68 marketing, creative, PR and digital agencies in 7 countries publish how they use AI on client work. We read every policy in full and marked what each one actually commits to, in the agency's own words. Filter by agency type or by commitment, read the line worth copying, and open the original.
promise a person reviews AI-assisted work
90%
say they tell clients how AI was used (37 up front, 14 when asked)
Marketing agency · New Zealand · policy page · undated
“If you do not want us to use AI, please inform us, and we will adjust the cost of deliverables accordingly.”
Brief client notice: AI writes copy, edits designs and transcribes; only public information goes in; clients can opt out, with deliverable pricing adjusted.
Human review
Explains when a client asks
Client data rule
Client can limit AI
In its own words
Human review
“All AI-generated information is reviewed by a human to ensure its accuracy and reliability.”
Tells clients
“Should you have any questions or concerns about our use of AI, please do not hesitate to contact us.”
Client data rule
“We refrain from using private company information. Only publicly accessible information or general business themes are utilised, along with data that will soon be publicly available”
Client can limit AI
“If you do not want us to use AI, please inform us, and we will adjust the cost of deliverables accordingly.”
Marketing agency · policy page · dated November 21, 2024
“AI assists in content creation but is never the sole creator. Every piece of content is human-developed and refined.”
Lists its approved tools by name, says when AI shaped content, never shares client data with AI without consent, and stops AI on request.
Human review
Tells clients up front
Client data rule
Says what AI never does
Client can limit AI
Names its tools
In its own words
Human review
“all outputs are rigorously reviewed and refined by our team to meet your expectations.”
Tells clients
“We are committed to being upfront about how we integrate AI into our work. This includes clearly stating when AI has been used to create or refine content.”
Client data rule
“Your data will never be shared with AI tools or external entities without your consent.”
Says what AI never does
“AI is never used to mislead, manipulate, or impersonate.”
Client can limit AI
“At any time a client requests Evenbound not use AI, we will honor this request.”
Names its tools
“We only use carefully vetted AI tools that meet our quality and security standards, including:”
Marketing agency · United States · policy page · undated
“Our company policy is that we should NEVER publish or send something that has been written entirely by AI without human development or review for quality and accuracy.”
Never sends anything written entirely by AI, keeps AI away from website code, legal agreements and client strategy, and approves Jasper as its tool.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Names its tools
In its own words
Human review
“Our company policy is that we should NEVER publish or send something that has been written entirely by AI without human development or review for quality and accuracy.”
Tells clients
“This includes acknowledging when AI has been used to create or modify content. This can be through a blanket statement on our website or integrated into contracts with clients.”
Client data rule
“do not submit customer data into AI tools or LLMs.”
Says what AI never does
“Do not use AI for the following:; Website coding; Legal Agreements; Client Strategy”
Ownership and originality
“It is our company policy that employees should not use AI to impersonate any person without their expressed permission.”
Names its tools
“Jasper, the AI copilot for marketing teams has best-in-class security and privacy policies including SOC2 compliance, SSO and U.S. data storage.”
Marketing agency · section of a legal page · undated
“Agency will not use Client Content to train or develop any general-purpose large language model, whether owned by Agency or a third party, and will not use Client Content for the benefit of any other Agency client.”
Contract addendum: AI used by default unless the SOW prohibits it; client owns AI-assisted deliverables, but the agency disclaims infringement and client holds it harmless.
Human review
Tells clients up front
Client data rule
Ownership and originality
Client can limit AI
In its own words
Human review
“Agency will use commercially reasonable efforts to ensure human review and validation of all outputs generated by AI Tools prior to presentation or delivery to Client.”
Tells clients
“This AI Addendum reflects Agency’s commitment to provide Client with transparency regarding how Agency uses artificial intelligence tools to support its work”
Client data rule
“Agency will not use Client Content to train or develop any general-purpose large language model, whether owned by Agency or a third party, and will not use Client Content for the benefit of any other Agency client.”
Ownership and originality
“Agency further assigns any and all intellectual property rights it has in any Deliverables or work product generated in connection with the Services to Client.”
Client can limit AI
“Unless expressly prohibited in a Statement of Work (“SOW”), Agency will use AI Tools in connection with its provision of Services and/or Deliverables in accordance with this AI Addendum.”
Marketing agency · United States · policy page · dated September 13, 2024
“Any content created using AI platforms will be reviewed and cleared for potential copyright or legal issues.”
Formal staff-wide policy: no PII in AI tools, data minimisation, copyright clearance of all AI content, misuse handled through disciplinary procedures, reviewed annually.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
In its own words
Human review
“We will validate the outputs of AI to avoid errors or inaccuracies in client deliverables and prevent the spread of misinformation.”
Tells clients
“We will be transparent about when and how AI is used, both internally and with clients.”
Client data rule
“We will not upload any personally identifiable information (PII) (like names, email addresses, etc.) or personal data of any individuals (like phone numbers) when using AI platforms, and anonymized data will be used whenever possible.”
Says what AI never does
“We will also avoid using AI in any manner that could be considered discriminatory, biased, or invasive.”
Ownership and originality
“Any content created using AI platforms will be reviewed and cleared for potential copyright or legal issues.”
Marketing agency · United States · policy page · dated June 16, 2025
“No sensitive or confidential client data (such as passwords, financial information, or personal identifiers) is ever entered into public AI systems.”
An AI privacy policy from an agency serving contractors: names seven AI tools, keeps sensitive client data out of public AI, and explains AI use when asked.
Human review
Explains when a client asks
Client data rule
Names its tools
In its own words
Human review
“Every client-facing deliverable is reviewed and approved by a qualified team member before delivery.”
Tells clients
“If you have questions about how AI contributed to your project, we’re happy to explain.”
Client data rule
“No sensitive or confidential client data (such as passwords, financial information, or personal identifiers) is ever entered into public AI systems.”
