Last Reviewed: September 2026

Artificial intelligence is increasingly used throughout advertising and marketing for research, analysis, audience development, media planning, optimization, customer service, personalization, content creation, production, reporting, forecasting, and other professional activities. These systems can improve speed and capability, but they can also introduce errors, bias, privacy risks, deceptive representations, intellectual-property concerns, synthetic media, unreliable automation, and new ways to make unsupported claims at scale.

The AAMA Responsible AI in Advertising & Marketing Guide provides a practical framework for using artificial intelligence while maintaining professional responsibility for the marketing decisions, claims, content, data, and customer experiences produced with it. It is designed for brands, agencies, marketers, advertisers, researchers, creators, media professionals, students, academics, technology providers, and organizations incorporating AI into marketing operations.

AI does not remove responsibility from the people and organizations using it. A marketer remains responsible for whether an advertisement is truthful, research is accurately represented, customer data are handled appropriately, intellectual property is respected, and automated systems are suitable for the decisions they are being asked to make.

This guide focuses primarily on professional practice in the United States. Laws, regulatory guidance, technology, platform rules, and professional standards continue to develop, and particular industries or jurisdictions may impose additional requirements. This resource is educational and should not be treated as legal advice.

Start With the Marketing Purpose

Organizations should begin by identifying why AI is being used rather than adopting it simply because the capability is available.

Potential uses may include:

• Research assistance
• Data analysis
• Content development
• Creative exploration
• Media optimization
• Customer segmentation
• Personalization
• Forecasting
• Customer service
• Translation
• Transcription
• Reporting
• Workflow automation
• Production assistance

For each use, ask what problem the system is expected to solve and whether AI is actually appropriate for that problem.

A faster process is not automatically a better process if the resulting work becomes less accurate, less transparent, or more difficult to verify.

Match Oversight to Risk

Not every AI-assisted task requires the same level of oversight.

Using AI to generate several internal headline ideas creates a different level of risk from using AI to determine who receives a credit offer, automatically respond to a customer complaint involving a financial transaction, make medical claims, or generate an advertising testimonial.

Higher-risk uses generally deserve greater:

• Human review
• Documentation
• Testing
• Approval
• Data controls
• Vendor evaluation
• Monitoring

The amount of oversight should reflect the consequences of being wrong.

Use a Risk Management Framework

Organizations developing substantial AI programs should consider a structured approach to identifying, evaluating, and managing AI risks.

The NIST AI Risk Management Framework provides a voluntary framework for managing risks associated with artificial intelligence. NIST also publishes a dedicated Generative AI Profile addressing risks and recommended actions associated with generative AI systems.

A marketing organization does not need to become an AI research laboratory to benefit from formal risk management. The core idea is to identify what can go wrong, determine which risks matter, establish appropriate controls, and monitor whether those controls actually work.

Maintain Human Accountability

AI systems can recommend, generate, classify, predict, and automate. They should not become an excuse for unclear responsibility.

Organizations should identify who is responsible for:

• Approving AI-generated advertising
• Verifying factual claims
• Reviewing research
• Protecting customer data
• Approving automated targeting
• Reviewing synthetic media
• Managing AI vendors
• Responding to errors
• Reviewing high-risk decisions

Statements such as “the AI did it” do not resolve professional responsibility.

The person or organization deploying the system should understand who owns the resulting decision.

Verify Factual Outputs

Generative AI systems can produce information that sounds credible while being inaccurate, incomplete, outdated, or entirely fabricated.

Marketers should verify material factual statements before publication.

Verification becomes especially important for:

• Statistics
• Research findings
• Historical claims
• Legal information
• Product specifications
• Competitor information
• Quotes
• Citations
• Prices
• Dates
• Scientific claims
• Financial information
• Health information

Do not assume that detailed language, citations, confidence, or polished formatting indicates accuracy.

Verify Sources Directly

When an AI system provides a source, statistic, quotation, or citation, locate the underlying source and examine it directly.

