Open-source tools for AI-visibility auditing
Curated, with an opinion attached rather than a bare link — including which of these you can skip.
A short, opinionated list. The criterion for inclusion is that it earns its place in a workflow, not that it exists.
Start with the curated lists, then leave them
The awesome-list corner of this topic is already occupied and the incumbents are decent: amplifying-ai/awesome-generative-engine-optimization is the broadest, DavidHuji/Awesome-GEO is the one to read if you want the research rather than the vendors, and discoveredlabs/awesome-aeo-seo-tools is the most engineer-facing.
Read one, take three links, close the tab. These lists grow faster than the field consolidates, and past a point they measure enthusiasm rather than tooling.
The one check worth automating first
Fetch your pages the way a non-rendering crawler does and diff that against what a browser shows.
curl and a headless browser, output through the same text extractor, diff the two. The gap is
your invisible content.
This takes an afternoon, needs no dependency, and finds more than any dashboard will — because most AI crawlers do not execute JavaScript, and the gap is usually larger than teams expect. Run it in CI, not on demand: a rendering regression that ships on a Tuesday is invisible until someone happens to look.
Schema validation, on every build
Validate structured data as a build step rather than a periodic audit. Schema that silently breaks is worse than no schema, because you stop checking something you believe is working. The failure mode is not an error page — it is a page that looks fine and is no longer machine-legible.
What to skip
Anything promising an “AI visibility score” without publishing its prompt set. A score derived from prompts you cannot see is not a measurement; it is a number. Ask which prompts, how many runs each, and which model snapshots — the answers separate the tools from the dashboards.
llms.txt generators. The file takes ten minutes to write by hand or one build step to generate, and the evidence that answer engines read it is thin enough to be worth reading before you invest. Generate it yourself, from your own content, and spend the saved time on the rendering diff above.