Names its tools
“Examples of tools we may use include: OpenAI (ChatGPT), Fireflies.ai, Microsoft CoPilot, Gemini, Jasper, Grammarly, Adobe Stock AI Imagery, and others”
Tools named
ChatGPT, Fireflies.ai, Microsoft Copilot, Gemini, Jasper, Grammarly, Adobe Stock AI imagery
Marketing agency · United States · policy page · dated August 2026
“What we don't do is disguise generated work as artisanal, or bill AI-speed work at hand-speed rates.”
Says AI runs through all its delivery, keeps client data out of consumer AI tools, and ships every client AI agent with an owner, scope and logs.
Human review
Tells clients up front
Client data rule
Named owner
In its own words
Human review
“every deliverable passes human review before it ships to a client.”
Tells clients
“AI runs throughout our delivery: research, drafting, code, and the agents we operate internally. We describe this at capability level”
Client data rule
“Client confidential data is never used to train models, never pasted into consumer AI tools, and never leaves systems governed by the engagement's data terms.”
Marketing agency · United States · blog post · dated April 24, 2026
“We input client data only into those AI tools that are contractually required to protect data security and confidentiality and that do not use such data to train or improve their underlying models.”
Client data goes only into AI tools contractually barred from training on it, IP is discussed before lightly edited AI output, and deepfakes are ruled out.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“We always review AI-generated content and apply human professional judgment, editing and refinement before it is included in a final work product.”
Tells clients
“We are transparent about our use of AI technology with teammates, clients and partners when it meaningfully contributes to the product we create.”
Client data rule
“We input client data only into those AI tools that are contractually required to protect data security and confidentiality and that do not use such data to train or improve their underlying models.”
Says what AI never does
“We do not use AI to create deceptive or misleading content, including deepfakes or synthetic media that misrepresents individuals, organizations or events.”
Ownership and originality
“If a client requests or authorizes use of AI-generated content with limited modification, we will discuss intellectual property and ownership implications and ensure agreement before moving forward.”
Named owner
“We evaluate and monitor the AI tools we adopt with a cross-disciplinary team to ensure alignment with our standards for legal compliance, risk mitigation, environmental impact, intellectual property controls, data security and avoidance of bias.”
Client can limit AI
“We always comply with client agreements, policies and directives in using AI technology”
Marketing agency · United States · policy page · dated August 2023
“AI-generated content will never be published in its entirety without edits and/or restructuring by the Content team”
Discloses every AI-generated piece to the client and keeps AI out of strategy, audience work, brand messaging, keyword research and published images.
Human review
Tells clients up front
Client data rule
Says what AI never does
In its own words
Human review
“AI-generated content will never be published in its entirety without edits and/or restructuring by the Content team”
Tells clients
“Whenever AI is used to generate content for our clients, we will clearly disclose it to our clients ensuring that the origin of the content is transparent.”
Client data rule
“When using AI to process any sensitive data, we will obtain explicit consent from the individuals involved or ensure that the data is anonymized and cannot be traced back to specific individuals.”
Says what AI never does
“Generative AI may NOT be used for the following steps in the content development process; Strategic direction; Audience identification; Email sequence structure; Brand messaging”
Marketing agency · United States · PDF · dated June 2, 2025
“We believe that humans remain accountable for all decisions and actions, even when assisted by AI. The human must remain in the loop in all AI applications.”
Adapted from the Marketing AI Institute manifesto: AI used in four named areas under human quality control, and three applications ruled out.
Human review
Says what AI never does
In its own words
Human review
“We currently use AI in the following areas, with human guidance and quality control”
Says what AI never does
“We commit to never knowingly use generative AI technology to deceive; to produce content for the sole benefit of financial gain; or to spread falsehoods, misinformation, disinformation, or propaganda.”
Marketing agency · United Kingdom · policy page · undated
“We still invest the full allocated time into your marketing activity.”
Promises AI time savings go back into the client's retainer as deeper strategy work, not fewer hours; every AI-assisted piece human-approved. Names no tools.
Human review
Explains when a client asks
Client data rule
Says what AI never does
In its own words
Human review
“Every piece of AI-assisted work is reviewed, edited and approved by a real member of the Koala team before it reaches a client.”
Tells clients
“If you have any questions about how we use AI within our services, we’re always happy to have an open conversation.”
Client data rule
“We do not input sensitive client information, confidential business data or personally identifiable information into AI systems unless appropriate safeguards and approvals are in place.”
Says what AI never does
“all major decisions, strategies, recommendations and client communication remain fully human-led.”
“We don’t use AI to create final client deliverables without human review.”
An inbound agency's 'where we draw the line' list: no confidential data in unapproved tools, human review of deliverables, rights checked on AI imagery, people make hiring decisions.
Human review
Client data rule
Says what AI never does
Ownership and originality
In its own words
Human review
“We don’t use AI to create final client deliverables without human review.”
Client data rule
“We never upload confidential or proprietary information into unapproved AI tools.”
Says what AI never does
“We don't use AI to make final hiring, promotion, compensation, or other personnel decisions.”
Ownership and originality
“We always verify rights before using AI-generated imagery, video, or content.”
Marketing agency · United States · policy page · undated
“we will never use AI to fully generate creative or strategic products without substantial human input, refinement and oversight”
An agency-wide AI advisory group vets tools, client-facing work gets multi-level review, and AI never creates logos, visual systems or brand standards.
Human review
Client data rule
Says what AI never does
Ownership and originality
Named owner
In its own words
Human review
“Multi-level review processes for client-facing implementation”
Client data rule
“We prohibit sharing sensitive or proprietary information with public AI platforms.”
Says what AI never does
“We maintain strict protocols against using AI for: … Creating brand identity elements including logos, visual systems or brand standards”
Ownership and originality
“We also maintain that our work is our own, and we will never use AI to fully generate creative or strategic products without substantial human input, refinement and oversight.”
Named owner
“Formation of an agency-wide AI advisory group that evaluates emerging tech and use cases”
Marketing agency · United States · policy page · undated
“Across all teams, AI-generated deliverables and services are reviewed and approved by a subject matter expert for accuracy, cultural sensitivity, bias, and brand alignment before use.”
