Give an AI marketing assistant a brief it can actually use.
The quality of an AI marketing assistant depends on the job it receives. This guide gives you a one-page brief, a worked example, and an acceptance test that turn a vague request into campaign work a person can check.
A useful AI marketing assistant brief makes the decision before it asks for the copy.
Give an AI marketing assistant one business outcome, one audience, one offer, approved evidence, the channels and formats, claim boundaries, source materials, approval rules, tracking names, a deadline, and an exact definition of done. That structure removes the guesses that create generic campaign copy. It also makes the result testable. The assistant can produce several creative angles, but each angle should trace back to the same audience problem, evidence and business goal.
Official prompt guidance from OpenAI recommends clear, specific instructions about context, outcome, length, format and style, with examples where the output shape matters. A marketing brief is that guidance translated into the decisions a campaign needs.
Start the AI assistant for marketing with one business outcome.
“Create a campaign” is an activity, not an outcome. Name what should change in the business and by when. A useful line is: “Generate 20 qualified consultation bookings from logistics companies with 20 to 200 employees in Germany by October 31, without raising cost per qualified booking above €180.” It identifies the event, audience, geography, deadline and cost boundary. The assistant can now reject ideas that do not serve the goal.
Choose one primary success metric before creative work begins. Google Analytics defines a key event as an action important to the business, and distinguishes that from an advertising conversion used to measure and optimise campaigns. The brief should use the event closest to value, such as a completed purchase, qualified lead or activated trial. Add diagnostic metrics like landing-page conversion rate only when they help explain the primary result. Google Analytics explains the distinction here.
Give the marketing AI assistant evidence and claim boundaries.
List the sources the assistant may rely on: the current product page, approved customer research, a price sheet, product documentation, signed-off testimonials, and prior campaign data. Put a source beside every factual claim. When evidence is missing, the output should mark the gap instead of filling it with a plausible sentence.
The Federal Trade Commission says advertising must be truthful and not misleading, and that advertisers need evidence for objective claims before they run them. Its substantiation policy also ties the required support to the claim being made. A brief should therefore name prohibited claims, required qualifications, regulated topics, comparison rules and who approves exceptions. See the FTC's online advertising guide and advertising substantiation policy.
A useful boundary can be plain: “Do not claim fastest, safest, guaranteed, compliant or best. Do not quote a number unless the approved source is linked beside it. Flag any legal, medical, financial, privacy or competitor claim for review.” The assistant still has room to write. It no longer has room to invent proof.
Specify channel, format and source assets before generation.
One idea does not become the same text everywhere. A search headline, a landing-page hero, a six-second video opening and a customer email have different space, intent and policy constraints. Name the channel, placement, character or time limits, language, reading level, required call to action and number of variants. Include one accepted example and one rejected example when tone is hard to describe.
For Google Ads asset groups, Google recommends organising creative around one theme or audience and providing related text, image and video assets. Its guidance also shows that the platform assembles assets across formats, which makes consistency between the final URL, audience and asset group essential. Review asset group best practices and the asset group setup guide.
Put tracking and the test plan inside the AI marketing assistant brief.
Decide the naming scheme before links are created. Google Analytics recommends using source, medium and campaign parameters and using content to distinguish creative. Its URL builder guidance warns that values are case-sensitive, so “LinkedIn” and “linkedin” fragment the reporting. Give the assistant the exact lowercase taxonomy and ask it to return the tagged final URLs as part of the deliverable. The official URL builder guide lists the required parameters.
Write one hypothesis: “For operations leaders, a specific scheduling-cost angle will produce more qualified bookings than a general automation angle.” Change one meaningful variable. Pick one or two success metrics before the test begins. Google Ads recommends the same discipline for experiments because changing several variables makes the result hard to interpret. It also advises keeping records so later tests can build on what was learned. See Google Ads experiment guidance.
A one-page AI marketing assistant brief template.
Business outcome: [one measurable result, target, deadline and cost or quality boundary]
Audience: [role, company or customer type, situation, problem, buying stage, geography, language and exclusions]
Offer and action: [what they receive, why it matters now, final URL and one call to action]
Approved evidence: [links to product facts, research, customer evidence, prices and prior results; source required beside every factual claim]
Claim boundaries: [banned claims, required qualifications, regulatory limits, privacy rules, competitor rules and approval owner]
Channels and formats: [placement, dimensions or character limits, language, reading level, number of variants and accessibility requirements]
Brand pattern: [three tone rules, one accepted example, one rejected example and required terminology]
Tracking: [primary business event, diagnostic metrics, UTM taxonomy and analytics owner]
Test: [hypothesis, one changed variable, control, traffic or budget split, review date and decision rule]
Definition of done: [the exact files, table or draft package to return; checks to run; items that must remain drafts; named approver]
A complete brief for a hypothetical B2B consultation campaign.
Outcome: generate 20 qualified consultation bookings by October 31 from operations leaders at logistics companies with 20 to 200 employees in Germany, with cost per qualified booking at or below €180. Audience: teams coordinating driver changes through email and spreadsheets, already researching scheduling software, English or German, excluding recruitment firms and businesses with fewer than five drivers.
Offer: a 25-minute workflow review that maps the current scheduling handoffs and returns one written bottleneck report. Evidence: use only the supplied service page, current price sheet and two approved anonymised workflow diagrams. Do not claim a time saving, error reduction, guarantee or customer count. Mark any missing proof.
Deliverables: four search-ad angles with three headlines and two descriptions each, two landing-page hero variants, and one follow-up email draft. Every angle must name the audience problem and use the same final URL. Return a table with angle, hypothesis, evidence source, channel limits, CTA and tagged URL.
Tracking and test: use lowercase source and medium values, campaign “de_logistics_workflow_q4”, and content values that identify the angle. The primary event is a qualified booking confirmed in the CRM. Test the scheduling-cost angle against the general workflow-review control, changing the message only. Review after the planned sample threshold, not after a fixed number of days. Nothing publishes and no budget changes until the campaign owner approves the package.
Review the first output like a campaign operator.
Check the package against the brief, line by line. Does every asset serve the same outcome and audience? Does every factual claim link to approved evidence? Are the final URL, offer, CTA and tracking values consistent? Do the variants represent different hypotheses, or are they synonyms? Are channel limits, accessibility requirements and approval gates intact? Can the measurement owner tell which creative produced the business event?
Reject the package if it hides a missing source behind confident language, changes several test variables at once, invents urgency, or optimises for a proxy that the business did not choose. Return the failed checklist items, not “make it better.” The assistant now has an observable correction target.
AI marketing assistant FAQs.
What should an AI marketing assistant brief include?
Include one business outcome, one audience, the offer, approved evidence, channel and format, source assets, claim limits, approval rules, tracking names, deadline, and the exact deliverable you expect.
How long should an AI marketing brief be?
Long enough to remove ambiguity and short enough to review. One structured page is usually better than a long brand deck. Link the source materials instead of pasting every document into the brief.
Should the brief include a target audience?
Yes. Name the person, situation, problem, buying stage, geography, language, and any exclusions. A broad label such as small businesses leaves the assistant to invent a buyer.
How do I stop an AI assistant from inventing marketing claims?
List the only evidence it may use, require a source beside every factual claim, ban unsupported superlatives, and send legal, medical, financial, performance, and comparison claims through human approval.
What marketing metric should go in the brief?
Use the closest business result the campaign can influence, such as qualified leads, completed purchases, booked consultations, or trial activations. Add one or two diagnostic metrics, but do not let clicks replace the business outcome.
Should an AI marketing assistant publish automatically?
Only inside a clearly approved rule. New claims, new audiences, paid budget, customer messages, account changes, and public publication usually deserve a human gate. Low-risk internal drafts can move faster.
How many creative variations should I request?
Ask for enough options to test one hypothesis, not a pile of interchangeable copy. Four to six meaningfully different angles are often more useful than twenty shallow rewrites.
How do I test whether the brief worked?
Review the first output against a written acceptance checklist, then run a controlled experiment that changes one variable and uses a preselected success metric. Keep the brief and result together for the next round.
The guidance behind the template.
OpenAI: Best practices for prompt engineering with the OpenAI API
FTC: Advertising and Marketing on the Internet
FTC: Statement of Policy Regarding Advertising Substantiation
Google Ads: Asset group best practices
Google Ads: Build an asset group
Google Analytics: URL builders
Bring one marketing workflow to a live demo.
Use the template above on a real campaign, then bring the brief and the source material. The useful question is whether the workflow can return a checkable result inside your approval rules.
For the operating model behind that workflow, read AI agent vs AI assistant and the AI operator blueprint.
Book a live demo