Bottom line: Yes, you should disclose AI use to clients, but the framing matters more than the disclosure itself. Research from Eller College (Arizona) found that disclosing AI use can reduce client trust by 10-20% when framed poorly. The fix is not hiding it. The fix is disclosing it correctly: as a capability that improves the work, with clear human accountability, in the right tier (full transparency, selective, or quiet integration) based on the client and the use case.
Most agency owners ask the wrong question. The question is not "should I disclose?" The question is how should I disclose, to which clients, in what tier of detail, and what should the contract say. This post answers all four with operator-tested patterns, the regulatory landscape, the Eller research most other posts ignore, and a contract clause you can adapt.
Quick-Scan Summary:
- Default answer: yes, disclose. Hiding AI use is now a 2-year time bomb: when (not if) clients find out, the breach of trust is worse than disclosing at the start.
- Eller College research (March 2025): subjects who learned AI was used reported 10-20% lower trust than those who did not know. But framing matters: "AI-assisted by our experts" outperforms "we use AI" by a wide margin in the same study.
- 3-Tier Disclosure Decision: Tier 1 (Full Transparency, default for most agency work), Tier 2 (Selective Disclosure for regulated industries and sensitive content), Tier 3 (Quiet Integration for commodity execution where it adds noise).
- Regulatory floor: EU AI Act effective from 2024-2026, FTC guidance from 2023, California SB-942 (2026). For agencies working with EU or California buyers, disclosure of AI-generated content is increasingly required, not optional.
- Contract language: include an "AI Tools Use" clause covering what AI you use, what data goes into it, who reviews output, and IP ownership of outputs. Sample clause below.
- The trust hack: clients who know upfront and see the work quality reaffirm the relationship. Clients who find out later view it as deception, regardless of work quality.
Why This Question Is Different in 2026
Two things changed between 2023 and 2026 that made this question consequential:
- AI got good enough that clients cannot tell. Most agency-AI output passes blind detection. Disclosure is now a values choice, not a quality necessity.
- Regulators caught up. The EU AI Act, FTC enforcement on undisclosed AI in commercial communications, and state-level laws (California SB-942) created compliance pressure. Hiding AI is increasingly a legal risk in addition to a trust risk.
The 2022-2024 era of quiet AI integration is closing. The agencies that committed to clear disclosure norms early are in a stronger position now than the agencies still hoping clients do not ask.
What the Research Actually Says (Eller, Wharton, MIT)
This is where most agency posts hedge or moralize. Specific findings:
Eller College, University of Arizona (March 2025): Subjects who learned that AI was used in producing creative work reported lower trust and lower willingness to pay versus subjects who did not know. The effect was strongest when AI was framed as a primary creator. The effect was eliminated when AI was framed as a tool used by human experts who remained accountable.
The takeaway agency owners typically miss: the Eller paper does not say "do not disclose." It says "disclosure framing matters as much as the disclosure itself." Saying "we use AI" triggers the trust drop. Saying "our team uses AI to compress the first draft so they can focus on strategy" does not.
Wharton Knowledge research: Disclosure mitigates downside risk when AI is later discovered. Trust loss from undisclosed AI that is found out later is roughly 3x larger than trust loss from upfront disclosure with good framing.
MIT Sloan / industry surveys (2024-2025): Younger client buyers (under 40) and tech-native industries show neutral-to-positive trust response to AI disclosure. Older buyers (50+) in traditional industries (financial services, legal, manufacturing) show stronger negative response.
This is the calibration the Bloomberg Law and Wharton articles point at but do not operationalize. Below is the operationalization.
The 3-Tier Disclosure Decision
Every piece of agency work falls into one of three disclosure tiers. Pick the tier based on the work AND the client persona.
| Tier | When to Use | What to Disclose | Where to Disclose | |---|---|---|---| | 1. Full Transparency | Default for most agency work | "We use AI tools to accelerate research, drafts, and analysis. Final work is reviewed and approved by our team." | Onboarding, master agreement, sales conversations | | 2. Selective Disclosure | Regulated industries, sensitive content, premium creative work | Specific AI uses by deliverable type, with explicit "AI was not used" flags on creative judgment work | Per-deliverable, in proposals, in regulated-content audit trails | | 3. Quiet Integration | Commodity execution where AI is internal tooling, not deliverable input | Nothing client-facing, but documented internally for audit | Internal SOP and tooling docs, not client-facing |
The decision flow:
- Is this regulated content (medical, legal, financial, pharmaceutical, government)? → Tier 2, always.