A B Corp's policy: a subject-matter expert approves every AI-assisted deliverable, leadership oversees AI use, and clients' own AI rules are followed.
Human review
Explains when a client asks
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“AI-generated deliverables and services are reviewed and approved by a subject matter expert for accuracy, cultural sensitivity, bias, and brand alignment before use.”
Tells clients
“We are open about our use of AI tools in our work and are willing to share when and how we use them.”
Ownership and originality
“We will not knowingly use AI generated content to infringe on the intellectual property rights of original content creators.”
Named owner
“Our leadership team oversees AI usage across the agency to ensure alignment with:”
Client can limit AI
“We endeavor to follow any AI policy requirements shared with us by clients.”
Marketing agency · United States · policy page · dated August 2026
“Portions of this content were developed with assistance from AI tools. Human reviewed.”
President is AI Oversight Lead; contracts state AI use and client opt-outs; standard 'Human reviewed' label; confidential data only in no-training environments; quarterly bias audit.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“For every use, we document the task, the data source, the human reviewer, and the verification step before any output reaches a client, partner, or member of the public”
Tells clients
“MPI discloses AI assistance in each proposal, scope of work, contract, deliverable, and external publication where AI materially shaped the content.”
Client data rule
“Client data and personally identifiable information are confined to MPI-contracted, professional-grade workspaces only”
Says what AI never does
“(2) generate visuals or audio that could be mistaken for authentic documentary evidence; or (3) use automated content engineered to manipulate emotional response.”
Ownership and originality
“clone or simulate the voice, likeness, or writing style of a real person without that person’s written, informed consent”
Named owner
“AI Oversight Lead (MPI President)”
Client can limit AI
“who owns the work, the human review that will happen, and when a client may opt out of specific AI-assisted tasks.”
Marketing agency · New Zealand · policy page · undated
“Clients may also ask us to limit or exclude AI assisted work on their account. We will confirm what is practical and note any effect on timeframes or cost.”
A table of where AI helps and who reviews it, the Director accountable, and a client opt-out that states the effect on time and cost.
Human review
Explains when a client asks
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“Client deliverables are reviewed by a person before release.”
Tells clients
“Clients may ask us how AI has been used on their work at any time.”
Client data rule
“We do not enter personal information, confidential client data or commercially sensitive material into public or consumer AI tools.”
Says what AI never does
“We do not use AI to fabricate results, reviews, testimonials or credentials.”
Ownership and originality
“We do not knowingly reproduce third party material without permission.”
Named owner
“Responsibility for AI use sits with the Director.”
Client can limit AI
“Clients may also ask us to limit or exclude AI assisted work on their account.”
Marketing agency · policy page · dated August 2024
“Podymos values transparency and client collaboration, and if we use AI for creative elements of a project, we will always seek our client's permission.”
Medical-device marketing agency: written consent before any client data enters AI, client permission before AI touches creative work, client owns all output.
Human review
Tells clients up front
Client data rule
Ownership and originality
Client can limit AI
In its own words
Human review
“All AI generated outputs are reviewed and validated by our internal team”
Tells clients
“if we use AI for creative elements of a project, we will always seek our client's permission.”
Client data rule
“No client data is shared with AI systems without explicit written consent”
Ownership and originality
“All creative content we produce is within the bounds of copyright laws, meaning that all the work we produce for you is owned by you.”
Client can limit AI
“if we use AI for creative elements of a project, we will always seek our client's permission.”
“AI-generated or AI-assisted work is reviewed, refined, and validated by rEvolution team members before it is shared with clients.”
A sports marketing agency: the team validates AI-assisted work, confidential data goes only into approved tools, AI use shared on reasonable request.
Human review
Explains when a client asks
Client data rule
Ownership and originality
Client can limit AI
In its own words
Human review
“AI-generated or AI-assisted work is reviewed, refined, and validated by rEvolution team members before it is shared with clients.”
Tells clients
“Upon reasonable request, rEvolution will share the general nature and extent of material AI usage related to the services we provide.”
Client data rule
“We do not input Client confidential information into AI tools unless the tool has appropriate data protection measures and has been approved for that use, either by the client or through a mutually agreed list of approved systems.”
Ownership and originality
“We take commercially reasonable steps to avoid knowingly infringing on third-party intellectual property, privacy, or usage rights when using AI tools.”
Client can limit AI
“has been approved for that use, either by the client or through a mutually agreed list of approved systems.”
“Particularly in the case of brand materials, thought leadership writings, and other substantive pieces of intellectual property, we will use AI tools only in preliminary stages of development”
Keeps AI to preliminary stages on brand and thought-leadership work, sorts work into copyright and copyright-free assets, and gives clients the final say.
Client data rule
Says what AI never does
Ownership and originality
Client can limit AI
In its own words
Client data rule
“Never enter personal, sensitive, or confidential information into any generative AI tools.”
Says what AI never does
“Particularly in the case of brand materials, thought leadership writings, and other substantive pieces of intellectual property, we will use AI tools only in preliminary stages of development”
Ownership and originality
“To keep everyone on the same page, we will categorize work as “copyright assets” vs. “copyright-free assets.””
Client can limit AI
“However, our clients always have final say how AI is used in the creation of their marketing assets.”
Marketing agency · United Kingdom · blog post · undated
“Sookio will never input sensitive information or private client data into generative AI tools.”
Announcement summarising a three-principle policy (quality, protection, privacy); the one concrete rule is never putting private client data into generative AI.
Client data rule
In its own words
Client data rule
“Sookio will never input sensitive information or private client data into generative AI tools.”
Marketing agency · United States · policy page · undated
“The human must remain in the loop in any AI-assisted applications.”
Publishes its AI tools by use case (ChatGPT, Writer, Canva, HubSpot, Moz) under principles adapted from the Marketing AI Institute's manifesto.
Human review
Tells clients up front
Names its tools
In its own words
Human review
“The human must remain in the loop in any AI-assisted applications.”
Tells clients
“Specifically, we’re using various AI tools for content ideation and style consistency as well as summaries, transcripts, or repurposes of our proprietary content.”
