The Capital Region website audit: what we found on local business sites

On the Capital Region business sites we audited in September 2026, the most common failures were the ones a customer meets first: 45 of 115 sites printed a phone number that was not a tap-to-call link, and 69 of 115 showed a broken preview when the link was texted or shared. Speed was worse than either, but speed is a lab number and those two are a thumb on a screen. Here is everything we counted, with the caveat first.

The caveat: the lab test runs on a simulated throttled phone, not real visitor data, and the sample is Capital Region home-service businesses from chamber directories and license rolls, not a random sample. Measured September 2026. We name no business, and every one of them can have its own report free by asking.

The failures a customer feels

The phone number that does not dial: 45 of 115. The number was on the page, but it was plain text, not a tap-to-call link. On an iPhone this matters less than you would think, because "by default, Safari on iOS detects any string formatted like a phone number and makes it a link that calls the number," unless the page turns that off (Source: Apple Safari HTML Reference). On Android, in email apps, and in any browser that does not guess, a plain number is a number you copy out yourself. So we say "not a tap-to-call link," never "cannot be tapped," because the second claim is false on half the phones.

The link preview that comes up broken: 69 of 115. When someone texts your site to a friend, the gray box with a title, a line of text and a picture is built from tags on your page; the Open Graph protocol names the required ones, og:title, og:type, og:image and og:url (Source: ogp.me). Sixty-nine sites were missing at least one of the three parts we check, and 60 of 115 had no preview image at all. A referral is a text message. A text message with a gray box and a raw address is a referral with no name on it.

These two are the ones to fix first because they are the ones a customer runs into on the way to calling you.

The pages that were missing

No privacy page: 85 of 115. Most sites in the sample had no privacy page at all. A contact form collects a name and a phone or email; the laws and terms that reach a site with a form are in our privacy policy post. This one is a page, not a project.

Lighthouse SEO score below 100: 82 of 115. Lighthouse's SEO audits check whether a page has a meta description, whether links have descriptive text, whether the canonical address and robots.txt are valid, whether tap targets are sized for a thumb, and whether structured data is valid (Source: Chrome for Developers). A score under 100 means at least one of those is missing. It is not a ranking and it does not predict one; it is a list of things Google's own tool asks for.

Then speed

Everything in this section is Google's lab test: Lighthouse, mobile, on a simulated slow connection of 1.6 Mbps down with 150 ms of latency and a 4x CPU slowdown (Source: Lighthouse throttling docs). It is a stress test. It is not what your customers see on a strong signal, and the unthrottled numbers below show the gap.

Main-image time, Lighthouse mobile: 102 sites measured, median 7.75 seconds. 98 of the 102 were over Google's 2.5-second line, and 37 of the 102 were over 10 seconds. The metric is Largest Contentful Paint, "the render time of the largest image, text block, or video visible in the viewport," and Google's line for good is 2.5 seconds or less (Source: Google web.dev). On many of these sites the largest thing was a headline, not a photo.

Page weight: 102 sites, median 3.3 MB, 55 over 3 MB. Weight is what the page downloaded. It is a measurement, not the cause of the wait; a heavy page with its image sized right can appear before a light page with a blocking script.

Unthrottled load on the studio computer: 104 sites, median 1.3 seconds, 11 over 3 seconds. One browser load of each site on our own connection, on a computer, with no slowing applied. Not a phone. This is the number that explains why a site can read nine seconds in the lab and feel fine when the owner opens it in the truck. Both numbers are true; they measure different things.

We do not have a number for what visitors do when a page is slow, and we are not going to borrow one.

What this audit is not

It is not a random sample, so none of these counts is a rate for the region. It is not real visitor data. It names nobody, and a lab number alone is never the headline of a report; a site that passes is not a prospect, and a decent site is not called broken. The speed figures are counted by one script (speed-stats.mjs) from the measurement stored for each site and appear the same way on our Albany page; the tap-to-call, preview, privacy and SEO counts come from a second script run on the same September gate reports.

Check your own site in five minutes

1. Open your site on your phone and tap the phone number. Did the dialer open, or did you have to copy it?
2. Text your own address to yourself. Is there a title, a line of text and a picture, or a gray box?
3. Scroll to the footer. Is there a Privacy link?
4. Open pagespeed.web.dev, enter your address, Mobile tab. Read Largest Contentful Paint against 2.5 seconds, and remember what the test is.

If you want the full report

The free audit runs the same checks we ran on these 115 sites, on your site, and sends you the report with every finding in plain words and the measurement beside it. It takes a minute to request and it does not commit you to anything. It is at /audit.