apify-ads-intelligence

Ads Intelligence Cluster

Answer natural language questions about ads, ad libraries, and competitor advertising activity by routing to the right Apify Actor and delivering a synthesized answer.

CLI rules: Always pass --user-agent apify-awesome-skills/apify-ads-intelligence, --json (or the relevant --format flag on datasets get-items), and 2>/dev/null. The --user-agent flag is critical for telemetry — never omit it.

Note on platform coverage

• Meta, Google, TikTok, LinkedIn: real public ad libraries with rich data (creatives, targeting, dates, reach where disclosed).
• X (Twitter): no public ad library exists. Coverage is a best-effort workaround that scrapes a brand's tweets and flags items with non-empty card field or source containing "Ads" as likely promoted. Always include the caveat in synthesis output.

Note on overlap with apify-ecommerce

That skill has an ads-intelligence intent that routes to apify/facebook-ads-scraper for shallow Meta-ad lookups. This skill is the deep dive across all five platforms. If you only need Meta ads as a side detail of an ecommerce question, stay in apify-ecommerce. If ads are the main task, use this skill.

Prerequisites

(No need to check it upfront)

• Apify CLI v1.5.0+ (npm install -g apify-cli)
• jq (recommended for response parsing and filtering; brew install jq on macOS, apt install jq on Linux)
• Authentication via one of:
  • apify login (OAuth, opens browser)
  • APIFY_TOKEN env variable (e.g. export APIFY_TOKEN=... or .env file)
  • Token from Apify Console → Settings → Integrations

Verify auth: apify info --user-agent apify-awesome-skills/apify-ads-intelligence — should show username and userId.

Workflow

Copy this checklist and track progress:

Task Progress:
- [ ] Step 1: Detect intent and select Actor(s)
- [ ] Step 2: Fetch Actor schema
- [ ] Step 3: Ask user preferences (output format, result count, country)
- [ ] Step 4: Run the Actor (or Actors in parallel for cross-platform-audit) and fetch results
- [ ] Step 5: Synthesize a direct answer (not a data dump)

Step 1: Detect Intent and Select Actor

Classify the user's message into an intent, then pick the right Actor.

Intent signals:

If multiple intents detected, ask: "Do you want [intent A] or [intent B]?"

Actor routing — always try Primary first, switch to Fallback only if it fails or returns 0 results:

X (Twitter) heuristic filter — after scraping, flag a tweet as likely promoted if any of the following hold:

• card field is non-empty (website cards / CTAs are commonly attached to promoted tweets)
• source field contains "Ads" (e.g. "Twitter Ads")

Surface results with the explicit caveat: "X has no public ad library; results below are tweets from the brand's own timeline that match promoted-content heuristics. They will miss promoted-only ads that appear in other users' feeds."

Step 2: Fetch Actor Schema

Fetch the Actor summary, input schema, and README:

# Summary (title, description, pricing, stats)

Replace ACTOR_ID with the selected Actor (e.g., apify/facebook-ads-scraper).

Step 3: Ask User Preferences

Before running, ask:

1. Output format:

   • Quick answer (default) — synthesized answer in chat, no file saved
   • CSV — full export saved to disk
   • JSON — full export saved to disk

2. Result count — defaults by intent:

   Intent	Default count
   competitor-ads	30
   keyword-ads	30
   top-creatives	20
   landing-page-audit	50
   cross-platform-audit	15 per platform

3. Country — default US. For TikTok library specifically, default DE (EU-only) and warn the user; for global TikTok use source: creative_center. X routes are global by handle/keyword, no country parameter.

Cost safety: Always set a sensible result limit in the Actor input (e.g., maxResults, resultsLimit, or the equivalent field per Actor schema). Warn the user before runs of 500+ ads — apify/facebook-ads-scraper charges per ad and X primaries charge per tweet.

Step 4: Run the Actor and Fetch Results

Two steps: run the Actor (blocks until done), then fetch dataset items in the requested format.

