How Atlas rebuilt its search foundations.
The studio’s own website is a practical test of its research discipline. We found inconsistent identity signals and missing navigation, repaired them, and connected measurement. The implementation is verified. Its effect on search visibility is still being measured.
Atlas Ad Agency is a research-led growth studio in Miami with a new cinematic website: a single-page film built on WebGL, launched the same morning. The fifteen supporting pages, the client portal and the contact function were unchanged and had to stay that way.
The owner reported a brand search showing the contact page, a generic favicon and the bare domain. That observation prompted an audit; it did not establish why Google chose that result. In the August 15 to September 11 Search Console window, the homepage recorded 176 impressions and 13 clicks. Contact recorded 116 impressions and four clicks. The exact query “atlas ad agency” was absent from the returned query rows. Missing query data is not proof that nobody searched.
- The file served at
/favicon.icowas the web framework’s template icon, declared first in the home’s head. The other pages declared a different SVG with a false size and no raster fallback. The declarations were inconsistent; Google’s choice of favicon had not been independently attributed to that defect. - The home carried no
WebSitestructured-data node, which is the primary site-name signal, and three brand names competed across surfaces. - The served home linked to none of the fifteen subpages. The home had inbound links, but visitors and crawlers could not follow it into the dedicated pages. This was a navigation gap, not proof of the contact page’s ranking cause.
- The metadata on the pages most likely to rank for the brand was boilerplate. URL Inspection reported all five service pages as discovered but not indexed, with no last-crawl time returned.
Repair the verified defects and preserve the approved film. The name policy was the hardest call: the domain, the LLC and Instagram all say Atlas Ad Agency, and shorter forms collide with several other companies. So one primary name, with the alternates declared in schema, set in one constant per codebase. Colour, motion and the watch stayed exactly as designed; performance had to come from engineering, not from removing the experience.
- One icon set from the official mark, byte-identical on every URL: a 16/32/48 ICO declared first, an SVG, 192 and 512 PNGs, a touch icon.
- One entity graph on all sixteen pages: Organization with a raster logo, a verified profile, the brand description, a WebSite node, typed page nodes and breadcrumbs.
- A site directory and a service-page row on the home, so every page is one click from the hub and back.
- Titles and descriptions rewritten to display length on eight pages; trailing-slash redirects; noindex headers on the application shells; honest sitemap dates; a 404 with its own title.
- A build gate that fails any release where a title, description, canonical, heading, icon or schema regresses, and a live check that proves the deployed domain after every release.
- For performance: the film refuses to run on software renderers and serves the still edition; a governed frame loop runs at 30 frames per second on modest devices; the film mounts only after the page has painted and gone idle.
- Search Console connected through a service account, with a weekly automated report; a stale sitemap removed; the canonical one resubmitted.
| Measure | Before | After |
|---|---|---|
| Performance (Lighthouse, mobile) | 49 | 66 |
| Largest contentful paint | 3.9 s | 2.5 s |
| Total blocking time | 7,740 ms | 1,750 ms |
| Speed index | 9.3 s | 5.2 s |
| Dedicated public pages linked from home | 0 of 15 | 15 of 15 |
| Favicon at the root | Framework template icon | The Atlas mark, 16 to 512 px |
| WebSite schema on the home | None | Present on all 16 pages |
| Brand names in use | Three | One, with declared alternates |
The performance figures are recorded mobile Lighthouse runs from September 13. They are lab measurements, not field Core Web Vitals or evidence of ranking improvement. The link comparison uses the same fifteen-page cohort; new Work and Capabilities pages were added later.
The September 13 checks confirmed the deployed metadata, icons, canonical URLs and navigation. URL Inspection reported the homepage, About and Contact as indexed; the five service pages remained discovered but not indexed. The available query window ends before the rebuild, so it cannot show its impact. We will compare later query windows and dated search-result observations. Ranking, favicon display and sitelinks are outcomes to observe, not promises of this work.
- Measure before theorising. Four earlier theories about a black screen were wrong; the renderer’s own statistics found the cause in one look.
- A brand result is an entity problem before it is a ranking problem. One name, one logo, one graph, one hub.
- A homepage needs clear routes into the pages that answer a visitor’s next question.
- Creative excellence and performance coexist when the engineering is honest: gate on hardware, govern the frame rate, hydrate the words first.
- Put the standard in the build. Our own gate rejected two of our own changes the same day, and it was right both times.
Search & Discoverability, applied to ourselves.
Read the field note: what discovered but not indexed actually tells us.