Check:

• Whether the source exists
• Whether the source actually says what the AI claims
• Publication date
• Author or organization
• Methodology
• Population
• Definitions
• Limitations
• Whether newer information exists

AI can assist with discovering sources. It should not replace reading the evidence supporting important claims.

Related AAMA Resource: Advertising & Marketing Research Sources Guide

Do Not Invent Citations

AI-generated citations can sometimes contain incorrect authors, titles, journals, URLs, dates, or publications.

Never place a citation into an academic paper, marketing report, client presentation, article, or public resource without confirming that the underlying source exists and supports the claim.

Fabricated citations undermine both the specific work and confidence in the organization producing it.

Distinguish AI Assistance From Evidence

An AI-generated explanation is not itself proof that a claim is true.

For example, an AI system may explain why a particular advertising approach could increase conversions. That explanation can generate a useful hypothesis, but it does not establish that the approach actually will increase conversions.

Evidence may instead require:

• Research
• Testing
• Customer data
• Experiments
• Measurement
• Academic literature
• Market data

Use AI to support thinking, not to manufacture evidence.

Do Not Overstate AI Capabilities

Organizations marketing AI-powered products or services should apply the same truth-in-advertising principles required for other technologies.

Claims about AI may involve:

• Accuracy
• Automation
• Prediction
• Personalization
• Productivity
• Cost reduction
• Detection
• Intelligence
• Security
• Revenue
• Customer outcomes

The Federal Trade Commission’s artificial intelligence resources document ongoing enforcement involving deceptive or unsupported AI-related representations.

If a company claims an AI system is more accurate, faster, safer, more effective, or capable of a particular function, appropriate evidence should support that claim.

“AI-Powered” Is Still an Advertising Claim

Adding artificial intelligence terminology to a product does not create an exemption from ordinary advertising standards.

Claims such as:

• AI-powered
• AI-driven
• Intelligent
• Predictive
• Autonomous
• Machine-learning powered

can create expectations about how the product actually operates.

Do not use AI language primarily to make an ordinary product sound more technologically advanced when the claimed capability is insignificant, nonexistent, or materially different from what customers would reasonably understand.

Substantiate Accuracy Claims

AI products often make claims involving accuracy, detection, prediction, or classification.

These claims require careful substantiation because performance can vary depending on:

• Dataset
• Population
• Language
• Context
• Input quality
• Threshold
• Model version
• Evaluation method

In 2025, the FTC announced an order involving Workado after alleging that its AI-content detector did not perform as accurately as advertised. The FTC’s Workado matter provides a useful reminder that AI accuracy claims require competent and reliable supporting evidence.

Do not convert performance observed under one narrow test into a broad claim about how the system performs everywhere.

Treat AI Detection Tools Carefully

Tools claiming to detect AI-generated writing, images, audio, or other media can produce false positives and false negatives.

Do not use an automated detector alone to make serious accusations involving:

• Academic misconduct
• Employee misconduct
• Plagiarism
• Fraud
• Customer deception
• Professional discipline

The consequences of an incorrect classification may substantially exceed the confidence justified by the detection system.

Use detection outputs as one piece of information rather than unquestionable proof.

Do Not Fabricate Customer Experiences

Generative AI should not be used to invent customer reviews, testimonials, or endorsements and present them as though they came from actual people.

The FTC’s Consumer Reviews and Testimonials Rule addresses fake and false reviews and testimonials, including certain AI-generated examples.

AI can assist with organizing authentic feedback or preparing internal analysis, but it should not create imaginary customers whose experiences are presented as genuine.

Related AAMA Resource: Advertising Ethics Guide

Synthetic Spokespeople Require Care

AI-generated avatars and virtual presenters can be legitimate creative tools. The ethical question is whether the presentation creates a misleading impression about who or what the audience is seeing.

Consider whether audiences may believe:

• The person is real
• The person is an actual customer
• A real expert made the statement
• A celebrity authorized the appearance
• A testimonial reflects genuine experience

The FTC does not prohibit AI avatars categorically, but existing deception and testimonial rules still apply.

If a synthetic person is simply functioning as an obvious fictional spokesperson, that is different from presenting the synthetic person as a real customer describing a real experience.

