Creative strategy clients do not ask to see designs. They ask to see three ads, the spend behind each, and an explanation of why it worked. Here is how to run an AI agent that assembles that from work you already have — or from competitor teardowns when you do not.
The evidence bar on a creative strategy post is written in the listing: three brands, three ads, spend attached, and why each worked. A gallery of polished visuals answers a question nobody asked, and the application gets skipped on the strength of it.
Postings in this niche ask for named brands, ads you personally contributed to, the spend behind them, and an account of why each worked. That is four pieces of evidence, not a gallery. Most applicants supply the gallery and are filtered without being read.
Freelancers frequently put a range on a project they are unsure about, and it fails in a specific way: a client who asks in the call finds the discrepancy and the whole application loses credibility. An honest 'this one I can attribute, these two I contributed to' holds up. An inflated number does not.
If you have never had account access, the honest position is competitor-based analysis — the ad, the hook, the angle, the format, and the reasoning. That is what the research half of the job actually is. What disqualifies it is presenting it as your own performance, not presenting it at all.
Copy it once and it takes the portfolio apart: which ads you actually contributed to, what spend sits behind each, what the evidence bar on a live posting demands, and what is missing before you apply.
Three stages, one goal. The agent builds the portfolio once, then checks it against each live posting's evidence bar, so you apply knowing whether you clear it rather than finding out from silence.
Three recent postings, quoted for what each asks for, reduced to the overlap and the outliers. That becomes the specification the portfolio is built against.
Each field on each project is labelled for whether it can actually be evidenced. An unevidenced number is a gap, not a rounding error — it gets tested in the call and it costs the whole application.
Where the evidence is genuinely missing, the agent writes the teardown the research half of the job actually produces — hook, angle, format, reasoning, adaptation — and labels it as competitor analysis rather than passing it off as a result.
Are you ready?
All without you getting tasks done on manual, missing out the opportunity you want, or putting your hard work in vain.
Try it with Zaira nowThe questions freelancers ask about building proof for this niche.
Three entries, each with: the brand, your exact role, the ad itself, the spend behind it where you can evidence it, and why it worked. That is what the postings ask for. A gallery of final designs does not answer any of those five questions and reads as a different discipline entirely.
Yes, built from teardowns. Take ads from the Meta Ad Library in the target category, and for each write the hook, the angle, the format, and your reasoning for why it is still running. Label them as competitor analysis. Clients read this as research discipline; what disqualifies it is presenting it as your own performance.
The category and the numbers, yes; the brand name, usually not. These are public ads running in the open, and clients expect to see which brand. What you cannot share is confidential strategy or unpublished performance data. Where you must redact, say so plainly rather than blurring — a visible redaction reads as professionalism, an unlabelled one reads as carelessness.
Three is the number the postings ask for, and three well-evidenced ads beat ten decorative ones. If you have more, keep three per category rather than a long gallery — clients ask for three, and a wall of assets makes it harder to find the three that answer their brief.
No, but it needs to match the thinking. A supplement ad proves you understand health claims and compliance; a B2B ad proves you understand long consideration cycles. The transferable proof is that you can diagnose an ad and explain why it worked — that is the same skill in every category, and stating which parts transfer is part of the application.
Upwork MCP server connection