From Competitor Ad to Creative Brief in 20 Minutes (API + LLM)

Your media buyer drops a competitor ad into Slack with three fire emojis and the words "we need our version of this." Everyone agrees. Three weeks later nothing has shipped, because nobody converted that screenshot into something a designer could actually build from. The competitor ad to creative brief step is where creative pipelines quietly die. Not in research. Not in production. In translation.

TL;DR: The translation step is now almost fully automatable. Resolve the brand and pull its longest-running ads through the AdLibrary API (1 credit per search), run AI enrichment to get a structured teardown with a scene-level transcript and persuasion analysis (1 credit), feed that JSON to an LLM with a fixed brief template, then spend five minutes editing what the model cannot know about your brand. Two credits and roughly 20 minutes per brief, on the Business plan (€329/mo).

This guide walks the full pipeline with working code, a complete worked example built on a fictional brand, and the exact LLM prompt template. It also marks the two steps that must stay human, because a brief that ships a clone is worse than no brief at all.

The gap between finding a winner and briefing one

Most teams have solved discovery. Between native transparency surfaces and how to find competitor ads workflows, locating a strong competitor creative takes minutes. The swipe file fills itself.

What stays unsolved is the handoff. A screenshot in a Slack channel carries maybe 10% of what a designer or editor needs. It shows the visual. It hides the hook structure, the offer architecture, the proof elements, the pacing of the first three seconds, and the reason the ad survived 80 days of spend when its siblings died in two weeks.

So the strategist becomes the bottleneck. Someone has to watch the ad five times, transcribe it, infer the targeting, reverse-engineer the persuasion logic, and write all of that up in a format production can execute. Done properly by hand, that is an afternoon per ad. The competitor ad analysis manual documents the full manual method, and it is rigorous work. Multiply it by a weekly creative cadence across three competitors and the math stops working.

The result is predictable. Briefs get thinner. Production gets vaguer direction. The creative strategist workflow collapses into "make something like this" plus a link.

None of this is an argument against taste. It is an argument against spending taste on transcription. The judgment calls that actually move performance, which angle to adapt and which to leave alone, get maybe ten minutes of a strategist's day under the manual system, because the other three hours went to watching ads at 2x speed with a notepad.

Why competitor ad to creative brief is a translation problem

A creative brief is a structured document. It has known fields: objective, audience, message hierarchy, mandatories, format specs, references. A competitor ad is unstructured media. Video, image, copy, and context fused together.

Converting one into the other is a translation between two well-defined formats, and that is precisely the shape of problem LLMs are good at. The model does not need to invent strategy. It needs to restructure evidence that already exists into fields that already exist.

The catch is input quality. Hand an LLM a raw video URL and a vague instruction and you get generic mush. Hand it a structured teardown, with a timestamped transcript, the identified hook, the offer mechanics, and the runtime evidence, and you get a draft brief that is 80% done. The competitor ad to creative brief conversion fails or succeeds on what you feed the model, which is why the pipeline below spends two of its three automated steps on data acquisition and only one on generation.

If you have never written briefs in a fixed format before, read how to brief a creative team for Meta ads first. The automation assumes you know what good output looks like.

The 20-minute pipeline at a glance

Five steps, three of them automated:

A note on data sources before the code. Meta's free Ad Library API is the original transparency interface, and for political and social-issue ads it remains the canonical one. Google publishes its equivalent through the Ads Transparency Center, TikTok through its Commercial Content Library, and LinkedIn exposes ad data via its API program. All of them are free. All of them are also siloed, and none attaches performance context to the creative.

Meta's free API is fine for one platform. The moment you want TikTok, YouTube, or LinkedIn ads in the same query, plus signals that separate a test from a scaled winner, you need a paid layer on top. That layer here is API access on AdLibrary, which searches eleven platforms from one endpoint and returns estimated spend, impression bands, heat score, and runtime on every ad. This pipeline is built on it.

Minutes 0–5: find the winning ad through the API

You usually start with a brand name, not an ad. Resolve it first. The advertiser lookup is free and maps a name to its platform IDs across Meta, Google, and LinkedIn in one call:

curl -G "https://adlibrary.com/api/advertisers/search" \
  -H "Authorization: Bearer adl_your_api_key" \
  --data-urlencode "q=glowdose" \
  --data-urlencode "country=US"

The response includes a best_match with the Meta page ID. Now pull the brand's ads, sorted by runtime. Runtime is the strongest single winner signal in ad intelligence: advertisers kill losing creatives within days, so anything still spending after 60+ days has earned its budget. Sort descending on days live:

curl -X POST "https://adlibrary.com/api/search" \
  -H "Authorization: Bearer adl_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "keyword": "glowdose",
    "metaPageId": "104283956710",
    "appType": "3",
    "adsType": ["2"],
    "sortField": "-days",
    "daysBack": 90
  }'

One credit per search page. The response carries a total count plus an array of ads, each with copy, creative URLs, engagement counts, an estimated spend figure, and days_count. Two reading rules. Impressions on Meta come back as ranges, so treat them as bands rather than precise counts. Spend is always an estimate, never an advertiser-reported figure. If you want to sanity-check what a given runtime implies in budget terms, the ad spend estimator gives you a quick cross-reference.

