What it does
GEO Builder scores a site's Generative Engine Optimization compliance — the structural work that decides whether an LLM crawler or AI search engine can parse and cite your pages at all. You supply the scan data, keyed by rule: schema.org structured data, semantic HTML landmarks, llms.txt presence, canonical link tags and heading hierarchy, each as a gap ratio. It scores them against explicit thresholds and returns worst-first findings with the concrete correction for each, plus a ready-to-review fix-PR proposal carrying a title, branch, description and the specific files to change. The ruleset is versioned and lives in its own config file rather than hardcoded in the analyzer, because GEO criteria are still moving. Every report states which ruleset version produced it, so a score from March and a score from September are comparable rather than silently different. The handling of incomplete input is the part worth knowing about, and it is unusually careful. A rule you did not measure is not quietly counted as compliant — it is excluded from the score and raised as a NOT SCANNED warning. A metric key the ruleset does not define is never graded either way, excluded from the score, given no file in the PR proposal, and named in the description as not scored. Inventing a file path for an unknown rule would be a fabricated fix, so it refuses to. That means the compliance score always arrives next to its coverage, and a high score on two measured rules cannot be mistaken for a healthy site. Each step is callable on its own: `compliance_score` for the number, `analyze` for the findings, `propose_fix_pr` for the PR artefact, `run_full` for all three. Scoring and the PR proposal are fully deterministic; the language model only explains why a gap hurts crawler comprehension, with a template fallback when no model is available. It proposes and never pushes. No connectors and no credentials — it runs on the scan data you provide. Built for frontend developers and tech leads, and directly useful to the marketing side that keeps asking for these fixes.
Example prompts
- Score this GEO scan and tell me what to fix first
- What is our overall GEO compliance, and which rules were not measured?
- Build a fix-PR proposal from these findings
