- findings in the audit
- 76
- pages with valid JSON-LD afterwards
- 83
- scenario images
- 16.3 → 5.9 MB
- logo that shipped on every page
- 2.85 MB
Where things stood
After April 26 the repository was quiet for seven weeks. Work picked up again on Sunday, June 21, with three pull requests, all on the marketing site, and it started with an audit rather than new pages.
What the audit found
We reviewed the site against four goals: search traffic, turning visitors into enquiries, page speed, and general polish. The result, committed to the repository as AUDIT.md, had 76 findings. The worst ones were plain mistakes, not matters of taste:
- The structured data was wrong on every page. Every page carried the same JSON-LD: a
LocalBusinessnode, which tells search engines you’re a local storefront, plus aSoftwareApplicationlisting for a product we didn’t sell. Aschemaprop that would have let a page add its own markup existed in the layout and was never passed through to the component that renders it, so no page used it. - The main call to action led nowhere. The site’s primary buttons pointed at an
/auditpage that had no form on it. - A 2.85 MB logo shipped on every page. The full-size crown image was used in the footer, so every page view downloaded it, while smaller copies sat unused in the repository.
What we changed
PR #5 worked through most of the list in one change (106 files):
- Structured data rebuilt. One
OrganizationandWebSitegraph for the whole site, and per-page breadcrumb, list, service, article and FAQ markup through the now-working prop. After the change, all 83 generated pages had valid JSON-LD. - Weight removed. The crown swapped for pre-scaled copies; scenario images downscaled from 16.3 MB to 5.9 MB (one screenshot went from 8 MB to 0.58 MB); 5.2 MB of videos nothing referenced deleted; Chart.js bundled by Vite instead of loaded from a CDN.
- Accessibility. A contrast sweep raised low-opacity text across 69 files, and the site gained
prefers-reduced-motionhandling, visible focus outlines, a skip link and keyboard-reachable menus. - Conversion. The buttons pointed at the real lead form, which went down to three required fields. New pages: an FAQ, an RSS feed and an
llms.txt.
PR #6 rewrote the copy across 55 files. It had drifted dense and jargon-heavy, against a brand guide that itself asked for the opposite. A shared glossary did most of the work: “CMEK” became “you hold the only keys that unlock your data”, and per-seat licensing became “a fee for every employee who logs in”. The invented brand terms stayed, each followed by an explanation.
PR #7 added a case studies section (a content collection, index and detail pages, and Article markup) with a first, anonymized study.
What we got wrong
We didn’t take them out until July, which is its own note.
What we’d tell someone else
Audit your own site the way you would a client’s, and start with what’s mechanically wrong: markup, dead ends, weight. Those fixes are cheap and you can verify them. Then treat every number on the page as a claim that needs a source, before you polish the sentence it sits in.