Marketing agency · United Kingdom · policy page · dated May 2026
“Our monthly reports note when AI-driven strategies (such as Smart Bidding or automated audience expansion) are in use.”
PPC agency flags Smart Bidding and other AI strategies in monthly reports, bans autonomous campaign changes, and reviews the policy quarterly.
Human review
Tells clients up front
Client data rule
Says what AI never does
Names its tools
In its own words
Human review
“Every AI-assisted decision at Visible Ads is reviewed and approved by a member of our team.”
Tells clients
“Our monthly reports note when AI-driven strategies (such as Smart Bidding or automated audience expansion) are in use.”
Client data rule
“Client campaign data is never shared with third-party AI training datasets.”
Says what AI never does
“We do not allow fully autonomous campaign changes without human review.”
Names its tools
“Automated bidding strategies and performance forecasting powered by platform-native AI (Google Smart Bidding, Microsoft auto-bidding) to maximise return on ad spend.”
Marketing agency · policy page · dated December 1, 2023
“We are committed to keeping “humans in the loop” for all AI-supported activities, combining tools’ capabilities with our seasoned industry professionals.”
Two lists, what AI will and will not do: research, outlines and transcripts yes, AI-only images, video or audio no. Shared under CC BY-SA.
Human review
Says what AI never does
In its own words
Human review
“We are committed to keeping “humans in the loop” for all AI-supported activities”
Says what AI never does
“Weidert Group will not use AI … for any image, video, or audio solely generated by AI”
Marketing agency · United States · policy page · dated July 20, 2026
“AI-generated output, whether it's a first draft, a data pull, or a piece of research, is a starting point, not a finished product.”
Agency that also sells its own AI suite (CARL); bars providers from training on client content, removes client data after engagements, discloses only when asked.
Human review
Explains when a client asks
Client data rule
Names its tools
In its own words
Human review
“Every piece of client-facing or published work that involves AI assistance goes through human review before it ships.”
Tells clients
“We'll tell you when and how AI was used on your project if you ask.”
Client data rule
“Where AI features rely on third-party AI providers to process content, we only send what's needed to complete that specific task, and we do not authorize those providers to use client content to train their models.”
Names its tools
“CARL Intelligence and the CARLverse suite (CARL AIO, CARL Writes, CARL Optimize, and CARL CMS)”
Tools named
CARL Intelligence, CARL AIO, CARL Writes, CARL Optimize, CARL CMS
Creative and branding agency · United States · policy page · dated October 2026
“Every AI-supported output, just like every human-generated one, undergoes a comprehensive review by one or more qualified Banner Day team members before being put to use or shared with our clients.”
A naming and branding studio's versioned policy: every AI-assisted output reviewed, the founder oversees AI governance, and client contracts checked for AI limits first.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“Every AI-supported output, just like every human-generated one, undergoes a comprehensive review by one or more qualified Banner Day team members before being put to use or shared with our clients.”
Tells clients
“From the first hello to the final high-five, we'll communicate clearly about how and why any AI tools we use contribute to a project's delivery and success.”
Client data rule
“Not inputting personally identifiable information or Proprietary Information into AI tools , unless we have documented permission from the information owner”
Says what AI never does
“Not using AI to filter, influence, or direct any decisions in which such bias may occur (for example, when hiring collaborators or staffing project teams)”
Ownership and originality
“For any deliverable where AI use raises a meaningful IP consideration, we'll raise it directly with our clients before proceeding.”
Named owner
“our AI governance is overseen directly by Banner Day's founder and creative director with input from key collaborators.”
Client can limit AI
“checking whether a client's contract or SOW includes AI-specific requirements or restrictions before inputting anonymized project details into an AI tool”
Creative and branding agency · Canada · policy page · dated August 3, 2026
“Evolving Media & Design does not use AI to generate or create designs or branding elements; we may use AI tools to brainstorm the creation of slogans.”
Canadian RGD-certified design studio posting a policy its association requires; logos and branding are hand-made, never AI-generated; clients told when AI creates elements.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Client can limit AI
In its own words
Human review
“Our team vets, edits, and approves all AI outputs to ensure quality and alignment with client expectations.”
Tells clients
“If AI is used to create elements of a deliverable, we inform clients and provide assurances that the output complies with intellectual property laws.”
Client data rule
“Any client-provided materials (e.g., logos, brand guidelines, proprietary content) are handled with strict confidentiality and are not used to train AI models or shared with third-party platforms.”
Says what AI never does
“Evolving Media & Design does not use AI to generate or create designs or branding elements; we may use AI tools to brainstorm the creation of slogans.”
Ownership and originality
“We cross-reference outputs with reputable databases and tools to confirm originality.”
Client can limit AI
“We welcome client feedback on AI usage and adjust our practices to meet their preferences or concerns on a per-client basis.”
Creative and branding agency · United States · policy page · dated January 1, 2025
“We take full responsibility for the accuracy and quality of anything we deliver, regardless of what tools were used in the process.”
Names the person who approves all copy, takes full responsibility whatever the tools, and asks consent before client data goes into AI.
Human review
Explains when a client asks
Client data rule
Says what AI never does
Named owner
In its own words
Human review
“All copy is reviewed, edited, and approved by Tony Salerno before it is presented to or deployed for a client.”
Tells clients
“If you ever have questions about a specific deliverable, we are happy to discuss how it was produced.”
Client data rule
“We do not submit identifiable client data, proprietary business information, or confidential materials into AI systems without explicit client knowledge and consent.”
Says what AI never does
“AI is a tool, not a creative director. The following are always handled by Tony Salerno and our team directly:”
Named owner
“All copy is reviewed, edited, and approved by Tony Salerno before it is presented to or deployed for a client.”
PR and communications agency · United States · blog post · dated September 17, 2023
“If a client has an existing policy surrounding AI, we will work together to respect the client’s policy so both parties agree with potential outputs.”
Seven short principles from a Florida PR firm: no confidential client contact data in collective AI without permission, and clients' own AI policies are respected.
