MyAIDevelopment · Product · Operations notes

AI Workflows · Risks

Marketing is one of the business areas where generative AI is used most widely. McKinsey's State of AI survey (2025-03) named marketing and sales as the function where organizations most often use generative AI. On the other hand, the IAB's State of Data 2025 (2025-03) reported that only 30% of agencies, brands, and publishers have fully integrated AI across the media campaign lifecycle. In other words, adoption is broad but depth still varies. What matters is where you use it and where a person checks it.

Map of Uses

Work Example AI use What a person must do
Research Summarizing interview transcripts, classifying reviews and inquiries Check the originals, check for bias
Positioning and messaging Generating phrasing options, summarizing competitor messages Choose and own the decision, verify evidence
Content Drafts, translation, summaries, variations Fact-check, add unique experience
Creative Variations of ad images and videos Check rights, check disclosure duties
Paid Automated bidding, targeting, and creative combinations (built into platforms) Conversion definitions, budget caps, brand safety
Lifecycle Personalizing copy per segment Scope of consent, frequency limits
Analysis Writing queries, summarizing data, detecting anomalies Verify calculations, interpret causality
Customer support FAQ answers, conversation summaries Accountability for wrong answers, a path to a human

Recommended Workflow

flowchart LR
    B[Brief] --> G[AI Draft]
    G --> F[Fact Check]
    F --> L[Legal and Brand Review]
    L --> P[Publish]
    P --> M[Measure]
    M --> B
    F -->|Fail| G
    L -->|Fail| G

What the Brief Must Contain

Goal: whom should it get to do what
Segment and situation:
Positioning summary and banned expressions:
Facts and numbers that may be used (with sources):
Format, length, tone:
Reviewer:

Do not let AI create facts; give it verified facts and let it handle only the wording.

Key Risks

NIST's Generative AI Profile (NIST AI 600-1, 2024-07-26) lists risks unique to generative AI, including confabulation (plausible wrong answers), harmful content, privacy, information security, and intellectual property. In marketing they often look like this.

Risk How it shows up in marketing Response
Confabulation Nonexistent features, wrong numbers, fabricated quotes Provide a fact list, cross-check sources
Privacy Entering customer data into external tools Define data that must not be entered, check contracts and settings
Rights infringement Output resembling others' works, likenesses, or trademarks Confirm usage rights, check similarity
Deceptive expression AI people who look real, fake reviews Follow disclosure duties, no fake reviews (document 09)
Homogenization Everyone writes similar copy with the same tools Add unique data and experience
Search quality Mass auto-generated pages May violate Google spam policies (document 04)
Over-trusting automation Scaling platform AI campaigns without validation Incrementality tests (document 06)

Bad Example / Good Example

Bad example:

Ask "Write 10 strengths of our product and 5 customer reviews" and publish the result as is.

Good example:

Provide real customer interview summaries and an approved feature list, and get five landing page copy options. A marketer picks two, the product team checks the feature descriptions, numbers are cross-checked against the source data, and the result is validated with an A/B test. Reviews use only original text from real customers who gave consent.

Adoption Checklist

Common Mistakes

References