AI Agents for Marketing: Content, Campaigns and Reporting

AI agents for marketing are most useful on five jobs: researching a topic or audience, drafting from a brief, repurposing one piece into many, checking a campaign before launch, and summarizing the results. In each, the agent gathers, drafts and checks, and a person approves anything published or sent, every claim about the product, and anything that looks like an endorsement or a review. In the US the FTC’s rules on endorsements, fake reviews and AI claims, and the CAN-SPAM Act for email, apply to content an agent writes exactly as they apply to content a person writes. Keep the brand-voice rules where every assistant reads them before it starts.

This page is about a marketing team’s work. The same approach for a whole small business, including social posts and review replies, is in AI agents for small businesses; outreach to named prospects belongs to AI sales agents. The rule across all three is the same: the agent drafts, a person decides.

Five jobs worth handing over

1. Research

Before a brief is written, the agent reads competitors’ public pages, your support tickets and sales call notes for the questions customers actually ask, and the search terms people use, and writes a one-page summary with a source for every fact. Ask for the sources, because a confident, unsourced claim about a competitor is how a campaign ends up saying something untrue. An agent that reads web pages can also be steered by text hidden in them; indirect prompt injection explains the risk and the limits to set.

2. Drafts from a brief

The brief is the instruction, so it decides the quality of the draft. A short one the agent can follow:

Piece:      Blog post, 1,200 words, for the October newsletter
Audience:   Office managers at US companies with 20-200 staff
Goal:       Understand why shared calendars double-book rooms
Key point:  One source of truth for bookings beats three
Claims:     Only those in the product fact sheet (attached)
Must not:   Name competitors; promise features on the roadmap
Voice:      Follow the brand rules; plain words, no hype
Call to action: Book a demo
Review:     Marketing lead approves; product checks any claim

The “Claims” line does the most work. An agent told to use only the fact sheet cannot invent a statistic, and a reviewer knows exactly what to check the draft against.

3. Repurposing

One approved piece becomes a newsletter section, three social posts, a short script and a sales one-pager. This is where agents save the most time, and the rule is simple: repurposing may cut and reword but never add. Any fact, number or promise not in the approved source goes back to review, because the approved piece is what was checked.

4. Campaign QA

Before launch, the agent checks what people miss at the end of a long week and reports a table of problems for a person to fix:

• Links: every link resolves, goes where the text says, and none points at a staging site.
• UTM tags: present and consistent. Google Analytics’ guide to campaign URLs↗ (opens in a new tab) says to always use utm_source, utm_medium and utm_campaign, and that values are case sensitive, so “Email” and “email” report as two sources.
• Alt text: every image has it, and it says what the image shows or does. The W3C’s images tutorial↗ (opens in a new tab) adds that purely decorative images take an empty alt="".
• Disclosures: every paid, gifted or affiliate mention carries a disclosure where it cannot be missed.
• Email basics: the postal address and unsubscribe link are in every commercial email.
• Claims: every number and superlative matches the fact sheet.

5. Reporting summaries

After the campaign, the agent turns analytics exports into a one-page summary: what ran, what each channel delivered against its goal, and what to try next. Give it the export rather than asking it to recall figures, have it quote each number with its source, and keep it to describing what happened. Whether the new subject line caused the lift is a judgment for a person who knows what else changed that week.

What a human must approve

• Anything published or sent: posts, pages, ads, emails and replies.
• Every claim about the product, price, results or competitors, and every number.
• Anything that could read as an endorsement or a review, and every disclosure of a paid or gifted relationship.
• Use of images, quotes, names and logos the team does not own.
• Audience lists, and who is emailed.
• Budget and spend changes in ad accounts.
• Replies to complaints or anything touching legal, health or financial topics.

Start with every item approved one at a time. Loosen only where drafts go out unchanged for weeks, and never for claims, endorsements or spend. Worked examples of where the person steps in, including publishing a campaign, are in human-in-the-loop AI examples.

US rules in plain words

This is a plain-language summary, not legal advice. State laws and industry rules add to it; talk to counsel before an agent publishes or sends anything on its own.

Endorsements and influencers

The FTC’s Guides Concerning the Use of Endorsements and Testimonials in Advertising, 16 CFR Part 255↗ (opens in a new tab), say endorsements must reflect the honest opinions, findings, beliefs or experience of the endorser. A connection that might materially affect how much weight people give an endorsement, such as payment, free products or a family or business relationship, must be disclosed clearly and conspicuously if the audience would not expect it. Advertisers can be liable for what their endorsers say, and the Guides expect them to give endorsers guidance and monitor what they post. An agent drafting influencer briefs should put the disclosure in the brief, not leave it to the influencer.

Reviews and testimonials

The FTC’s Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465↗ (opens in a new tab), finalized in August 2024, makes it a violation for a business to write, create or sell a review or testimonial that misrepresents that the reviewer exists, that they used the product, or what their experience was. The FTC’s announcement of the rule named AI-generated fake reviews among the practices it bans. The rule also covers buying reviews that must express a particular sentiment, undisclosed reviews by officers and managers, suppressing negative reviews, and fake indicators of social media influence such as bot followers. The plain rule for an agent: it never writes a review or testimonial in a customer’s voice, and it never drafts a request for reviews that asks for a positive one.

Claims about AI

If you market a product that uses AI, the AI claims are ad claims like any other. When the FTC announced Operation AI Comply↗ (opens in a new tab) in September 2024, its chair said there is “no AI exemption from the laws on the books”. The cases included a service marketed as a “robot lawyer” and an AI writing tool that let users generate fake reviews. Have the agent flag every sentence that says what the product’s AI does, and have someone who can prove it approve each one.

Email: CAN-SPAM

The FTC’s CAN-SPAM compliance guide↗ (opens in a new tab) says the law covers all commercial email, business-to-business included. Headers and subject lines must not mislead, the message must be identified as an ad, it must include a valid physical postal address and a clear way to opt out, and opt-outs must be honored within 10 business days. Hiring someone else to send the email does not move the responsibility off you. For an agent, that means drafting only into a template that already carries the address and unsubscribe link, and never writing a subject line that promises what the email does not deliver.

Brand-voice rules every assistant reads first

Voice drifts when the rules live in one person’s prompt. Write them down once where every assistant reads them before it starts. On fenbs that place is the Decisions and rules page: a rule is a decision that holds from now on, it shows in full at the top of every connected AI assistant’s context, and the decider is always a person, never an assistant. Good marketing rules are short and checkable:

• “Never publish or send anything without the marketing lead’s approval.”
• “No product claim that is not in the current fact sheet.”
• “American English; sentence case in headlines; no exclamation marks.”
• “Never write a review, testimonial or quote in a customer’s name.”
• “Paid and gifted mentions always carry a disclosure in the first line.”

Facts nobody chose, such as product names, the fact sheet’s location or the UTM naming scheme, go in AI context notes, which every assistant also reads. Approvals become tasks: the assistant files “approve the October newsletter draft” with the draft linked in the note and a priority from 1 to 10, and a person moves it through To Do, Next Up, In Progress and Completed. History records each change under the name of whoever made it, assistant or person. fenbs has no due dates or content calendar, so keep publishing dates in the calendar you already use and the approvals on the board.