Client data rule
Says what AI never does
Client can limit AI
In its own words
Client data rule
“Without permission, we will not enter confidential or personal client contact information into any collective AI system.”
Says what AI never does
“We will never use generative AI technology to deceive or spread misinformation or propaganda.”
Client can limit AI
“If a client has an existing policy surrounding AI, we will work together to respect the client’s policy so both parties agree with potential outputs.”
PR and communications agency · South Africa · policy page · undated
“Do not use generative AI tools to create imagery, likenesses, or avatars that create the appearance of diversity instead of working with diverse talent.”
Staff rulebook adapted from PR Council guidance: no AI images as final client assets, no AI translation, no voice mimicry; disclosure only to line managers.
Human review
Client data rule
Says what AI never does
Ownership and originality
In its own words
Human review
“Utilise diverse perspectives within the agency to review content created by generative AI tools to ensure no bias is accidentally overlooked or shared externally.”
Client data rule
“Use caution when putting confidential client information into a generative AI tool or platform. For example, do not use AI to create the first draft of a press release about a new product”
Says what AI never does
“Do not use generative AI images as final creative assets for client campaigns.”
Ownership and originality
“Always check for inadvertent plagiarism, copyright infringement, or trademark infringement in AI-generated output.”
PR and communications agency · United States · policy page · undated
“To ensure the originality of generative AI work products, Gregory submits machine-generated content through anti-plagiarism software to identify issues.”
PR firm runs AI output through plagiarism software, anonymizes client data, opts out of ChatGPT training, and keeps copyright-critical content human-made.
Client data rule
Says what AI never does
Ownership and originality
Client can limit AI
Names its tools
In its own words
Client data rule
“Preventing confidential or private client content to be entered into AI prompts.”
Says what AI never does
“We ask clients to inform us of content that requires copyright protection so that we can ensure it is not machine-generated.”
Ownership and originality
“To ensure the originality of generative AI work products, Gregory submits machine-generated content through anti-plagiarism software to identify issues.”
Client can limit AI
“If clients prefer that we not use generative AI, we are agreeable to that decision with the understanding that it may impact the SOW, pricing, and delivery timelines.”
Names its tools
“Opting out of allowing ChatGPT to use our prompts in training future models.”
PR and communications agency · United Kingdom · policy page · undated
“We will not claim AI-generated work as our own, and use of AI for client work will always be discussed and agreed with our clients.”
AI on client work is always agreed with the client first, new AI tools are vetted for data, IP and liability, and senior leadership owns the policy.
Human review
Tells clients up front
Client data rule
Says what AI never does
Named owner
Client can limit AI
In its own words
Human review
“Employees are expected to review, validate, and, where necessary, edit AI-generated content to ensure it aligns with client needs, company standards, and industry best practices.”
Tells clients
“In situations where there is a risk our use of AI might be seen to be misleading, or where agreed with a client, we will disclose the use of AI alongside the generated content.”
Client data rule
“Employees must avoid inputting sensitive, confidential, or personal client data into AI tools unless explicit consent is obtained and appropriate safeguards are in place.”
Says what AI never does
“JBP Associates commits to avoiding the use of AI for unethical purposes, including the creation of deceptive, harmful, or manipulative content.”
Named owner
“The senior leadership team is responsible for overseeing the implementation of this policy and ensuring compliance.”
Client can limit AI
“use of AI for client work will always be discussed and agreed with our clients.”
PR and communications agency · policy page · undated
“We disclose when and how AI is used in client work, and every PAN client receives our internal AI policy and use guidelines.”
PR agency's five-pillar 'PAN Pact': every client receives its internal AI policy, NDAs extend to generative AI, client materials stay in enterprise tools.
Human review
Tells clients up front
Client data rule
Names its tools
In its own words
Human review
“every strategic recommendation and deliverable is ultimately created, approved, and owned by a human.”
Tells clients
“We disclose when and how AI is used in client work, and every PAN client receives our internal AI policy and use guidelines.”
Client data rule
“We do not upload client materials into public models. All usage is routed through secure, enterprise-grade platforms (e.g., ChatGPT Enterprise) that do not retain or train on client data.”
Names its tools
“We use tools like ChatGPT Enterprise and Wordtune to generate early outlines for blogs, POVs, emails, and CTAs.”
PR and communications agency · policy page · dated November 2025
“No AI system may be used autonomously to make strategic, editorial, or reputational decisions without human judgment and approval.”
A PR firm's tenets: no autonomous AI decisions on strategy, editorial or reputation, no confidential data in external AI systems, no deepfakes.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
In its own words
Human review
“Every AI-assisted output is reviewed and validated by our team.”
Tells clients
“We will be transparent about the use of AI tools in our PR activities, clearly disclosing when AI is employed in content generation, data analysis, or decision-making processes.”
Client data rule
“We do not input confidential, proprietary, or identifying data into AI systems that store or share such information externally.”
Says what AI never does
“We reject any use of AI that may mislead, manipulate, or harm individuals or communities, including the creation or distribution of deepfakes, false information, or deceptive content.”
Ownership and originality
“Rosica never knowingly uses AI-generated material that infringes on third-party rights or misrepresents authorship.”
PR and communications agency · United Kingdom · policy page · dated October 2026
“We disclose AI use to clients and seek consent. Where content is AI-generated, such as synthetic media, we disclose that to audiences too. Where AI features in a project, it’s recorded in project documentation and surfaced through routine reporting.”
UK comms agency: client consent plus audience disclosure of synthetic media, AI use logged in project documentation, EU AI Act deployer framing, environmental-cost test.
Human review
Tells clients up front
Client data rule
Says what AI never does
Client can limit AI
In its own words
Human review
“Editorial control stays with people. AI can be useful for drafting, but it doesn’t decide what we say.”
Tells clients
“We disclose AI use to clients and seek consent. Where content is AI-generated, such as synthetic media, we disclose that to audiences too.”
Client data rule
“We use tools and settings that keep client or project data out of model training pools.”
Says what AI never does
“We don’t publish AI-only outputs as finished deliverables.”
