Score niche tool site ideas from keyword, SERP and market data

Naved, posting on the n8n community forum, built a workflow to find small single-purpose web tools worth building, such as calculators and converters. Given a niche, it expands seed queries through the Twinword keyword API, filters to tool-shaped queries and merges aliases, then scores each on volume, CPC, build simplicity and cluster strength with per-category quotas. For the shortlist it pulls the live top ten results from Serper and penalises strong incumbents, checks Chrome Web Store competition through Apify, and pulls sale comps from a Flippa scraper. Only then does OpenAI, with a strict JSON schema, name the site and write a brief per tool. A validation step drops any tool whose keyword is not in the scored data, and results are written to Notion with dedup keys.

Each run takes about 20 minutes, costs about $5 and produces a ranked blueprint of more than 40 tools with real volumes, CPCs and evidence. Per-node retries with continue-on-error kept single dead sources from killing long runs. The keyword API's rate limit needed a 7 second batch interval; 5 seconds failed intermittently. He nearly shipped a scoring bug by misreading an ad-density field as SEO difficulty. A second workflow with the same shape scans launch sites for traction. He gives no figures on whether any tool was built from the output.

The design keeps every number in deterministic code and uses the model only at the end for naming and briefs, plus a check that rejects briefs citing keywords not in the data. Field names from the keyword API are odd (one has a literal space), so validate them before scoring.