Do Not Create False Experts

An AI-generated doctor, lawyer, engineer, financial advisor, academic, technician, or other professional figure should not be presented in a way that causes audiences to believe a real qualified expert endorsed a claim when no such endorsement exists.

Professional appearance can create credibility.

If a synthetic character is used, avoid creating credentials, institutions, professional histories, or expert experiences that never existed unless the fictional nature is unmistakable.

Protect Real People’s Identity

AI can replicate or imitate:

• Faces
• Voices
• Writing styles
• Performances
• Likenesses
• Personal characteristics

Do not assume that technical ability to reproduce someone’s identity creates permission to use that identity commercially.

The U.S. Copyright Office’s Artificial Intelligence Study has examined digital replicas and other AI-related copyright and policy questions, illustrating the broader legal and policy concerns surrounding synthetic representations of real people.

Use appropriate permissions, contracts, releases, and legal review when real identities are involved.

Do Not Impersonate Businesses or Government Entities

AI can make impersonation easier through realistic text, voice, imagery, websites, chat interfaces, and automated communication.

The FTC’s Impersonation of Government and Businesses Rule prohibits certain deceptive impersonation practices.

Marketing should never create the false impression that communication originates from:

• A government agency
• Another company
• A competitor
• A bank
• A recognized institution
• An official representative

AI-generated realism does not make impersonation acceptable.

Be Careful With Voice Cloning

Voice cloning can be useful for approved production workflows, localization, accessibility, and authorized creative work.

Organizations should establish clear permission before creating or using a synthetic version of a person’s voice.

Consider:

• Consent
• Permitted uses
• Duration
• Markets
• Languages
• Compensation
• Future reuse
• Modification
• Revocation
• Security

A voice model can become an enduring digital asset. Treat access to it accordingly.

Identify Synthetic Media When Necessary

Not every AI-assisted image or video requires a disclosure simply because AI contributed to production.

Disclosure becomes more important when the synthetic nature of the content affects how audiences would interpret what they are seeing or hearing.

Examples can include:

• Fabricated events
• Synthetic customer testimonials
• Simulated product demonstrations
• AI-generated news-like footage
• Digital replicas
• Synthetic experts
• Materially altered documentary imagery

The question should be whether an ordinary audience could form a materially false impression if the synthetic nature remains undisclosed.

Preserve Content Provenance Where Practical

Organizations producing substantial amounts of synthetic media should consider systems for recording how content was created and modified.

NIST’s work on digital content transparency and synthetic content discusses technical approaches including provenance information, metadata, watermarking, and detection.

No single technical mechanism solves the problem completely, but provenance can help organizations document production history and respond to later questions about authenticity.

Keep Original Assets

For significant campaigns involving AI-generated or heavily AI-modified media, preserve relevant source materials.

These may include:

• Original photographs
• Original video
• Recorded voice
• Prompts
• Generated versions
• Human edits
• Approval records
• Model information
• Production notes

Documentation can become particularly useful when questions arise about what was real, what was altered, and who approved the final output.

Review AI-Generated Product Images

Generative image tools can create polished product photography that does not accurately represent the actual product.

Check:

• Shape
• Size
• Color
• Features
• Packaging
• Accessories
• Quantity
• Included components
• Product performance

An attractive image can still be deceptive if customers reasonably believe it represents the product they will receive.

Where a visualization is conceptual rather than representative, consider whether additional context is needed.

Review AI Product Demonstrations

Synthetic video can show products performing actions they never actually performed.

Do not use generated or altered demonstrations to imply unsupported:

• Speed
• Strength
• Durability
• Appearance
• Functionality
• Results

The same standards that apply to traditional advertising demonstrations should apply when AI makes fabrication easier.

Use AI for Creative Exploration, Not Automatic Approval

AI can rapidly generate:

• Headlines
• Concepts
• Scripts
• Layout directions
• Images
• Storyboards
• Variations

Creative teams should still evaluate whether the output fits:

• Strategy
• Audience
• Brand
• Ethics
• Legal requirements
• Production reality
• Cultural context

Volume is not a substitute for judgment.

