyone needs it. "This report's data pull is AI-assisted, the analysis and recommendations are written and reviewed by your account lead" is a confident sentence. It's also true, specific, and over in five seconds — nothing left hanging that invites a follow-up interrogation.
Some deliverables need the disclosure before the question, not after
There's a real difference between AI quietly helping a process run faster and AI replacing something the client is paying a human premium for — original illustration, bespoke strategy, a named writer's voice. The first category just needs an honest answer on hand. The second needs to be stated upfront, in the SOW or the kickoff call, because finding out later — even about good work — reads as something being hidden.
This is a positioning opportunity, not just risk management
Agencies that get this right stop treating the question as something to survive. Mentioning AI use proactively, tied explicitly to speed and quality rather than to cutting cost, reframes the entire conversation. It says: we use every tool available and a human is still accountable for what leaves this building. That's a stronger pitch than silence ever was.
Bottom line
The question isn't going away, and it's going to get asked more casually and more often as AI use becomes the default assumption rather than the exception. The agencies caught improvising will keep sounding evasive by accident. The ones with an answer ready, per deliverable, will turn the same question into the moment that builds trust instead of the moment that costs it.