Method

Everything on this site comes from one script in the open. This page describes what it does, so you can check the work, disagree with it, or run it yourself somewhere else.

Which businesses are in scope

We query the Google Places API for 10 categories across 5 cities (Port Arthur, Nederland, Groves, Port Neches and Beaumont), and keep every independent business it returns. A business appears exactly once, filed under the first city that returned it.

National chains are excluded by name. A McDonald's website tells you nothing about how Southeast Texas builds for the web, and a few dozen corporate sites with big budgets would drag every average in this study sideways.

What we measure

• Whether a website exists. Taken from the business's own Google listing. A business with a Facebook page and no site counts as having no website. That is the finding, not an oversight.
• Whether it responds. We request the URL once and record the HTTP status and the final URL after redirects. Two separate facts come out of this: whether the server answered at all, and whether it served the page. A server that answers 403 is running perfectly well and simply refused our crawler, which many sites behind a firewall or CDN do. Counting that as a broken website would report something about our own robot as though it were a fact about somebody's business, so we do not.
• Whether it is secure. HTTPS on the final URL, plus whether an http:// address upgrades on its own.
• Whether it is built for a phone. Presence of a viewport meta tag, a floor rather than a full mobile audit.
• How fast it loads. Google PageSpeed Insights on the mobile strategy: performance, SEO and accessibility scores, largest contentful paint, first contentful paint, total blocking time and cumulative layout shift.

What we deliberately do not publish

No star ratings, no review text, no photos, no “worst sites in town” list, no rankings of one business against another by name. Those either belong to Google or exist only to embarrass somebody. The dataset is a measurement of infrastructure, and it is presented in alphabetical order for that reason.

On data handling: Google's Places terms allow keeping a place identifier indefinitely but not archiving their content, so the durable record here is the identifier plus our own measurements: the load times, status codes and HTTPS results we produced by fetching public websites. Business names and addresses are re-derived from Places each time an edition is built rather than kept as a permanent archive.

Where these numbers can be wrong

They can be, and it is worth being specific about how:

• Places is not a census. It is the best available list of local businesses, not a complete one. A business that has never been listed on Google is invisible to this study.
• One measurement, one moment. Each site was loaded once, onJuly 26, 2026. A site that was having a bad day looks worse here than it deserves.
• A single run is not precise. Re-measuring the same site minutes later routinely moves its load time by a second or more, because the test simulates a network and a device rather than observing real ones. Treat an individual figure as an order of magnitude, not a stopwatch reading. The aggregate numbers are far steadier than any single row.
• Stale listings. A business that moved to a new domain without updating Google appears here as broken. That is a real problem for their customers, but it is not the same problem as a slow site.
• A viewport tag is a low bar. Passing it does not mean a site is good on a phone. Failing it does mean it is bad.
• The sites we could not measure are probably the slow ones. A handful of sites answered our request but produced no PageSpeed result, usually because the test itself failed to finish loading them. That failure mode correlates with being extremely slow, so those sites are missing from the speed figures rather than dragging them down. Every average and median on this site is therefore likely to be a little optimistic, not harsh.
• Lab data, not field data. PageSpeed's mobile scores simulate a mid-tier phone on a throttled connection. Real users on good hardware will have a better time than these numbers suggest.

Corrections

If something about your business is wrong here, write to info@9thavenue.dev and it gets fixed in the next edition, or removed entirely, no questions asked, via the opt-out page.

Reuse

The dataset is free to download, quote, chart and republish, including commercially. A link back is appreciated and not required.