Competitor Ad Intelligence

Competitor Ad Intelligence

Composite teardown: Meta scrape + Google scrape → creative-pattern analysis → LP/funnel reconstruction → strategic gap/vulnerability analysis → report. Scripts do the deterministic scraping/parsing; you (the agent) do all clustering, scoring, funnel inference, and the teardown narrative.

When to use

• "What ads are my competitors running?" / "Tear down <competitor>'s ad strategy."
• "Find new creative angles" / "Reverse-engineer <competitor>'s paid funnel."
• "What hooks/formats are dominant in our space?" / "Find weaknesses in their ad strategy."

Sub-skills it chains

• google-ad-scraper capability (Google Transparency Center) — call its script by path.
• Bundled scrape_meta_ads.mjs (Meta Ad Library) and fetch_landing_page.py (LP extract).

How to run

One-time browser setup:

cd ${SKILL_DIR}/scripts && npm install && npx playwright install chromium

1 — Scrape ads per competitor (channels: meta / google / both)

# Meta Ad Library (this skill's bundled scraper)
node ${SKILL_DIR}/scripts/scrape_meta_ads.mjs --query "Notion" --country US --max-ads 60 --output ${WORKSPACE}/meta_notion.json

# Google Ads Transparency Center (reuse the google-ad-scraper capability)
node ${SKILL_DIR}/../google-ad-scraper/scripts/scrape_google_ads.mjs --domain notion.so --max-ads 50 --country US --output ${WORKSPACE}/google_notion.json

Captured per Meta ad: libraryId, visualType, adText, cta, landingUrl, startDate, platforms, daysRunning. daysRunning (first-seen → still-running) is the key "what's working" signal — longer runs weight higher in the teardown.

If a library blocks the scraper: if APIFY_API_TOKEN is set → use the Apify ad-library actor (or the Apify fallback in google-ad-scraper) for the blocked library; if not → degrade to a site:facebook.com/ads/library "<competitor>" / site:adstransparency.google.com "<domain>" web search (default; less structured, lower coverage — note it). That degrade search itself can use the optional SERP upgrade: if DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) is set → structured SERP API; else → the agent's keyless web search.

2 — Cluster hooks, tabulate formats & CTAs (you, the agent)

Read the scraped JSON. Classify each ad's hook into Fear / Outcome / Question / Social-proof / Contrarian / Empathy / Product-led; build hook-distribution, format-distribution, and CTA-taxonomy tables per competitor. Surface the longest-running ads.

3 — Fetch each unique landing page

# Collect unique landingUrl values into urls.txt, then:
python3 ${SKILL_DIR}/scripts/fetch_landing_page.py --urls ${WORKSPACE}/urls.txt --output ${WORKSPACE}/lps.json

Returns hero/subhead/primary-CTA/proof/form-field-count/page stats per LP. For JS-rendered or gated LPs, escalate that URL to a Playwright fetch (reuse scrape_meta_ads.mjs's browser pattern) or capture a screenshot so visual continuity reflects what a clicker sees.

4 — Cluster into campaigns, infer funnels & budget, write the teardown (you)

Cluster ads into inferred campaigns by LP destination + theme. Per campaign infer intent, persona, positioning bet, hook, conversion path, longevity, and A/B signals; infer platform budget allocation from ad volume × platform. Then do the strategic pass: creative white-space, overcrowded angles, format gaps, and vulnerabilities. In deep mode, add historical positioning change by fetching web.archive.org snapshots of competitor LPs (fetch_landing_page.py --url "https://web.archive.org/web/<ts>/<lp>") and propose counter-plays with headline/body/LP strategy.

5 — Render

Write competitor-ad-intel-<YYYY-MM-DD>.md to ${WORKSPACE} and attach it to the Agent Teams channel. Optionally persist scraped ads (dedup by libraryId/creativeId + LP) so monitoring across runs shows true new vs. retired campaigns.

Outputs

competitor-ad-intel-<YYYY-MM-DD>.md — coverage summary, Meta + Google ad analysis (hook distribution, longest-running ads, CTA taxonomy, format distribution), per-campaign funnel map, budget-allocation inference, creative gap analysis, vulnerability report, and (deep mode) counter-plays.

Credentials / env

• Required: none for the scripts — both ad libraries are public; Playwright + a proxy carry the scrape. Hook clustering, funnel inference, and the teardown are the agent's reasoning (no LLM key needed in scripts).
• Optional (each with a keyless default fallback):
  • APIFY_API_TOKEN — if set → Apify ad-library actor when a library page is too anti-bot; else → the keyless Playwright scrapers, degrading to site: web search (default).
  • SUPABASE_URL/SUPABASE_KEY — if set → persist ad history for cross-run new-vs-retired monitoring; else → single-run workspace JSON (default).
  • DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) — if set → structured SERP API for the site: degrade/archive searches; else → the agent's keyless web search (default).
  • HTTPS_PROXY — Robomotion Proxy for the Playwright scrapers.

Notes & edge cases

• Meta Ad Library and Google Transparency Center are heavy JS SPAs with active anti- scraping — randomize timing, rotate the proxy, set the library country to your target market. If blocked, degrade to site: snippets and note lower coverage.
• Apify Meta-Ad-Library actors are unreliable under Meta's countermeasures — last resort.
• Longevity is the load-bearing signal; capture daysRunning per ad.
• Dedup ads by library/creative id + LP across runs so monitoring is accurate.