SEO and content agency · policy page · dated September 2026
“We may use AI to aid in research, brainstorming, outlining, and analyzing, but from the first draft to the final deliverable, human creativity and oversight will always be involved.”
A healthcare content agency: AI for research, ideas and outlines, human editors have the final review, plagiarism and AI-detection checks, no proprietary data in open AI.
Human review
Client data rule
Ownership and originality
Names its tools
In its own words
Human review
“Human editors always have the final review to ensure accuracy, tone, and engagement.”
Client data rule
“Our policy prohibits uploading proprietary data to open AI systems to prevent unauthorized training on sensitive content.”
Ownership and originality
“We run plagiarism and AI detection checks to ensure originality.”
Names its tools
“we originally outlined and reviewed this policy with help from ChatGPT-4 Teams”
SEO and content agency · United States · blog post · undated
“For the duration of the work, your guidelines become our guidelines.”
Content agency's ethics code: names every AI tool used if asked, gets permission before proprietary data meets a model, and adopts each client's AI rules.
Human review
Explains when a client asks
Client data rule
Ownership and originality
Client can limit AI
In its own words
Human review
“A person reviews every output of our AI tools before it reaches you. Nothing ships to a client straight from a model.”
Tells clients
“So if you want to know what we used to make something, just ask. We will tell you, and that includes every AI tool involved.”
Client data rule
“When a project involves proprietary information, we confirm your permission before it goes near a model.”
Ownership and originality
“We use AI tools that generate work from approved references or from scratch. We also attribute human and AI contributions whenever that matters.”
Client can limit AI
“We adapt to whatever level of AI involvement fits your policies and your comfort.”
SEO and content agency · Australia · policy page · dated October 30, 2025
“Mahlab respects the rights of artists by never using AI to recreate the style or approach of an existing artist.”
An Australian content agency's policy mapped to government AI principles: client consent before client data enters AI, no imitation of an artist's style.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“Mahlab ensures that AI-generated content and output is fact-checked, verified, reviewed and edited to ensure relevance, originality and compliance before dissemination.”
Tells clients
“Mahlab uses AI in all aspects of our business and our service delivery.”
Client data rule
“Mahlab does not share personal information (as defined in the Privacy Act 1988 (Cth)), or confidential information, into any AI model unless we are certain that data will not be used for training purposes.”
Says what AI never does
“Mahlab respects the rights of artists by never using AI to recreate the style or approach of an existing artist.”
Ownership and originality
“Mahlab ensures that all services and deliverables have significant independent intellectual effort and authorship by human team members.”
Named owner
“Mahlab has appropriate leaders with specific obligations for safe and responsible development and use of AI across our organisation.”
Client can limit AI
“Mahlab will obtain explicit consent from clients before using client proprietary data in AI systems.”
“Each client will complete an Acceptable Generative AI Use Questionnaire which will specify allowed uses”
Writing guidelines for an SEO agency: a per-client AI-use questionnaire, clients can ban AI, and AI never writes an entire first draft.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Client can limit AI
Names its tools
In its own words
Human review
“We will review, fact-check, and revise all AI output we submit for publishing”
Tells clients
“if AI was used heavily to produce a piece of content, (assuming the client agrees), it will be publicly disclosed alongside or within the published content piece”
Client data rule
“We will not share proprietary or sensitive client or Netconcepts information with AI tools.”
Says what AI never does
“We will not use AI to provide an entire “first draft.””
Ownership and originality
“the writer is expected to be the writer, not a prompt engineer.”
Client can limit AI
“We will comply with a client's request to prohibit or restrict the use of AI and not employ AI tools when working for that client.”
Names its tools
“We use the paid versions of ChatGPT and Claude for higher-quality AI output.”
SEO and content agency · United States · policy page · undated
“At NgageContent, we specialize in writing human-focused copy designed to encourage conversions for our clients. We will never use AI for a final deliverable as it has multiple limitations.”
Says it never uses AI for a final deliverable, then lists where AI helps and the tools behind it; principles adapted from the Marketing AI Institute.
Human review
Says what AI never does
Names its tools
In its own words
Human review
“The human touch must be present in all finished work completed for a client.”
Says what AI never does
“We will never use AI for a final deliverable as it has multiple limitations”
“AI is strictly implemented as a tool with intentional input from our clients' brands, established strategies, and marketing objectives. All AI-generated output undergoes thorough team review.”
A short do and do-not list: no AI testimonials, logos or full campaigns, and every AI output reviewed by the team.
Human review
Client data rule
Says what AI never does
In its own words
Human review
“All AI-generated output undergoes thorough team review.”
Client data rule
“How We Do NOT Use AI … Share data or any sensitive company information”
“We have a brilliant editorial team and any written content we deliver to clients will come from them, not ChatGPT.”
Written deliverables come from its editorial team, not AI; anything AI-generated gets source and brand checks; no confidential information goes into AI tools.
Human review
Tells clients up front
Client data rule
Says what AI never does
In its own words
Human review
“Anything generated using AI (at any point in our production process) will undergo thorough human review and editing, such as source-checking, brand guidelines alignment and quality assessment.”
Tells clients
“We believe in demystifying AI magic by providing our clients with a clear understanding of how these technologies enhance our collaboration and how we're planning to use them on a project.”
Client data rule
“We don't input any confidential information into AI tools and only use tools with strong privacy and security controls.”
Says what AI never does
“We have a brilliant editorial team and any written content we deliver to clients will come from them, not ChatGPT.”
Social media agency · United States · blog post · undated
“If you used AI in a process, you own the output. That means you reviewed it, approved it, and are accountable for it.”
Discloses AI when it materially shaped a deliverable rather than for routine features, and asks five questions before any new AI use.
Human review
Tells clients up front
Client data rule
Client can limit AI
In its own words
Human review
“If you used AI in a process, you own the output. That means you reviewed it, approved it, and are accountable for it.”
Tells clients
“When AI materially shaped a deliverable (AI-generated imagery in a social post, conclusions an AI reached after analyzing data, a concept that came out of an AI ideation session) we tell clients upfront.”