Generating 500 options does not guarantee that any of them solve the communication problem.

Preserve Human Creative Direction

AI can assist creative professionals without replacing the need for intentional creative decisions.

Humans should continue determining:

• What the work is trying to communicate
• Why the idea matters
• What the audience should experience
• Which execution is appropriate
• What should be rejected
• What requires revision

Creative work should not become a process of publishing whichever output an AI system produced first.

Avoid Homogenized Brand Communication

Organizations using the same generative tools, prompts, structures, and visual conventions can begin sounding and looking alike.

Review AI-assisted work for:

• Generic corporate language
• Repetitive sentence structures
• Familiar AI visual conventions
• Empty adjectives
• Predictable imagery
• Unnecessary summaries
• Loss of distinctive brand vocabulary

Efficiency has little value if it erodes the characteristics that make the brand recognizable.

Maintain Brand Voice

AI systems should be trained or instructed using approved brand guidance where permitted and appropriate.

Human reviewers should still evaluate:

• Vocabulary
• Tone
• Formality
• Humor
• Terminology
• Sentence style
• Calls to action

Do not assume an AI system understands a brand because it was given several examples.

Related AAMA Resource: Brand Voice Development Guide

Protect Confidential Information in Prompts

Do not paste confidential information into an AI system without understanding how the provider handles that information.

Potentially sensitive materials can include:

• Unreleased campaigns
• Client strategy
• Financial information
• Customer records
• Contracts
• Research
• Credentials
• Proprietary data
• Personal information
• Trade secrets

Organizations should establish which AI systems are approved for which categories of information.

A public consumer AI service and an enterprise system operating under negotiated data protections may present very different risk profiles.

Do Not Put Customer Personal Data Into Unapproved Systems

Customer data deserve particular protection.

Before using customer information with an AI system, understand:

• What data are being transmitted
• Where they are stored
• Whether they are retained
• Whether they are used for model training
• Who can access them
• What contractual protections exist
• What deletion mechanisms exist
• Which jurisdictions may apply

Marketing convenience should not override established privacy and security requirements.

Minimize Data

Only provide AI systems with information necessary for the task.

If an analysis can be performed using:

• Aggregated data
• Anonymized data
• Synthetic examples
• Reduced datasets

consider whether there is any reason to expose identifiable customer information.

Data minimization reduces the consequences of both technical failure and human error.

Review Vendor Data Practices

Organizations should not assume that an AI vendor’s marketing language provides a complete picture of its data practices.

Evaluate:

• Privacy documentation
• Contract terms
• Security controls
• Retention
• Training use
• Subprocessors
• Access controls
• Data location
• Incident procedures

For higher-risk uses, procurement, privacy, security, legal, and marketing teams may all need to participate in vendor evaluation.

Verify Extraordinary Vendor Claims

AI vendors frequently market dramatic improvements in:

• Productivity
• Revenue
• Conversion
• Cost reduction
• Automation
• Accuracy

Treat vendor claims as advertising claims that require evaluation.

Ask:

• What was measured?
• Against what baseline?
• With what customers?
• Over what period?
• Under what conditions?
• Is the result typical?
• Can the evidence be independently reviewed?

A product being labeled AI does not make extraordinary performance claims self-validating.

Be Skeptical of Surveillance Claims

Claims that AI can infer consumer interests from hidden, extraordinary, or unusually invasive data sources deserve especially careful scrutiny.

In August 2026, the FTC finalized orders involving Cox Media Group and two marketing firms after alleging they falsely represented an “Active Listening” service as using conversations captured from consumers’ smart devices and falsely represented consumer opt-in. The FTC’s Active Listening enforcement announcement is directly relevant to marketers evaluating AI-powered targeting vendors.

Do not repeat a vendor’s technical or privacy claims to clients simply because the vendor included them in a sales presentation.

Marketing organizations should verify what data actually exist, where they came from, how they were collected, and whether the advertised capability genuinely functions as represented.

Use AI Targeting Responsibly

AI can help identify audiences, predict response, optimize delivery, and allocate advertising budgets.

These systems can also reproduce or amplify patterns contained in historical data.