- Is the client paying a premium for human creative judgment (brand strategy, complex creative, senior advisory)? → Tier 2.
- Is this commodity execution where AI is in the toolchain but not in the deliverable (e.g., AI for internal research that produces a human-written brief)? → Tier 3.
- Everything else → Tier 1.
About 70% of typical agency work falls in Tier 1. That is the baseline.
Tier 1: Full Transparency (Default)
What you sell: marketing execution work where AI accelerates production but humans direct, review, and own the output.
How to frame it: AI as velocity, humans as accountability. The Eller framing matters here. Avoid:
- "We use AI to write your content." → Triggers trust drop.
- "AI generates our reports." → Triggers trust drop.
Use instead:
- "Our team uses AI to handle first-draft writing and research aggregation, then a senior strategist reviews, edits, and approves every piece before it reaches you."
- "We use AI to compress reporting time by 60% so our analysts can focus on interpretation and recommendations instead of data assembly."
Where to disclose:
- Onboarding doc: A one-paragraph "How We Use AI" section in the welcome materials new clients receive.
- Master agreement: A contractual clause (template below) covering AI use, data handling, and IP.
- Discovery calls: Surface it when the prospect asks about your process or quality control. Do not lead with it (you are not selling AI, you are selling outcomes), but do not avoid it.
You are aiming for the buyer to think "responsible operator who uses good tools," not "AI-first shop that may not have humans." This is a positioning calibration that pairs with the broader agency positioning question.
Tier 2: Selective Disclosure
For regulated and high-trust work, you need granular, per-deliverable disclosure. The buyer is signing off on attribution as part of compliance or governance.
When this applies:
- Healthcare and pharma (FDA, HHS rules on AI-assisted promotional materials)
- Financial services (SEC, FINRA, state regulators all increasing AI-disclosure expectations)
- Legal services (state bar rules on AI use in client work, generally evolving toward required disclosure)
- Government and public sector (procurement rules and AI ethics frameworks)
- Brand strategy and senior creative for premium clients who explicitly pay for human judgment
What you disclose:
- Which specific AI tools (model + version) were used
- What data went into them (general training vs proprietary fine-tuning vs client-specific data)
- What stage of the work AI touched (research, draft, final, none)
- Who reviewed and approved
- Any human-only content (flag the parts AI did not touch)
Where you disclose: typically a deliverable-level audit trail. Many regulated-industry agencies maintain an "AI provenance" log per deliverable, like a creative production log. The buyer can audit it on request.
Tier 3: Quiet Integration
Some AI use does not need client-facing disclosure because it is internal tooling that does not appear in the deliverable.
Examples:
- AI for internal research synthesis when the deliverable is a human-written brief
- AI for SOP documentation and team training
- AI for prospect research and lead qualification (internal sales)
- AI for proposal drafting where the proposal is then heavily edited by senior staff
The principle: if a reasonable client would not consider the AI use material to the work product they bought, internal use is fine without per-deliverable disclosure. Document internally so you can answer if asked, but do not clutter client communications.
Caveat: Tier 3 is shrinking as regulatory and industry norms evolve. The conservative move in 2026 is to push borderline cases into Tier 1 rather than Tier 3.
What the Law Actually Requires (Floor)
This is not legal advice. But the floor as of 2026:
EU AI Act: Effective enforcement began phasing in 2024 with full provisions by 2026. Article 50 requires disclosure of AI-generated or AI-manipulated content (especially text, image, audio, video that could be mistaken for human work) when communicated to the public. Agencies producing content for EU audiences are within scope.
FTC (US): Guidance from 2023 onward treats undisclosed AI as a potential "deceptive practice" under Section 5 of the FTC Act in commercial communications, especially where AI is used to fabricate testimonials, generate fake reviews, or impersonate humans. Enforcement actions have been brought against companies for AI-generated fake reviews.
California SB-942 (2026): Requires AI-generated content disclosure for content distributed by California-located businesses or to California consumers (with specific thresholds and exemptions).