Client data rule
“never input client confidential information into any AI tool without first knowing which version you’re using, verifying its data policy, and opting out of training where available.”
Client can limit AI
“Clients have different tolerances for AI use. Some have explicit policies. Some have strong preferences. Some are enthusiastic. We honor wherever they land, not wherever we’d prefer they land.”
Web and digital agency · policy page · dated June 20, 2025
“For example, when Fathom is used to record and summarize Zoom calls, every call participant should be verbally informed of this. Do not rely on the Zoom notifications to ensure that everyone is notified.”
Staff rules: novel AI uses need leadership sign-off, AI note-takers are announced aloud on calls, and client source data needs explicit approval first.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Names its tools
In its own words
Human review
“AI tools are a starting point, not an endpoint. Always review and refine AI-generated outputs before sharing internally or externally.”
Tells clients
“When AI is used in the work setting, that should be clearly communicated to coworkers and clients.”
Client data rule
“uploading any client "source" data, such as client lists, project files, or confidential communications, requires an internal conversation and explicit approval before being inputted into an AI tool.”
Says what AI never does
“Generating the likeness of real people without consent”
Ownership and originality
“Do not generate the visual likeness of brands or individuals (e.g., celebrities, public figures, clients) without their explicit consent.”
Named owner
“If it's a new or novel application, check in with 4Site leadership first.”
Names its tools
“If the use of AI in a deliverable is something we've done before (e.g., using ChatGPT to help draft copy), feel free to use it.”
Web and digital agency · United Kingdom · policy page · dated May 3, 2026
“New AI tools require client approval before being added, on the same 30-day notice / 14-day objection cycle as any other sub-processor”
Development agency naming Claude, Claude Code and DeepSeek; AI tools are client-approved sub-processors, production data anonymised first, read-only AI credentials, opt-out via SOW.
Human review
Tells clients up front
Client data rule
Says what AI never does
Client can limit AI
Names its tools
In its own words
Human review
“All AI-generated code, content, configuration, or analysis is reviewed by a qualified team member before being incorporated into client deliverables, deployed to production, or shared with the client.”
Tells clients
“We disclose them as part of our standard sub-processor disclosure”
Client data rule
“No production personal data to AI APIs without prior anonymisation”
Says what AI never does
“AI tools do not make decisions affecting client projects, data, contracts, or commercial relationships.”
Client can limit AI
“If a client engagement requires that BuiltByGo not use AI tools at all in connection with their work, this can be specified in the Statement of Work and we will operate without AI assistance for that engagement.”
Names its tools
“named specific AI tools (Claude / Claude Code / DeepSeek)”
Web and digital agency · United Kingdom · policy page · undated
“We also appreciate that AI, like many things in digital, requires processing power. Therefore, we avoid using it unnecessarily in order to reduce our consumption and subsequent environmental impact.”
Three principles with a bias checklist for prompts and outputs, ISO 27001 data handling, and a rule against using AI where it adds nothing.
Explains when a client asks
In its own words
Tells clients
“If you have any questions related to our use of AI please reach out to your Crafted contact”
Web and digital agency · policy page · dated March 26, 2026
“Design Powers does not use GenAI to build websites, design logos, or graphics for client projects.”
Tool-by-tool inventory per channel (website, social, client work); no AI for client sites, logos or graphics; AI imagery used once, with explicit client consent.
Human review
Says what AI never does
Names its tools
In its own words
Human review
“all GenAI-generated content is thoroughly edited to ensure it reflects our expertise and point of view”
Says what AI never does
“Design Powers does not use GenAI to build websites, design logos, or graphics for client projects.”
Names its tools
“We leverage Claude (formerly used ChatGPT ) to draft initial versions of blog posts and newsletters”
Tools named
Claude, Gemini, Grammarly AI Writing Assistant, Ubersuggest, SEMrush, Adobe Firefly
Web and digital agency · United Kingdom · policy page · undated
“If you’d like us to avoid AI on your project, tell your project manager and we’ll agree on an approach.”
Clients can ask for no AI on their project, drafts are human-approved before use, and AI is never the final author of client work.
Human review
Explains when a client asks
Client data rule
Says what AI never does
Client can limit AI
In its own words
Human review
“Creating first drafts or outlines for content - always reviewed, edited and approved by a human before use”
Tells clients
“If you want to know where AI has been used, ask - we’ll be transparent.”
Client data rule
“What we don’t do … We don’t upload confidential client information into public AI tools unless we have a clear, agreed process and the right safeguards in place”
Says what AI never does
“What we don’t do … We don’t use AI as the final author of client-facing work without human review and accountability”
Client can limit AI
“If you’d like us to avoid AI on your project, tell your project manager and we’ll agree on an approach.”
“We use AI as a tool to research and understand, not to mass-produce low-quality content.”
Principles list plus AI for niche, persona and buyer-journey research, with human-edited output aimed at Google E-E-A-T. Silent on data, tools, disclosure.
Human review
Says what AI never does
In its own words
Human review
“Our final output is always heavily edited and refined by human experts to ensure it meets the standards of E-E-A-T”
Says what AI never does
“We use AI as a tool to research and understand, not to mass-produce low-quality content.”
Web and digital agency · Australia · policy page · dated April 28, 2026
“We tell clients when AI contributes materially to deliverables, explain the role it played, and identify the human review and sign-off process.”
A Drupal and GovCMS agency aligned to Australian Government AI guidance: the CTO owns an AI use-case register, and AI-assisted code gets licence checks.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
In its own words
Human review
“Every AI-assisted output that changes code, content, analysis, advice, or client deliverables must be reviewed by an appropriately skilled person before release.”
Tells clients
“We tell clients when AI contributes materially to deliverables, explain the role it played, and identify the human review and sign-off process.”
Client data rule
“We avoid entering personal, confidential, commercially sensitive, privileged, or secret material into public AI services.”
Says what AI never does
“Prohibited uses include uploading secrets or unapproved personal data to public tools, bypassing security review, deploying autonomous production changes without approval gates, impersonating humans, or using AI to make high-impact decisions without prior assessment and approval.”