Review targeting systems for potential effects involving:

• Exclusion
• Discrimination
• Geographic bias
• Economic bias
• Sensitive characteristics
• Vulnerable audiences
• Proxy variables
• Unequal access to opportunities

Optimization toward a numerical objective does not guarantee an ethically acceptable outcome.

Audit Outcomes, Not Only Inputs

An organization may never explicitly instruct an AI system to discriminate and still produce systematically unequal results.

Review actual campaign outcomes when appropriate.

Questions can include:

• Who receives the advertising?
• Who is excluded?
• Who receives different offers?
• Are particular groups disproportionately filtered out?
• Is the result consistent with the campaign’s legitimate purpose?

Responsible AI governance requires looking at consequences as well as intentions.

Be Careful With Lookalike & Predictive Audiences

AI-generated audiences can use complex correlations that marketers may not fully understand.

Before using them for sensitive categories, consider whether the underlying attributes or proxies could create inappropriate targeting.

A model may discover that a particular location, browsing behavior, or purchase pattern correlates with a protected or sensitive characteristic even if that characteristic was never entered explicitly.

High-impact categories deserve stronger review.

Avoid Exploiting Vulnerability

AI can identify patterns associated with urgency, distress, compulsive behavior, financial pressure, insecurity, or other vulnerabilities.

The fact that a model can identify someone as highly persuadable does not mean the organization should exploit that condition.

Additional ethical review is appropriate when marketing involves:

• Financial distress
• Health concerns
• Children
• Addiction
• Crisis
• Major life events
• Other sensitive circumstances

AI should improve relevance without turning vulnerability into a targeting advantage.

Personalization Should Remain Appropriate

AI can personalize messages at a scale that was previously impractical.

Personalization can become uncomfortable or manipulative when it reveals how much the organization knows about a person.

Consider whether a message would cause the customer to ask:

How did they know that about me?

The most sophisticated personalization is not always the most appropriate personalization.

Do Not Manufacture False Intimacy

Generative AI can communicate conversationally and imitate empathy, familiarity, or personal understanding.

Organizations should be careful about deliberately encouraging customers to believe an automated system has emotions, relationships, memories, or personal concern that it does not actually possess.

Human-like communication can make interfaces easier to use. It should not be designed primarily to manipulate customers into overestimating the nature of the relationship.

Identify Automated Agents When Identity Matters

Not every chatbot needs a dramatic warning that it uses artificial intelligence.

Transparency becomes more important when a customer could reasonably believe they are interacting with a human and that misunderstanding could materially affect the interaction.

This can be particularly important in:

• Customer complaints
• Sales conversations
• Negotiations
• Sensitive services
• Expert advice
• High-stakes decisions

Use clear language appropriate to the situation.

Do Not Create Fake Human Scarcity

If an automated sales system is capable of handling unlimited conversations, do not falsely imply that:

• A human representative personally selected the customer
• An executive is typing the message
• A representative has only one appointment remaining
• A salesperson personally wrote a mass-generated message

Automation should not depend on fictional human involvement to create urgency or credibility.

Human Review Should Be Meaningful

A “human in the loop” provides little protection if the human simply approves every AI output without enough time or expertise to evaluate it.

Meaningful oversight requires:

• Appropriate training
• Access to relevant information
• Authority to reject the output
• Reasonable review time
• Clear escalation procedures

Human review should change outcomes when necessary.

Avoid Automation Bias

People can become overly willing to trust recommendations produced by algorithms.

This is particularly dangerous when a system presents outputs with precise probabilities, scores, rankings, or predictions.

Users should understand:

• What the system measures
• What it does not measure
• Relevant limitations
• Appropriate confidence
• When human judgment should override it

Precision in presentation does not guarantee precision in reality.

Test Before Deployment

AI systems should be evaluated in the context in which they will actually be used.

Testing may examine:

• Accuracy
• Reliability
• Bias
• Privacy
• Security
• Brand consistency
• Factuality
• Customer experience
• Failure modes
• Edge cases

A model performing well in a vendor demonstration may behave very differently with actual organizational data and users.

