Google Search Ads Builder

Google Search Ads Builder

Composite: keyword research → architecture → structure → copy → negatives → bid → export. Scripts cover the deterministic research primitives (HN terminology, autocomplete expansion) and the export (RSA-limit validation + Google Ads Editor CSV). The architecture, intent/funnel classification, ad-group design, RSA copywriting, negatives, and bid strategy are the agent's reasoning — then validated and serialized by the export script.

When to use

• "Set up Google Search Ads for <product>" / "Build a PPC campaign."
• "Do keyword research for Google Ads" / "What keywords should we bid on?"
• "Generate Google Ads copy" / "I need an import file for Google Ads Editor."

How to run

1 — Keyword research

Seeds + competitor/review/community mining are agent-driven via web search (Robomotion Proxy, geo): site:<competitor>, <competitor> alternative/vs, best <category> tools, site:g2.com/site:capterra.com, site:reddit.com. SERP path: if DATAFORSEO_LOGIN/ DATAFORSEO_PASSWORD (or SERPER_API_KEY) is set → structured SERP API and real keyword volume/difficulty; if not → the agent's keyless web search with directional volume (default). Reuse the google-ad-scraper capability for competitor ad terms, and fetch_landing_page.py from ../ad-to-landing-page-auditor/scripts/ (or web search) for the own-site / competitor content audit.

Deterministic research primitives:

# Technical-buyer framing from Hacker News (keyless)
python3 ${SKILL_DIR}/scripts/hn_terms.py --query "workflow automation" --days 365 --output ${WORKSPACE}/hn.json

# Autocomplete / modifier / long-tail expansion (keyless Google Suggest)
python3 ${SKILL_DIR}/scripts/autocomplete_expand.py --seeds "workflow automation,rpa software" \
  --geo US --alpha --output ${WORKSPACE}/expanded.json

For Reddit/review depth beyond search: if APIFY_API_TOKEN is set → use an Apify actor (better depth); if not → degrade to site: web search over public pages (default).

2 — Architecture, structure, copy, negatives, bid (you, the agent)

From the pooled keywords: classify funnel stage + intent, estimate competitive density (from SERP density + autocomplete + HN/review frequency — paid keyword-volume/difficulty is an optional enrichment: if DATAFORSEO_* is set use its real metrics, else derive directional volume from the keyless signals), score keywords, pull a quick-wins list. Build ad groups (≤15 single-theme keywords each, one LP per group). Write 3 RSAs per ad group: 15 headlines ≤30 chars, 4 descriptions ≤90 chars. Build negatives (universal + category + intent + competitor + cross-ad-group). Recommend bid strategy + budget split by monthly_budget tier.

Emit the campaign as a JSON tree (see build_import_csv.py header for the shape) to ${WORKSPACE}/campaign.json.

3 — Validate limits + emit the import CSV

python3 ${SKILL_DIR}/scripts/build_import_csv.py --input ${WORKSPACE}/campaign.json \
  --csv ${WORKSPACE}/google-ads-import-$(date +%F).csv \
  --report ${WORKSPACE}/validation.json

This enforces RSA char limits and ad-group caps before writing the CSV so the import won't reject rows. If ok:false, fix the flagged headlines/descriptions and rerun — the CSV is not written on validation failure (exit code 1).

4 — Render the strategy doc

Write google-search-campaign-<YYYY-MM-DD>.md to ${WORKSPACE} (overview, research summary, competitive density map, campaign tree, keywords with match type/funnel stage/ intent/priority, RSA copy, negatives, bid + budget split, top-10 quick-wins, launch checklist) and attach both it and the CSV to the Agent Teams channel.

Outputs

• google-search-campaign-<YYYY-MM-DD>.md — the strategy doc.
• google-ads-import-<YYYY-MM-DD>.csv — Google Ads Editor import file.

Credentials / env

• Required: none — keyword research runs on web search + the keyless HN API + Google Suggest; the export is stdlib. Architecture/copy/bid are the agent's reasoning.
• Optional (each with a keyless default fallback):
  • APIFY_API_TOKEN — if set → Apify Reddit/review actor for depth; else → site: web search over public pages (default).
  • DATAFORSEO_LOGIN/DATAFORSEO_PASSWORD (or SERPER_API_KEY) — if set → structured SERP API + real keyword volume/difficulty; else → keyless web search + directional volume from SERP density + Google Suggest + HN/review frequency (default).
  • HTTPS_PROXY — Robomotion Proxy for geo-correct SERP/autocomplete/research.

Notes & edge cases

• Paid search-volume is optional enrichment: with DATAFORSEO_* set, use real keyword volume/difficulty; without it (default), derive volume/competition directionally from SERP density + autocomplete + HN/review frequency.
• RSA limits are enforced by build_import_csv.py (headline ≤30, description ≤90); fix and rerun on failure rather than shipping a rejecting import.
• Cap each ad group at 15 single-theme keywords; cross-negative between groups to stop cannibalization.
• Use a proxy + geo so SERP/review/autocomplete results reflect the target market.