Ownership and originality
“AI-assisted contributions are reviewed for licence compatibility, attribution needs, copied code, and client IP obligations before release.”
Named owner
“The Chief Technology Officer (CTO) owns AI governance, maintains the AI system and use case register, and reports material risks or incidents to the leadership team.”
Production company · United Kingdom · policy page · undated
“We strive to use services which have been trained on licensed or synthetic data and will only use such services for any elements used within final productions.”
Animation studio keeps storytelling, performance and artistic vision human, and uses only AI trained on licensed or synthetic data in final productions.
Human review
Says what AI never does
Ownership and originality
In its own words
Human review
“All key moments within a production will be reviewed by humans.”
Says what AI never does
“For us, this means storytelling, performance and artistic vision remain in the hands of humans.”
Ownership and originality
“We strive to use services which have been trained on licensed or synthetic data and will only use such services for any elements used within final productions.”
Production company · policy page · dated September 29, 2026
“We do not generate speech that a person we interviewed did not say.”
A video producer that never generates words an interviewee did not say, raises any AI content with the client up front, and human-checks AI captions.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Client can limit AI
In its own words
Human review
“We will use AI-captioning in our videos with a human check on the captions to ensure their quality.”
Tells clients
“we will communicate up front with our clients about their own AI policies and any AI-generated content being considered for a video project”
Client data rule
“To respect that and stop AI applications from using such proprietary information, we will turn off such “use my information to train the model” features.”
Says what AI never does
“We have no plans to use AI video editing software or apps”
Ownership and originality
“We do not generate speech that a person we interviewed did not say.”
Client can limit AI
“we will communicate up front with our clients about their own AI policies and any AI-generated content being considered for a video project, so that they can have agency in the decision”
Production company · United Kingdom · policy page · undated
“A named member of our team is responsible for AI use on every project.”
A post-production house: consent first for synthetic voices and faces, one named person accountable per project, and clients can opt out of AI entirely.
Human review
Tells clients up front
Client data rule
Says what AI never does
Ownership and originality
Named owner
Client can limit AI
In its own words
Human review
“Every frame, edit and mix that leaves our facility is checked and signed off by a person.”
Tells clients
“we will tell you what was used and why, so you can make informed decisions, including any disclosures you may owe to your own audiences, broadcasters or platforms.”
Client data rule
“We will never upload client footage, audio, scripts, unreleased content or any confidential project material to public or consumer generative AI tools, where data may be retained or used to train models.”
Says what AI never does
“We do not produce deceptive synthetic media”
Ownership and originality
“We will not create synthetic voices, faces, performances or digital doubles, or alter a performer's likeness using AI, without explicit, informed and documented consent from the individual or their representatives”
Named owner
“A named member of our team is responsible for AI use on every project.”
Client can limit AI
“If you would prefer no generative AI be used on your project at all, simply let us know and we will work accordingly.”
Nearly all promise review. Few say who answers for it.
The commitment agencies make most is human review (61 of 68), then client data rule (54) and tells clients (51). The rarer commitments are the ones a client can check: 18 name the person or group accountable for AI use, 26 let a client limit or exclude AI on their work, and 21 name the tools they use. Only 34 of 68 put a date on the policy, so a reader cannot tell whether the rest still describe how the agency works.
Telling clients splits two ways: 37 agencies say they disclose AI use without being asked (labelling work or telling clients up front), and 14 promise a full answer whenever a client asks. Both are honest positions; what matters is saying which one you take. Every one of those statements, grouped by when it applies, is in AI disclosure statement examples, and how agencies put AI into the contract itself is in AI contract clause examples. Labeling content for its audience is set by the platforms and the law: AI disclosure requirements by platform and law.
The AI disclosure checker reads your homepage, legal pages and policy addresses the way a client would and quotes what it finds.
Is your agency's policy missing?
Send the link to hello@bigpipe.agency and we will read it and add it. The same address takes corrections if your policy has changed.
How we built the library
We searched the web for AI policies that agencies publish about their own work: dedicated policy pages, AI sections of legal pages, policy PDFs and the occasional blog post that publishes the policy itself. We kept 68 from 7 kinds of agency where the agency states how it uses AI on client work, and read each in full (October 25 to 30, 2026). A commitment is ticked only when the policy states it as the agency's own practice; values with no practice behind them, advice to others and templates are not counted. Every tick carries the agency's exact words, under “In its own words” on each card.
This is a library of agencies that chose to publish, not a sample of all agencies: most agencies have no public AI policy at all. For how many publish one, and for the holding groups (WPP, Publicis, Dentsu, Havas, Monks), see our audit of 104 small agency websites.
Questions agencies ask
What should an agency AI policy include?
Judging by the 68 agencies here, the core is human review, client data rule, tells clients: 90%, 79% and 75% of the policies commit to them. Fewer name an accountable owner (26%), let clients limit AI on their work (38%) or name the tools they use (31%), which is where a policy can stand out.
Should an agency publish its AI policy?
Clients and procurement teams increasingly ask how an agency uses AI before they sign. A public page answers that before the call, gives you a link to put in proposals, and holds you to what it says. Most small agencies have not published one yet: in our audit of 104 small US and UK agencies, 2 had an AI policy page.
Is there a free AI policy template for agencies?
Yes. The free AI terms generator writes a client AI statement and contract clauses from where your agency actually uses AI, and the same documents are available as a Notion template and as open templates on GitHub under CC BY 4.0.
How did you choose these policies?
We searched the web for AI policies that agencies publish about their own work, read each one in full (October 25 to 30, 2026, the date of each reading is in the CSV), and kept those where an agency states how it uses AI on client work. A commitment is ticked only when the policy states it as the agency's own practice; the quote under each tick is the agency's own wording.
Data: the library as CSV (68 policies, one column per commitment and the date each was read). Our coding of each policy is free to reuse under CC BY 4.0 with a link to this page; the quoted words belong to each agency and are quoted to show what its policy says. Policies change: each card links to the original.
Agency AI Policies: 68 Published Policies Compared | BigPipe