Test After Model Changes

AI systems change.

Vendors may modify:

• Models
• Safety systems
• Training
• Interfaces
• Retrieval systems
• Default behavior
• APIs

Organizations using AI in important workflows should consider whether major changes require renewed evaluation.

A system that worked acceptably six months ago should not be assumed to behave identically forever.

Monitor Performance Over Time

Responsible deployment continues after launch.

Monitor for:

• Accuracy changes
• Unexpected outputs
• Customer complaints
• Bias
• Security issues
• Disclosure failures
• Brand problems
• Vendor changes
• New regulatory requirements

Create a process for reporting and investigating recurring problems.

Maintain an AI Use Inventory

Organizations using AI across several teams should maintain a basic record of significant use cases.

An inventory may identify:

• System
• Vendor
• Purpose
• Department
• Data involved
• Risk level
• Human reviewer
• Approval status
• Last review date

Without an inventory, organizations may not know where AI is being used until a problem occurs.

Create Approved & Prohibited Use Categories

Organizations can reduce uncertainty by defining common use categories.

For example:

Generally Appropriate With Normal Review

• Brainstorming
• Internal summaries
• Draft outlines
• Formatting assistance
• Non-sensitive ideation

Requires Additional Review

• Public factual content
• Customer personalization
• Research summaries
• Advertising claims
• Synthetic media
• Customer-facing chatbots

Restricted or Prohibited Without Specific Approval

• Sensitive personal data
• Legal decisions
• Unapproved voice cloning
• Fake testimonials
• Confidential client information in public tools
• High-impact automated decisions

The categories should reflect the organization’s actual work and risk tolerance.

Establish an AI Approval Process

Higher-risk uses should have clear approval requirements.

Potential reviewers may include:

• Marketing
• Creative
• Legal
• Privacy
• Security
• Research
• Compliance
• Technology
• Leadership

The objective is not to make every AI use bureaucratic. It is to ensure that high-consequence uses receive appropriate scrutiny.

Train Employees

Responsible AI depends on people understanding both capabilities and limitations.

Training should cover:

• Verification
• Confidentiality
• Approved tools
• Customer data
• Copyright
• Synthetic media
• Claims
• Bias
• Human oversight
• Escalation

Employees should also know whom to contact when they encounter a use case the policy does not address.

Establish Prompt & Data Guidelines

Organizations should define what employees may place into AI systems.

Guidelines can distinguish among:

• Public information
• Internal information
• Confidential information
• Customer data
• Regulated data
• Trade secrets

A clear policy is easier to follow than simply telling employees to “be careful.”

Address Intellectual Property

Generative AI raises evolving questions involving training data, generated output, ownership, licensing, similarity, and human authorship.

The U.S. Copyright Office’s AI initiative provides current federal analysis of several of these issues.

Organizations should understand that the legal status of AI-generated material can differ from conventionally authored work.

Do not assume that every generated output:

• Is copyrightable
• Is exclusive
• Is free from third-party rights
• Can be registered
• Can safely imitate existing creative work

Appropriate review becomes increasingly important as commercial stakes increase.

Understand Human Authorship

The U.S. Copyright Office has concluded that copyright protection depends on sufficient human authorship. AI can assist human creative work, and human selection, arrangement, or modification may contribute protectable authorship, but prompts alone do not automatically make the resulting AI expression copyrightable.

Marketing organizations should consider this when AI-generated assets are intended to become valuable long-term intellectual property.

A low-cost generated image may be useful for a short-lived social post while presenting different concerns when intended to become a permanent brand character, packaging system, campaign platform, or proprietary identity.

Avoid Intentional Style Imitation When Risk Is Unnecessary

Generative systems can be prompted to imitate recognizable artists, photographers, writers, designers, campaigns, or brands.

Even where a particular use may present unresolved legal questions, organizations should consider whether direct imitation is professionally necessary.

Creative direction can usually be described through:

• Era
• Medium
• Technique
• Composition
• Mood
• Lighting
• Genre
• Visual characteristics

without building the concept around copying the identifiable work of a living creator or competitor.

