The scan said the cheapest fix wasn't on their website

Anonymization is deliberate and permanent for this story (receipts policy): numbers stay specific, names stay out. The same case is published with full confounders at sandcastlelabs.ai/receipts.

What changed

Fourteen README pull requests and identity metadata standardized across 16 public repositories: one canonical sentence about what the product is, a comparison table, an FAQ. About two person-days of work. No paid placement, no site redesign, no new content program.

Two weeks later, own-domain citation rate read 5.8%, up from 1.25% — and the category rank moved from outside the top 25 to #10.

Why the fix worked

The baseline scan showed a product with real adoption whose most-retrieved surfaces said almost nothing about what it is. For an open-source company, AI engines read the GitHub README as a first-class identity document — often the most current, most structured description they can retrieve. The fix plan targeted exactly that surface and nothing else.

Why we believe the number

The client runs a third-party AI-visibility tracker alongside Collimer. Neither tool was tuned to the other, and both moved in the same week: their tracker’s visibility rank rose 15 places to #10, and average position moved from 8.5 to 4.1. Independent corroboration is what makes this a receipt rather than a self-graded claim.

One honest caveat, stated here the way we state it everywhere: the client’s site team shipped internal linking work in the same window, so we call this strong evidence rather than proof.