Run the Actor — returns run metadata as JSON; extract defaultDatasetId for the next step:

From the output use .id (run ID), .status (should be SUCCEEDED), and .defaultDatasetId.

Fetch results — pick the variant based on the user's preference:

Other --format options: jsonl, xlsx, xml, rss, html. Use --offset N to paginate large datasets.

Tip: for anything more than a quick peek, save the dataset to a local file first (with > file.json / > file.csv) and run further analysis from disk. apify datasets get-items always streams over the network, so piping it straight into jq re-downloads the whole thing every iteration.

Cross-platform audit (parallel runs): For cross-platform-audit, kick off Meta + Google + TikTok + LinkedIn primaries in parallel by backgrounding each apify actors call ... invocation with & and calling wait before fetching results. Example:

Combining with jq for quick extraction:

Treat jq as a complement to apify datasets get-items, not a replacement: server-side --limit / --offset / --format keeps cost and bandwidth down. Use jq on a sample item or on a file you already saved.

Step 5: Analyze Results and Deliver Answer

Synthesize, don't dump. Patterns by intent:

Suggested follow-ups — keyed off the intent that just ran:

Quirks

• TikTok keyword search is loose. Searching "Nike" can return ads from unrelated advertisers (Interactive Brokers, Shopify in our test). Always post-filter by advertiserName matching the user's intended brand; warn the user if zero matches after filter.
• TikTok Ads Library is EU/EEA/UK only. The library source needs an EU country code (DE / FR / IT / ES / NL / PL / SE etc.). For US/global coverage, switch to creative_center source — different fields (CTR, impression ranges, no targeting data).
• dz_omar/google-ads-scraper requires resultsPerQuery >= 10. Smaller values fail validation. Always set 10+ even for small intents.
• apify/facebook-ads-scraper takes URLs, not keywords. For competitor-ads: build https://www.facebook.com/<PageName> from the brand name. For keyword-ads: build a Meta Ad Library URL with q=<keyword>&country=<XX>.
• apify/google-search-scraper paid-ads mode has a built-in retry (up to 3) when no paid results are found — sometimes a query genuinely has no paid results. Treat empty paidResults as a valid answer, not an error.
• LinkedIn Ad Library URL construction: company URL https://www.linkedin.com/company/<slug>/ is allowed but slow and ignores filters. For competitor-ads use https://www.linkedin.com/ad-library/search?accountOwner=<slug>&countries=<XX>. For keyword-ads use ?keyword=<term>&countries=<XX>.
• X has no public ad library. Coverage is heuristic only. The route uses apidojo/twitter-scraper-lite to scrape a brand's own tweets (or keyword search results), then flags items with non-empty card field or source containing "Ads" as likely promoted. This will miss promoted-only tweets that never appear in the brand's own timeline.
• X session sensitivity. If the primary X Actor returns only noResults sentinels, switch to the fallback before declaring zero results.
• Pricing. Most primaries are FREE in our pricing tier; apify/facebook-ads-scraper charges per ad ($0.001 - $0.0058); X primaries charge per tweet (~$0.0004 / 1k). Default counts (30 / 20 / 50) keep cost negligible. Warn before runs of 500+ ads.

Error Handling

• Auth error → run apify login, or set APIFY_TOKEN env var
• Actor not found → check Actor ID against the routing table
• Run status FAILED → open the console URL (.consoleUrl from run metadata) for logs
• Timeout / very long run → pass --timeout <seconds> to apify actors call, or reduce result count
• 0 results → switch to the Fallback Actor; if still 0, try a different country code
• TikTok library: no EU country supplied → default to DE and warn the user
• dz_omar/google-ads-scraper: validation error on resultsPerQuery → bump to 10+
• X scraper: only noResults sentinels → switch to the fallback X Actor
• proxy is required error → add "proxy": {"useApifyProxy": true} to the input