Lesson objective

Diagnose where responsibility for AI-assisted work currently rests inside an agency, and separate responsibility, accountability and authority for claims, endorsements, email and the agency's own statements.

Plain-language explanation

Why this matters now

Level 3 built AI-assisted agency operations: a mapped workflow, a risk tier for each kind of work, an SOP a reviewer can enforce, a quality sample and a stop condition. Each control had an owner at the level of the operation. This level moves up a layer. The question is no longer "is this content workflow sound?" It is the question a client's general counsel, the FTC, a state attorney general, a National Advertising Division challenger, an insurer or a journalist will eventually put to the people who run the agency: who here is answerable for how AI is used, and what would you show me?

Agencies often answer that question with "the client approves everything." The sources in this module do not support that answer.

  • Professional guidance (FTC, "Advertising FAQ's: A Guide for Small Business", dated April 2001, last edited January 2025; staff guidance, not binding): on ad agency liability, the FTC looks at "the extent of the agency's participation in the preparation of the challenged ad" and "whether the agency knew or should have known that the ad included false or deceptive claims." The guide says plainly that "the advertising agency also may be held legally responsible for misleading claims in ads."
  • Professional guidance (FTC Policy Statement Regarding Advertising Substantiation, 23 November 1984; describes the FTC's enforcement position under section 5): the statement reaffirms that "advertisers and ad agencies have a reasonable basis for advertising claims before they are disseminated." Agencies are named in the sentence, not added by inference.

Neither document mentions AI. The guide was written in 2001; the policy statement in 1984. They matter because AI changes how much an agency participates in making a claim, not whether the old tests apply. When a studio uses a model to generate forty headline variants, the agency is writing more claims, faster, than it ever wrote by hand.

This programme's register holds no figure for how many agencies or PR firms use AI. There is none to cite, and none is needed. The case for naming owners rests on exposure that already exists.

A question to sit with

If a client's general counsel asked on Monday who at the agency approved the tool that drafted last month's product claims, and what that person checked, would the honest answer be a name, a department, the freelancer, the client, or a silence?

What goes wrong

Everyone points at the client. The account team says the client approved the copy. The client approved what it was shown, which did not include the fact that a model had added a performance figure nobody supplied. The FAQ's test asks about the agency's participation and what it knew or should have known. Client sign-off does not answer either question.

The studio thinks the account team checked; the account team thinks the studio did. Creative produced the variants. Account management sent them. Each assumed the other read them against the claims file. Both did part of the work, and nobody made the decision.

Freelancers sit outside the system. A freelance copywriter uses a personal AI account and delivers finished copy. The agency's SOP never reached them. The agency still disseminates what they wrote.

The tool arrives without a decision. A generative feature switches on inside the design suite, the social scheduler or the email platform through an update. Nobody assessed it, because nobody bought anything.

The principal signs the invoice and nothing else. A subscription approval is treated as a governance approval. Nobody approved which work the tool may touch, who reviews its output, or what happens when it is wrong.

One person wears four hats. In a 25-person agency the managing director may also be head of new business, the person who buys software and the last reader of every major claim. That can work, but only if the agency writes down which role each decision was made in and who checks the decisions that person makes about their own work.

The practical answer

Separate three words that agencies blur.

Word Meaning in this module Who can hold it
Responsibility Doing a piece of the work properly Anyone doing the work, including freelancers, the studio and a vendor
Accountability Answering, by name, for the outcome A named person, competent for the judgement involved
Authority Saying yes or no and having it stick A named person or role, written into agency governance

Most agencies have responsibility everywhere, accountability nowhere and authority with whoever is closest to the client. Governance puts accountability and authority in the same named place and writes it down.

Map the diffusion before you fix it. Trace one AI-assisted deliverable through the agency and note who touched it.

Stage Typical holder What they usually assume
Brief and claims file Client and strategist "The claims file is complete"
Drafting and generation Copywriter, studio, freelancer, tool "Someone will check the facts"
Internal review Creative director, account lead "Creative checked the claims"
Client approval Client marketing team "The agency knows the rules"
Dissemination Media, social, email, PR "It was approved, so it is fine"

Five stages and five assumptions. Each assumption is reasonable on its own; together they leave a claim with no owner.

Match the owner to the kind of exposure. The next lesson sets out each source in detail. Here the point is that the accountable person depends on what is at stake.

Exposure Who at the agency owns it Where the lesson treats it
Claims in ads the agency helps make Named account or creative lead for the account Lesson 2
Endorsements, creators and reviews Named lead for the creator and reviews programme Lesson 2
Native and sponsored content Named lead for content and media placement Lesson 2
Commercial email the agency sends Named lead for CRM and email operations Lesson 2
What the agency tells clients about its own AI services A principal or named new-business owner Lesson 3
The AI programme as a whole A named principal with written authority Lesson 4, Module 02

Six rows, six owners. A small agency may put three rows on one person. That is allowed; leaving a row blank is not.

Apply four tests to anyone proposed as accountable. These tests are an AI Coalition Network recommendation, not a legal requirement.

  1. Named. A person, not a department or a committee. The charter describes the role; daily operation needs a name and a deputy.
  2. Competent for the judgement. Whether an AI-drafted claim is supported is a claims judgement. Whether a creator post needs a clearer disclosure is an endorsement judgement. Put each judgement with someone who can make it.
  3. Authorised. The person can refuse a tool, pause a use and hold a deliverable over an account lead's or a client's deadline, within a written scope the principals agreed.
  4. Resourced. Time, budget and a deputy for launch weeks. A title with no hours attached is not accountability.