Review Outputs for Third-Party Material

Generated content can occasionally resemble or reproduce recognizable material.

Review commercial outputs for potential:

• Logos
• Trademarks
• Characters
• Product designs
• Text
• Images
• Brand elements
• Known creative work

Do not assume that because an AI system generated something, no third-party rights are implicated.

Review AI Translation

AI translation can greatly improve speed and access, but incorrect translation can alter:

• Claims
• Legal disclosures
• Product instructions
• Tone
• Cultural meaning

Important public-facing translations should receive review by someone competent in the language and subject matter.

This is especially important for regulated claims or contractual communication.

Review AI Accessibility Outputs

AI can assist with:

• Alt text
• Captions
• Transcripts
• Descriptions
• Simplification

These outputs should still be reviewed.

An automatically generated image description that incorrectly identifies a person, product, chart, or action can reduce accessibility rather than improve it.

AI Should Not Replace Research Participants

Synthetic personas and AI-generated “consumers” may help teams brainstorm possible questions or hypotheses.

They are not substitutes for evidence from real customers when the research question concerns what actual people think, feel, know, or do.

An AI-generated persona cannot establish:

• Customer preference
• Purchase intent
• Brand awareness
• Satisfaction
• Consumer motivation

Treat synthetic respondents as ideation tools, not research samples.

Do Not Fabricate Focus Groups

A simulated focus group generated by AI may be useful for anticipating questions.

It should not be presented as qualitative research involving real participants.

If a report states that “customers said” something, actual customer evidence should support that statement.

The distinction between simulation and research should remain explicit.

Use AI Carefully in Survey Analysis

AI can help categorize or summarize large volumes of open-ended survey responses.

Researchers should validate whether the system accurately represents:

• Themes
• Frequency
• Sentiment
• Nuance
• Minority viewpoints

Do not allow automated summarization to erase contradictory or less common responses simply because the model considers them less central.

Related AAMA Resource: How to Design a Marketing Survey

Document AI-Assisted Research Methods

When AI materially influences a research process, document how it was used when that information affects interpretation.

This may include AI-assisted:

• Coding
• Classification
• Translation
• Summarization
• Data cleaning
• Analysis

Research transparency helps others understand how conclusions were produced.

Do Not Let AI Create Unsupported Insights

AI systems are particularly good at turning patterns into plausible narratives.

A correlation in campaign data may produce an elegant explanation that the evidence cannot actually support.

Separate:

Observation: Customers exposed to Channel A converted at a higher rate.

from:

Explanation: Channel A caused customers to trust the brand more.

The second statement may require additional research.

Maintain Advertising Claim Standards

AI-generated advertising copy should pass the same claims review as human-written copy.

Review:

• Objective claims
• Comparisons
• Superlatives
• Performance claims
• Health claims
• Environmental claims
• Financial claims
• Testimonials
• Statistics

An AI system may invent attractive claims because they sound persuasive.

The advertiser remains responsible for whether those claims are supportable.

Review AI-Generated Fine Print

Do not rely on generative AI to create legal disclosures or terms without qualified review.

A disclosure can sound professionally written while:

• Omitting required information
• Using obsolete requirements
• Applying the wrong jurisdiction
• Contradicting the main claim

Use appropriate legal or compliance expertise where interpretation is required.

Do Not Generate Fake Social Proof

Avoid using AI to manufacture:

• Fake comments
• Fake followers
• Fake engagement
• Fake customer photographs
• Fake user posts
• Fake testimonials
• Fake endorsements

Social proof is persuasive because audiences believe it represents the behavior or opinions of other people.

Fabricating those people is deception, not optimization.

Disclose AI When the Fact Is Material

AAMA does not recommend placing an “AI-generated” label on every piece of content touched by an AI tool.

Instead, ask whether knowing about the AI involvement would materially affect how the audience understands or evaluates the communication.

Disclosure deserves greater consideration when AI is used to create:

• A person who appears real
• A testimonial
• An expert
• Documentary-style evidence
• News-like footage
• A simulation presented as actual performance
• An automated advisor presented as a human

Transparency should address the potential misunderstanding rather than become a meaningless universal badge.