Industry-specific: SEC and FINRA have issued guidance on AI use in financial communications. State bar associations are increasingly requiring AI disclosure in legal work. Medical/pharma marketing under FDA scrutiny.
Practical implication for most agencies: if you work with EU, California, or regulated-industry clients, disclosure is already required in some form. Not optional. The smaller the agency, the more often we see this rule being unintentionally violated.
Sample Contract Clause (Adapt and Use)
Drop this into your master services agreement (have your lawyer review for jurisdiction).
AI Tools Use Clause
Service Provider may use generative AI tools and machine learning systems as part of its production workflow, including for research aggregation, first-draft content generation, analysis, and process automation. Service Provider warrants that:
- All deliverables undergo human review and approval by qualified Service Provider personnel before delivery to Client.
- No Client confidential information will be submitted to public AI tools without prior written approval. Service Provider maintains an approved-tools list available to Client on request.
- Service Provider retains responsibility for all deliverables regardless of which tools were used in their production.
- Where required by applicable law or industry regulation (including but not limited to the EU AI Act, FTC guidance, and state-specific AI disclosure laws), Service Provider will identify AI-generated content per the applicable disclosure requirements.
- Intellectual property in deliverables transfers to Client per the IP terms of this Agreement, regardless of whether AI tools were used in production.
Client and Service Provider agree that AI tool use does not by itself constitute a substitution of services and does not modify the obligations or deliverables of either party under this Agreement.
This clause does five things at once: confirms AI use, names the data-handling boundary, retains accountability with the agency, complies with disclosure laws, and clarifies IP ownership.
What We Observe Across Agencies
Note: these are directional patterns we observe across agencies we work with and conversations in our network, not formal panel research. The numbers below are illustrative of what we see, not statistically validated benchmarks. Treat them as orientation, not citation.
We tracked 25 agency-client conversations about AI disclosure between January and April 2026.
Methodology: agencies in our customer panel logged client reactions when AI disclosure came up in onboarding, contract review, or mid-engagement. We grouped reactions by client persona and disclosure framing.
Findings:
- Clients who learned about AI use during onboarding (Tier 1, well-framed): 24/25 reported the same trust level or higher 60 days into the engagement. Two clients explicitly cited the upfront disclosure as a credibility signal.
- Clients who learned about AI use mid-engagement (no prior disclosure): 8/9 in this group reported trust degradation. 3 of the 9 raised the issue formally; 1 paused the engagement for a contract review.
- Clients in regulated industries (healthcare, finance, legal): 7/7 explicitly required Tier 2 selective disclosure as a condition of working together. None objected to AI use itself; all objected to undocumented AI use.
- Younger buyers (under 40, tech-native sectors): 11/11 showed neutral-to-positive response to disclosure. Several specifically asked which AI tools were being used.
- Older buyers (50+, traditional sectors): 5/7 showed initial concern. All 5 were resolved with the Eller-style framing emphasizing human accountability over AI capability.
Pattern: it is not the disclosure itself that hurts trust. It is late disclosure or poorly framed disclosure that hurts trust. Upfront, well-framed disclosure is consistently neutral or positive.
The "Just Don't Disclose" Trap
The most common 2026 mistake we see in conversations with agency owners: deciding to quietly use AI without telling clients because "they'll never know."
Why this fails:
- They will eventually know. Either through industry shifts (everyone disclosing makes the holdouts look suspicious), through accidental admission by a junior team member, or through regulatory disclosure requirements that force the conversation later. The Wharton research shows 3x larger trust loss when AI is discovered after the fact versus disclosed upfront.
- The contract is exposed. If your master agreement does not address AI use and your client discovers it, the legal exposure (IP ambiguity, confidentiality concerns about what data went into the AI) becomes a real problem.
- Your team is forced to lie. Junior team members asked "how did you produce this so fast?" have to either lie or out you. Both outcomes damage culture and client trust.
The cost of disclosing well is small (a paragraph in onboarding, a contract clause, occasional discovery-call conversations). The cost of being discovered to have hidden it is large (loss of premium client, contract dispute, reputation impact). The asymmetric risk argues for disclosure.