Use a member code only as far as it reaches. Professional guidance (PRSA Board of Ethics and Professional Standards, ESA #22, "AI Agents in Public Relations", dated July 2026; advisory guidance for PRSA members, not law): "AI systems are tools, not ethical actors. Public relations professionals are accountable for how the AI tools they use are designed, deployed, and supervised." That binds no one outside PRSA's membership. It is still a useful statement of where accountability sits, and an agency whose PR staff are members should expect them to act on it.

Remember this

  • An agency spreads AI work across client, strategist, studio, freelancers, account team and tools, and that spreads decisions until nobody holds them.
  • The FTC's small-business FAQ looks at the agency's participation and whether it knew or should have known a claim was false; client approval answers neither question.
  • The 1984 substantiation policy names ad agencies alongside advertisers in the reasonable-basis requirement.
  • Responsibility, accountability and authority differ; governance puts the last two with one named, competent person.
  • A member code such as PRSA's ESA #22 binds its members only, but it says plainly that people, not tools, are accountable.

Marketing agency example

Northlight Creative Co. is a fictional 32-person agency. Its client Marrowfield Organics, a fictional snack brand, asks why a paid social ad said its granola bar had "40% more protein than leading brands." Nobody at Marrowfield supplied that figure. Managing director Helen Asante traces the ad. The strategist wrote a brief and attached the claims file, which contains no comparative protein claim. A freelance copywriter, Tom Reyes, used his own AI account to generate thirty headline variants and delivered twelve. The studio laid out six of them. The account lead, Priya Mensah, sent the six to Marrowfield, assuming the creative director had checked them against the claims file. The creative director says he reviews tone and brand fit, not claims. Marrowfield's marketing manager approved the ads because "the agency knows the rules." The media buyer launched them. Helen lays the answers side by side: six people and one tool touched the ad, and each thought someone else owned the claim. She runs the three-desk test. The copywriter names the account lead, the account lead names the creative director, and the designer says "the client." Helen pulls the ad that morning, tells Marrowfield honestly how the figure arrived, and does not describe it as the client's error. At the principals' meeting the longest argument is not about the tool. It is whether a named claims reviewer may hold a deliverable over an account lead's launch date. The principals decide yes, in writing, with appeal to Helen. Priya is named as claims reviewer for the Marrowfield account with a deputy for launch weeks, freelancers are required to disclose any AI use on delivery, and Helen is named owner of the AI programme as a whole.

Responsible-use warning

Do not answer "who is accountable for AI here?" with "the client approves everything." The FTC's staff guide looks at the agency's own participation and at whether it knew or should have known a claim was false or deceptive, and the 1984 policy statement expects agencies as well as advertisers to hold a reasonable basis before a claim is disseminated. Client approval is a step in the process, not a transfer of the agency's exposure. Name the competent person at the agency, give them authority and a deputy, and record it.

Human-review requirement

The accountable person depends on the exposure: a named account or creative lead for claims on each account; a named lead for creators and reviews; a named lead for native placements; a named CRM lead for commercial email; a principal for what the agency tells clients about its own AI services; and a named principal for the programme as a whole. That programme owner checks that the owners do not contradict one another, records each decision and its basis, and never accepts a department, a freelancer, a vendor or the client as the owner.

Practical exercise

Use a fictional agency, or your own with nothing confidential.

  1. Pick one recent AI-assisted deliverable, such as a paid social set, a press release or an email sequence. List every person, freelancer, tool and client contact who touched it.
  2. For each, write what they assumed someone else had checked. Use their words if you can get them.
  3. Mark each entry R (does the work), A (answers by name) or Au (can say no and have it stick). Note entries with authority but no accountability, and the reverse.
  4. Run the three-desk test with a copywriter, an account lead and a designer, and record each answer verbatim.
  5. For each of the six exposure rows in the lesson table, write the name of the person who owns it today, or "blank".
  6. List every AI feature that arrived through a software update in the last year and note who, if anyone, decided how it may be used.
  7. Draft the four-test profile (named, competent, authorised, resourced) for the claims reviewer on your largest account. Note which test your principals would find hardest to agree.
  8. Write one sentence for the principals to decide on, naming a role and its authority to hold a deliverable over a launch date.
  9. List what that person would need to answer a client's general counsel with records rather than recollection.

Knowledge check

  1. What two things does the FTC's small-business advertising FAQ say the FTC looks at when deciding whether an ad agency is liable?

    Answer guidance: The extent of the agency's participation in preparing the challenged ad, and whether the agency knew or should have known that the ad included false or deceptive claims.

  2. Does the 1984 substantiation policy statement apply only to advertisers?

    Answer guidance: No. It reaffirms that advertisers and ad agencies must have a reasonable basis for advertising claims before they are disseminated, naming agencies in the same sentence.

  3. What is the difference between responsibility, accountability and authority in this lesson?

    Answer guidance: Responsibility is doing a piece of the work properly and can sit with anyone, including a freelancer or a vendor. Accountability is answering by name for the outcome. Authority is the power to say yes or no and have it stick. Governance puts accountability and authority with the same named, competent person.

  4. Whom does PRSA's ESA #22 bind, and what does it say about AI tools?

    Answer guidance: It is advisory guidance for PRSA members, not law. It says AI systems are tools, not ethical actors, and that PR professionals are accountable for how the AI tools they use are designed, deployed and supervised.

Sources and further reading