Ad To Landing Page Auditor

Ad → Landing Page Auditor

Composite: ad inventory parse → LP fetch/extract → message-match scoring → friction analysis → prioritized fixes + rewrites. Scripts handle the deterministic parse + LP extraction; you (the agent) do the 6-dimension message-match scoring, friction classification, prioritization, and rewrites.

When to use

• "Why are my ads getting clicks but no conversions?"
• "Audit my ad-to-landing-page flow" / "Check message match on our campaigns."
• "My conversion rate is low — help me figure out why."

How to run

1 — Parse ads into an inventory + collect unique LP URLs

# From an ad-platform CSV export:
python3 ${SKILL_DIR}/scripts/parse_ads.py --csv ${WORKSPACE}/ads_export.csv \
  --urls-out ${WORKSPACE}/urls.txt --output ${WORKSPACE}/inventory.json

# Or from a JSON ad list:
python3 ${SKILL_DIR}/scripts/parse_ads.py --json ${WORKSPACE}/ads.json --urls-out ${WORKSPACE}/urls.txt

Each row: {ad_id, platform, headline, headlines[], body, cta, landing_url, conv_rate}. --urls-out writes the deduped LP list so each LP is fetched once.

2 — Fetch + extract each unique landing page

python3 ${SKILL_DIR}/scripts/fetch_landing_page.py --urls ${WORKSPACE}/urls.txt --output ${WORKSPACE}/lps.json

Per LP: hero headline, subhead, primary CTA, above-fold-CTA flag, benefit lists, proof signals, image/asset count (load-weight proxy), form_field_count, has_video, nav-link count, status. If an LP returns ok:false (failed to load), mark its friction "Unknown" and flag it as a Critical fix — a broken destination — rather than scoring it.

For JS-rendered / A/B-cloaked / consent-walled LPs, escalate that URL to a Playwright fetch + screenshot (reuse the browser pattern in ../competitor-ad-intelligence/scripts/scrape_meta_ads.mjs) so the audit reflects what a real clicker sees. Use a proxy + geo matching the ad's target region. For an LP behind heavy anti-bot: if APIFY_API_TOKEN is set → fall back to an Apify fetch (last resort); if not → the keyless urllib + Playwright path is the default and remains the only LP fetch.

3 — Score message match (you, the agent)

For each ad→LP pair, score 6 dimensions 1–10 (→ /60): promise continuity, language match, visual continuity, CTA alignment, specificity match, emotional match. Cite the specific mismatch per dimension. Visual continuity is N/A for plain search ads — exclude it from the /60 average rather than penalizing (use the screenshot when one exists).

4 — Friction analysis (you, over the extracted stats)

Per LP, classify Red/Yellow/Green on: load weight (image/asset count), form length (form_field_count), CTA clarity, above-fold conversion, proof placement, nav distraction, mobile issues. Estimate conversion impact.

5 — Prioritize + rewrite (you)

Rank disconnects by severity × conversion impact into Critical / Important / Nice-to-have. Draft a matched ad-headline or LP-headline rewrite for the worst-matching pair.

6 — Render

Write ad-lp-audit-<YYYY-MM-DD>.md to ${WORKSPACE} and attach to the Agent Teams channel.

Outputs

ad-lp-audit-<YYYY-MM-DD>.md — per ad→LP Message Match Score (/60) with dimension breakdown + disconnect + fix; landing-page friction report; prioritized fixes with estimated conversion impact; rewrite suggestions for the worst pair.

Credentials / env

• Required: none — parse_ads.py and fetch_landing_page.py are keyless. The scoring, friction classification, and rewrites are the agent's reasoning (no LLM key in scripts).
• Optional (each with a keyless default fallback):
  • HTTPS_PROXY — Robomotion Proxy for geo-matched / escalated LP fetch.
  • APIFY_API_TOKEN — if set → Apify fetch as a last resort for an LP behind hostile anti-bot; else → keyless fetch_landing_page.py (urllib) escalating to Playwright fetch + screenshot (default).

Notes & edge cases

• Prefer the keyless fetch_landing_page.py for static / server-rendered LPs; escalate to Playwright only when the hero/CTA render client-side or the page is consent/geo-walled.
• Watch for LPs that A/B-rotate or cloak by referrer — capture a screenshot via Playwright so the audit reflects the real post-click experience.
• A failed LP load is a Critical finding (broken destination), not a zero score.