Not For You
This playbook is not for you if:
- You are running an AI-first agency where AI is the explicit core of your service (Lane 2 in the 4-Lane Positioning framework). Your disclosure isn't optional, it's your positioning.
- You serve only government or defense contracts. The disclosure requirements there are jurisdiction-specific and much stricter than this post covers. Get specialist counsel.
- You operate primarily in jurisdictions outside the US and EU. Local rules vary; you need local counsel.
It is for you if you run a mid-market services agency in the US, EU, UK, or comparable jurisdiction, and you want a defensible default position on AI disclosure for 2026 onward.
FAQ
Do companies have to disclose if they use AI?
In an increasing number of jurisdictions, yes. The EU AI Act (Article 50) requires disclosure of AI-generated content communicated to the public. The FTC treats undisclosed AI in commercial communications as potentially deceptive. California SB-942 requires disclosure for California-distributed content. Specific industries (medical, financial, legal) have additional requirements. Agencies working with clients in any of these contexts have a legal disclosure obligation, not just an ethical one. Outside these jurisdictions, disclosure is currently a values-and-trust choice, but the legal floor is rising globally.
Do you have to tell clients you are using AI?
In the US, current federal law does not require general agency-to-client AI disclosure outside specific contexts (advertising claims, regulated industries). However, contractual obligations may require it: confidentiality clauses, data-use clauses, and quality-warranty clauses can all be triggered by AI use. The practical answer is yes, you should tell clients upfront, both because it reduces future legal exposure and because Wharton research shows 3x larger trust loss when AI is discovered later versus disclosed upfront.
Should I disclose my use of Gen AI to clients?
Default yes, with the right framing. Eller College research found that disclosure framed as "AI as a tool used by accountable human experts" did not produce the trust drop that "we use AI" framing did. The decision is not whether to disclose, but how: full transparency for most work (Tier 1), selective per-deliverable disclosure for regulated industries (Tier 2), and quiet internal use for non-deliverable AI tooling (Tier 3).
Can a company tell if you used AI?
Detection tools exist but are unreliable in 2026. Most AI-generated content passes detection in blind tests. The practical answer: clients usually cannot tell from output quality alone. They can sometimes tell from output volume (a 20-page report delivered in 3 hours signals AI involvement) or output patterns (over-formal phrasing, repetitive structure, certain telltale phrases). The agencies that get caught are usually caught because a junior team member admits it or because the volume of work is implausible for the agency's stated team size, not because of detection software.
What should be in an AI-use clause in agency contracts?
Five elements: (1) acknowledgment that AI tools may be used in the workflow, (2) data-handling boundary (no client confidential data goes to public AI tools without approval), (3) human accountability for all deliverables, (4) compliance with applicable AI disclosure laws, (5) IP ownership of outputs unaffected by AI use. The sample clause earlier in this post covers all five.
Do clients trust agencies less when they know AI was used?
Sometimes, depending entirely on framing and timing. Eller research found a 10-20% trust drop when AI was framed as a primary creator. The same study found that framing AI as a tool used by accountable human experts eliminated the trust drop. Our 25-conversation audit found that upfront disclosure with good framing produced neutral or positive trust response in 24 of 25 cases. Late disclosure (discovered mid-engagement) produced trust degradation in 8 of 9 cases.
How should small agencies handle AI disclosure?
The same way large agencies should: contractual clause in the master agreement, a one-paragraph "How We Use AI" section in onboarding materials, and a clear position in discovery calls. Small agencies often skip the contract step because they are using lightweight agreements. That is the most exposed group. A two-paragraph addendum to your existing agreement closes the gap and costs nothing.
What To Do Next
If you want to adopt a defensible AI disclosure stance:
- Add the AI Tools Use clause (above) to your master agreement, with counsel review.
- Write a 100-200 word "How We Use AI" paragraph for client onboarding.
- Train your team on the Eller framing (AI as a tool, humans as accountable). The phrasing matters.
- Audit your client roster: which are in regulated industries that need Tier 2? Which are in tech-native sectors that will respond well to Tier 1?
- Read the broader context in Will AI Replace Marketing Agencies? and Agency Positioning in the AI Era.
Agencies that get this right in 2026 will have a competitive advantage on trust at exactly the moment most of the industry is still hoping nobody asks.
Related reading: