How to Brief AI on Your Ad Campaigns: A Prompt Guide for Media Buyers

The difference between AI that gives you useful analysis and AI that gives you generic non-answers is almost entirely in how you frame the request. Most media buyers who try AI tools and give up within a week do so because they typed a vague question, got a vague answer, and concluded the tool doesn't work.

The tool works. The brief was the problem.

This guide covers how to structure AI prompts for the four core campaign management tasks: performance analysis, budget optimization, campaign creation, and client reporting. Every example is immediately usable - copy, adapt, paste.

The single most important principle: context before request

AI models don't have access to your account history unless you give it to them. A prompt like "analyze my Meta Ads performance" gives the model nothing to work with. You'll get a generic framework for how one might analyze Meta Ads performance - not an analysis of yours.

The structure that works:

Context (what the model needs to know) + Task (what you want it to do) + Format (how you want the output) + Constraint (what to avoid or prioritize)

Part 1: Performance analysis prompts

Prompt for weekly performance summary

Context:
- Client: [name or category - e.g. "e-commerce fashion brand, PLN market"]
- Platform: Meta Ads
- This week: Spend €3,400 / Budget €3,800 / ROAS 2.6x / CPA €18.40 / Conversions 185
- Last week: Spend €3,600 / Budget €3,800 / ROAS 3.1x / CPA €15.20 / Conversions 237
- Campaign changes made this week: Launched two new Prospecting ad sets with broad audiences on Monday. Paused one Retargeting ad set with CPA >2x target on Wednesday.

Task:
Write a 4–5 sentence performance summary for a client. Explain what happened, why, and what we're doing about it.

Format:
Plain language, no jargon. Client is not a marketing expert. One paragraph.

Constraint:
Do not make it sound alarming. The ROAS drop is expected given the new Prospecting campaigns in learning phase.

Why this works: You've given the model the numbers, the changes you made, the context (new campaigns in learning phase), and the tone constraint (not alarming).

Prompt for anomaly investigation

Context:
- Meta Ads account for a D2C supplement brand
- Average daily spend: €850, average daily ROAS: 3.4x
- Yesterday: spend €820, ROAS 1.7x - 50% below normal
- No campaign changes made yesterday
- No creative pauses or budget changes
- iOS 17.5 update rolled out yesterday in EU markets

Task:
Give me 5 possible explanations for a sudden 50% ROAS drop with no campaign changes, ordered by likelihood. For each, describe how I would verify it.

Format:
Numbered list. Each entry: hypothesis, how to verify in Meta Ads Manager, estimated time to verify.

Constraint:
Do not include "check your targeting settings" - I've already verified those are unchanged.

Why this works: Specific numbers, specific context (iOS update), and a request for ordered hypotheses with verification steps. You'll get actionable investigation steps, not a generic "here are reasons ROAS might drop" list.

Prompt for cross-platform budget allocation analysis

Context:
- Total monthly ad budget: €15,000
- Current allocation: Meta €8,000, Google €5,000, TikTok €2,000
- Last 30-day performance:
  Meta: ROAS 2.8x (7-day click, 1-day view attribution)
  Google Brand Search: ROAS 9.2x
  Google Non-Brand Search: ROAS 3.8x
  Google Shopping: ROAS 4.1x
  TikTok: ROAS 1.6x
- Business context: Q3 launch of a new product line, want to maximize reach to new audiences

Task:
Recommend a budget reallocation for next month that balances efficient performance with reach goals for the new product launch. Explain the trade-offs.

Format:
Recommended allocation (table), explanation of each change and the trade-off, one key risk to watch.

Constraint:
Don't suggest cutting TikTok entirely - client has brand awareness goals there.

Part 2: Optimization prompts

Prompt for identifying which campaigns to scale

Context:
- Google Ads account, e-commerce
- I have 8 campaigns. Here is last 30 days performance data:
  [paste table: Campaign | Spend | Conversions | CPA | ROAS | CPC | Conv Rate]
- My target CPA is €22. My budget cap is €18,000/month total.
- Campaigns in learning phase (launched <2 weeks ago): [name them]

Task:
Identify which campaigns I should increase budget for, which I should leave as-is, and which I should consider pausing or restructuring. Give a specific recommended budget for each.

Format:
Three-section list: Scale (with budget recommendation), Maintain (with reason), Review/Pause (with reason). Then a total that sums to €18,000.