Pick your target ad with a simple filter: longest runtime, video format, and engagement that is not embarrassing. For recurring competitor work, wire this exact query into a schedule instead of running it by hand. How to monitor competitor ads covers diff-detection, and the competitor ads research playbook covers the broader cadence.

One more selection rule, because it trips people up. Do not pick the ad you personally like. Strategists fall for clever creative, then brief a clone of something the market already ignored. The data disagrees with taste more often than feels comfortable. An ugly static ad at 70 days of runtime beats a beautiful video at 12, every time, because the advertiser's own budget already ran the test for you. Trust the runtime column over your eye.

Grab the winner's ad_key and its media URLs. That is all step 2 needs.

Minutes 5–8: enrich the ad into structured analysis

This is the step that used to be your afternoon. The enrichment endpoint runs a full creative teardown on one ad and returns it as structured data:

curl -X POST "https://adlibrary.com/api/enrichment" \
  -H "Authorization: Bearer adl_your_api_key" \
  -H "Content-Type: application/json" \
  -d '{
    "ad": {
      "ad_key": "meta_724455910283341",
      "platform": "facebook",
      "advertiser_name": "Glowdose",
      "video_url": "https://...",
      "body": "Your skin barrier called...",
      "landing_page_url": "https://glowdose.example/serum"
    }
  }'

One credit for text and any video up to 180 seconds. Longer videos cost more, scaled by duration. What comes back is the input layer your LLM actually needs:

• A scene-level transcript. Everything spoken and shown, timestamped, so the model can quote the ad instead of hallucinating it.
• The strategic read. Product, funnel stage, awareness stage, inferred audience, core message.
• The persuasion teardown. The hook and why it stops the scroll, the offer structure, the proof stack, the psychological levers in play, each point quoted from the ad itself.
• A replication-oriented breakdown tuned to the detected format, whether that is a static image, a UGC-style video, or a cinematic cut.

The AI ad enrichment feature is the same engine the AdLibrary app uses, exposed as an endpoint, which matters because you get the analysis as JSON you can pipe forward rather than prose you have to copy out of a UI.

Save the full enrichment response to a file. It becomes the evidence block in your prompt.

Minutes 8–14: the LLM prompt that writes the first draft

Now the generation step. The prompt has a fixed shape: role, evidence, template, constraints. Paste the enrichment JSON into the evidence slot and your brand facts into the constraints slot. Here is the template in full:

You are a senior creative strategist writing an internal creative brief.

EVIDENCE (structured teardown of a competitor ad — quote it, never invent):
<paste the full enrichment JSON here>

PERFORMANCE CONTEXT:
- Runtime: <days_count> days and still live
- Estimated spend: <estimated figure> (estimate, not reported)
- Format: <video/image/carousel>, <duration>, <aspect ratio>

OUR BRAND (override anything in the evidence that conflicts):
- Product: <your product, one line>
- Audience: <your ICP, one line>
- Voice: <three adjectives + one "never do">
- Proof we own: <real reviews, stats, certifications>
- Offer we can run: <your actual offer>
- Hard exclusions: <claims we cannot make, competitors we cannot name>

TASK: Write a creative brief with EXACTLY these sections:
1. WHY THIS REFERENCE WON — 3 bullets, each quoting the evidence
2. THE ANGLE WE ARE TAKING — our version of the underlying angle,
   adapted to our product. State what we are NOT copying.
3. HOOK — 3 options for the first 3 seconds, written out word for word
4. STRUCTURE — beat-by-beat outline with target timestamps
5. MESSAGE HIERARCHY — one primary claim, max two support claims,
   each mapped to proof we own
6. MANDATORIES — voice rules, legal exclusions, logo/CTA placement
7. FORMAT SPECS — duration, aspect ratio, platform, captions y/n
8. REFERENCE NOTES — what the editor should watch in the original
   and what to deliberately do differently

RULES:
- Every claim in sections 1 and 8 must quote the transcript or analysis.
- Section 2 must name the difference between our version and theirs.
- Write hooks in our voice, not the competitor's.
- 600 words maximum. A brief nobody reads is not a brief.

Two notes on running it. First, model choice matters less than evidence quality here. Any current frontier model produces a usable draft from a good enrichment payload, because the task is restructuring rather than reasoning. Some prompt engineering basics still apply: the EXACTLY in the task line and the quote-the-evidence rule do most of the anti-hallucination work.