Avoid AI Disclosure Theater

A generic statement such as “AI may have been used” can create the appearance of transparency without explaining anything useful.

When disclosure matters, identify the material fact.

For example:

“This demonstration contains AI-generated imagery.”

is more useful than:

“Created with technology.”

Transparency should clarify, not obscure.

Develop an Incident Response Process

Organizations should decide what happens when an AI system creates a material problem.

Possible incidents include:

• Publication of false information
• Confidential-data exposure
• Offensive content
• Unauthorized likeness
• False claim
• Customer harm
• Discriminatory output
• Vendor failure

An incident process should identify who can stop the system, correct public material, notify stakeholders, preserve evidence, and investigate the cause.

Correct Errors Promptly

AI allows content to be produced quickly, which can also allow errors to spread quickly.

When an important error is discovered:

1. Stop further distribution when appropriate.
2. Correct the content.
3. Determine whether customers or clients need notification.
4. Identify the cause.
5. Update the process to reduce recurrence.

Speed of generation should be matched by speed of correction.

Recommended AI Governance Structure

A practical marketing AI governance system can include:

1. Approved AI Tools
2. AI Use Inventory
3. Risk Classification
4. Data Rules
5. Human Oversight Requirements
6. Content Verification Standards
7. Claims Review
8. Synthetic Media Policy
9. Intellectual-Property Review
10. Vendor Evaluation
11. Testing Requirements
12. Monitoring
13. Employee Training
14. Incident Response
15. Periodic Review

The system should be proportionate to the organization’s size and use of AI.

A small agency does not need the same governance structure as a multinational corporation, but both should know what tools are being used and who is accountable for the results.

Responsible AI Review Checklist

Before deploying AI in an advertising or marketing workflow, confirm:

• The business purpose is clear
• AI is appropriate for the task
• Risk level has been considered
• An accountable human owner exists
• The system has been tested
• Material factual outputs will be verified
• Claims will be substantiated
• Confidential information is protected
• Customer data are handled appropriately
• Vendor data practices have been reviewed
• Potential bias has been considered
• Synthetic people or media will not mislead audiences
• Reviews and testimonials are genuine
• Intellectual-property issues have been considered
• Human review is meaningful
• Material AI involvement will be disclosed when necessary
• Performance will be monitored
• Errors can be escalated and corrected
• Documentation will be maintained where appropriate

The checklist should help teams identify the issues that deserve attention rather than serve as automatic approval for every use.

AI Should Extend Professional Capability

Artificial intelligence can make professionals faster, broaden the number of ideas they can explore, improve access to information, automate routine processes, and support analysis that would otherwise require substantially more time.

The strongest use of AI does not eliminate professional expertise. It allows professionals to apply that expertise more effectively.

A strategist still needs judgment. A researcher still needs methodological discipline. A copywriter still needs to understand language and audience. A media professional still needs to understand measurement. A creative director still needs taste and direction.

AI changes the tools available to the profession without eliminating the standards of the profession.

Responsibility Cannot Be Automated Away

The central principle of responsible AI in advertising and marketing is accountability.

Organizations can automate generation, prediction, personalization, media buying, analysis, and customer interaction. They cannot automate away responsibility for what those systems communicate, whom they affect, what data they use, or what decisions are made because of them.

The appropriate question is not simply:

Can AI do this?

Professional practice also requires asking:

Should we use it here, what could go wrong, how will we know, and who is responsible if it does?

Related AAMA Resources

Continue exploring responsible professional practice with the Advertising Ethics Guide, Influencer & Sponsored Content Disclosure Guide, Advertising Claims Checklist, Marketing Data Ethics Checklist, Marketing Research Methods Guide, How to Design a Marketing Survey, A/B Testing Guide for Marketers, Brand Voice Development Guide, and Advertising Copy Review Checklist. These resources provide additional guidance for evaluating evidence, claims, data, automation, creative communication, and professional decision-making.

The AAMA Resource Library will continue reviewing this guide as AI technology, regulatory guidance, professional practices, and technical standards evolve.