Constraint:
Do not recommend pausing campaigns launched <2 weeks ago - they're still in learning phase.

Prompt for ad copy generation

Context:
- Product: Project management software for architecture firms
- Target audience: Architects and project managers at firms with 10–50 people
- Pain points: Projects going over budget, version control on drawings, client communication chaos
- Tone: Professional but not corporate. Direct. They're busy.
- Platform: Meta Ads, single image ad
- Ad objective: Lead generation (free trial signup)
- What's worked before: Headline "Stop managing projects in email" - 4.2% CTR

Task:
Write 8 headline/primary text combinations for a Meta Ads lead generation campaign. Each combination should address a different pain point or angle.

Format:
Numbered list. Each entry: Headline (max 40 chars) / Primary text (max 125 chars) / Angle label.

Constraint:
Avoid generic "save time" language. Be specific to architecture. No exclamation marks.

Part 3: Client reporting prompts

Prompt for monthly performance narrative

Context:
- Client: Online furniture retailer, mid-market positioning
- Month: May 2026
- Meta Ads: Spend €12,400 / ROAS 3.2x / Revenue €39,680 / Conversions 318
- Google Ads: Spend €8,600 / ROAS 4.1x / Revenue €35,260 / Conv 247
- Total: Spend €21,000 / Revenue €74,940 / Blended ROAS 3.57x
- vs. April: Blended ROAS was 3.21x (+11%), spend was €19,800 (+6%)
- Key events: Bank holiday week had unusually high conversion rate. Creative refresh on May 14 - new lifestyle photography. Shopping campaign restructured May 8.
- Target for May: ROAS 3.0x - achieved.

Task:
Write a monthly performance report narrative for the client. They are not technical. They want to know: did it work, why, and what's next.

Format:
Three paragraphs: (1) Results headline + verdict, (2) What drove performance, (3) What we're doing in June. Then a bullet list: "Key actions in May" and "June priorities."

Constraint:
No internal jargon (no "Creative refresh" - say "new ad photos"). Attribute the bank holiday performance to seasonal context, not campaign management skill.

Part 4: Strategic planning prompts

Prompt for campaign structure recommendation

Context:
- New client: SaaS tool for HR managers
- Monthly budget: €5,000 (Meta Ads only, to start)
- Product: €89/month subscription, 14-day free trial, no credit card
- Target audience: HR managers and HR directors at companies with 50–500 employees
- Geography: Poland + Germany + Netherlands
- Business goal: 50 free trial signups per month at CPA ≤ €100

Task:
Recommend a Meta Ads campaign structure for the first 60 days. Include campaigns, objectives, audience strategy, and budget allocation per campaign. Explain the rationale for the structure.

Format:
Campaign structure table (Campaign | Objective | Audience | Budget | Goal), then 3–4 paragraphs of rationale explaining the phased approach.

Constraint:
This is a B2B SaaS with a long consideration cycle. Don't recommend a conversion campaign from day 1 - the pixel needs data first.

Common mistakes that kill prompt quality

• Too vague. "What should I do about my Meta Ads?" → no data, no context, no specific question. You'll get generic advice.
• Missing numbers. "ROAS dropped a lot this week." → How much? From what baseline? What's the spend? Without numbers, the model is guessing.
• No business context. "Should I increase budget?" → Where in the funnel? What's the goal? What's the target CPA? Budget decisions without goals are meaningless.
• Forgetting constraints. Not telling the model what not to do results in outputs that include things you already know are wrong for your situation. Add "don't suggest X because Y" - it saves a back-and-forth.

Frequently asked questions

Can I use these prompts in Consultad's AI chat? Yes - Consultad's AI chat is specifically designed for campaign management context. When your accounts are connected, you can skip the "Context" section of most prompts because the AI already has access to your real account data. You'd write: "Identify which campaigns I should scale next month, given my €18,000 budget cap and €22 CPA target" - the data context comes from the live account.

How specific should I be about client details in prompts? Use categories rather than identifying details when possible: "e-commerce fashion brand" rather than the brand name. The quality of the output depends on accuracy of the numbers and context, not the client's name.

What if the AI gives me an answer I disagree with? Push back with specifics: "That recommendation doesn't account for the fact that this campaign is in learning phase - what would you recommend instead?" AI tools respond well to specific disagreement. Vague disagreement ("I don't think that's right") produces a vague revision.

How long should a prompt be? As long as it needs to be to provide the context required. The examples in this guide are on the longer side - that's intentional.