Second, this scales past one-at-a-time chat. Teams run this prompt inside agent setups where the model calls the API itself, which is the pattern documented in Claude Code AdLibrary API workflows. If you want the research layer available to every agent session you run, building an AdLibrary MCP server wraps these same endpoints as tools.

Minutes 14–18: the human edit pass

The draft will be structurally complete and strategically naive. Your four minutes go to the things the model cannot know.

Differentiation. The model adapted their angle to your product, but only you know whether that angle is already saturated in your category or off-strategy for your positioning. Section 2 of the brief is where you earn your salary. Sharpen what "we are NOT copying" actually means. A creative angle borrowed wholesale produces an ad that performs like a rerun.

Claim safety. The competitor may be making claims you legally cannot. The model only knows the exclusions you listed. Read the message hierarchy against your actual substantiation, not against vibes.

Voice. LLM hook drafts trend toward competent and flat. Rewrite at least one of the three hooks yourself. Keep the model's structure and replace its words where your value proposition deserves better language.

Offer truth. Verify the offer in the brief is one your margin survives. Models cheerfully brief a discount you never approved.

What you do not do in this pass is restructure. The sections are fixed for a reason. Editors and designers learn the format once and then read every brief in 90 seconds. If you find yourself rewriting structure weekly, fix the template, not the briefs. The creative strategist role does not get automated away by this pipeline. It gets concentrated into these four minutes.

Handoff is the last two minutes. The brief goes into your PM tool with the original ad URL, the enrichment JSON attached for the curious, and a deadline. Done. A brief delivered the same day the winner was spotted, while the manual version would still be a calendar block next Tuesday.

Worked example: Fernweh Botanicals vs. Glowdose

Both brands are fictional. The numbers are realistic but invented for illustration.

You run creative strategy at Fernweh Botanicals, a DTC skincare brand doing €180k/mo on Meta. Your sharpest competitor is Glowdose, whose barrier-repair serum keeps showing up in your audience's feeds.

Minute 0. Advertiser lookup on "glowdose" returns a best match with Meta page ID 104283956710 and 41 active ads. Free call.

Minute 2. Search with metaPageId, video only, sorted by -days. The top result jumps out: a UGC-style video at 84 days runtime, estimated spend in the mid five figures, impressions in the 1M+ band, 2,300 reactions. The next-longest video sits at 31 days. An ad outliving its siblings by 53 days is not luck. One credit.

Minute 6. Enrichment on that ad returns the teardown. The transcript shows the hook is a negative-confession open: "I wasted $400 on serums before I learned what a skin barrier actually is." The analysis flags the structure as problem-agitation against a PAS pattern, a proof stack built on one dermatologist quote plus two before/after cuts, and a soft CTA framed as education rather than discount. Funnel read: cold traffic, problem-aware audience. One credit.

Minute 9. You paste the enrichment JSON and Fernweh's brand block into the template. The brand block lists your proof (4.8 stars across 3,100 reviews, a clinical hydration study), your voice (warm, precise, never clinical-cold), and one hard exclusion (no dermatologist endorsement, you do not have one).

Minute 13. The draft comes back. Section 1 quotes the confession hook and the 84-day runtime as the evidence. Section 2 proposes keeping the wasted-money confession structure but swapping the authority proof for your review volume, since you cannot match the dermatologist. Three hooks, a six-beat structure for a 45-second 9:16 video, message hierarchy mapped to the hydration study.

Minute 15. Edit pass. You kill one hook that leans on a "repair in 7 days" claim your substantiation does not cover. You rewrite the second hook into Fernweh's voice. You tighten section 2 to name the real differentiation: their ad sells the ingredient, your version sells the routine. For the production route, your editor will build it as an AI-assisted UGC shoot, and the AI UGC ads guide covers that half of the pipeline.

Minute 19. Brief lands in Notion, tagged to next week's shoot, original ad linked. Total cost: 2 credits and 19 minutes. The same translation done manually, by the workflow in from ad library research to creative brief in 60 minutes, is excellent work and three times slower, which matters when you brief five of these a week.

The weekly loop is the point. One winner per competitor per week, three competitors, 6 credits, one focused hour. Set up automated competitor ad monitoring to surface the candidates and the hour shrinks further. Teams running this on a schedule typically wire it through n8n recipes so the search and enrichment steps finish before the strategist sits down.

The competitor ad to creative brief template

This is the download section, minus the email gate. Copy the block below into your prompt library and fill the angle brackets per run. It is the same template from minute 8, condensed to its skeleton:

ROLE: senior creative strategist, internal brief
EVIDENCE: <enrichment JSON>
PERFORMANCE: <runtime> / <estimated spend> / <format>
BRAND: <product> / <audience> / <voice> / <proof> / <offer> / <exclusions>

OUTPUT SECTIONS (exactly):
1. Why this reference won (3 bullets, quoted evidence)
2. The angle we are taking (+ what we are NOT copying)
3. Hook (3 verbatim options)
4. Structure (beats + timestamps)
5. Message hierarchy (1 primary, ≤2 support, proof-mapped)
6. Mandatories (voice, legal, CTA)
7. Format specs (duration, ratio, platform, captions)
8. Reference notes (watch this, change that)

RULES: quote-only claims · name the differentiation ·
our voice, not theirs · ≤600 words

Adapt the section list to your shop, then freeze it. The value of a fixed competitor ad to creative brief format compounds weekly. Your editors learn it, your LLM outputs become comparable across runs, and brief quality stops depending on who wrote it at 5pm on a Friday.

What this costs and how to set it up

The arithmetic is short. Each brief consumes 2 credits, one for the search page and one for the enrichment of a standard-length ad. Searches are 1 credit per page regardless of filter count, and a failed search refunds automatically.

Two failure modes are worth budgeting for. Enrichment on a video past 180 seconds costs extra, scaled by duration, so a five-minute brand film runs 3 credits instead of 1. Check the duration before you enrich if you are watching the meter. And a search that comes back empty still costs its credit, because the API did its job even when the competitor had no winners that week. Neither changes the arithmetic much. A heavy month of daily searches across five competitors, plus a couple of enrichments per week, still lands under 200 credits, a fifth of the Business allowance.

API access sits on the Business plan at €329/mo with 1000+ credits per month, which prices a single automated brief at well under a euro of credits. Business also includes free integration help, so if the auth or the response shapes fight you, the team wires it up with you. Keys are created in your dashboard, start with adl_, are shown once, and you can hold up to ten, so give the brief pipeline its own key. Rate limits are 10 requests per minute, which a weekly brief cadence will never notice. Full endpoint behavior lives in the API documentation and implementation guide.

Whether the plan pays for itself is a production-economics question, not a data question. If a strategist afternoon costs you €250 in loaded salary and this pipeline saves four of them a month, the plan is cheap. Run your own numbers against the break-even ROAS calculator for the testing side, and start with API access for the data side. Multi-platform coverage means the same two-credit pipeline works when the winner you spotted is on TikTok or LinkedIn instead of Meta, through the same multi-platform search.

Frequently Asked Questions

How do I turn a competitor ad into a creative brief automatically?

Three API calls and one prompt. Resolve the brand with the free advertiser lookup, pull its longest-running ads via /api/search sorted by -days (1 credit), run /api/enrichment on the winner to get a structured teardown with transcript and persuasion analysis (1 credit), then feed that JSON to an LLM with a fixed brief template. A human edit pass for differentiation and claim safety finishes the brief in about 20 minutes.

What does the AdLibrary API return that Meta's free Ad Library API doesn't?

Meta's free API is built for transparency and is limited to political and social-issue ads in most regions, on Meta platforms only. The AdLibrary API is a paid layer that returns commercial ads across eleven platforms with performance signals attached: estimated spend, impression bands, heat score, and runtime, plus an AI teardown endpoint that converts any ad into structured analysis ready for a brief.

How much does an automated creative brief workflow cost per brief?

Two credits per brief: one for the search page that finds the winner and one for enriching a text ad or video up to 180 seconds. Longer videos cost more, scaled by duration. On the Business plan (€329/mo, 1000+ credits, API access included), a weekly cadence of five briefs uses about 40 credits a month.

Which LLM should I use to write creative briefs from ad analysis?

Any current frontier model works, because the hard part is input quality rather than reasoning. The enrichment endpoint supplies a timestamped transcript and persuasion teardown as structured data, and the model restructures that evidence into brief sections. Strict output formatting and a quote-the-evidence rule in the prompt matter more than the model choice.

Is it ethical to base a creative brief on a competitor's ad?

Studying competitor ads through official transparency surfaces and licensed data is standard competitive research. Cloning is the line. The template forces an explicit "what we are NOT copying" section, and the human edit pass exists to make sure you ship an adapted angle with your own proof and claims, never a copy of someone else's creative.

Ship briefs, not screenshots

The competitor ad to creative brief gap was never a creativity problem. It was a labor problem, and labor problems get automated. The pipeline is two credits, one prompt, and a four-minute edit. What stays human is exactly the part you wanted to spend your time on anyway: deciding which angle deserves your brand's version and making the words sound like you.

Spotted a competitor ad this week that deserved a brief and got an emoji instead? That is your first candidate. Get an API key through API access on the Business plan, or compare tiers on pricing if you want to run manual searches first and automate once the workflow proves itself.