I Surveyed 88 Local Businesses. Three In Four Had A Website That Was Not Doing Them Justice.

Between July and August 2026 I surveyed the websites of 88 independent businesses around Newcastle-under-Lyme and Stoke-on-Trent. Not a design opinion, a technical one: does the website load, is it served securely, does it work on a phone, and can a search engine or an AI assistant actually make sense of what is on it.

Sixty-five of the 88, roughly three in four, had a website that was not doing their business justice.

I want to be careful about what that means, because it is not a criticism of anybody. Almost none of these were bad businesses with bad websites. They were good businesses whose websites were quietly underselling them, usually because of something invisible from the front, and in sixteen cases because the web address had stopped working altogether without anyone noticing.

The scores at a glance

Share of surveyed websites scoring full marks, by criterion

Criterion	Full marks
Mobile-friendly	68 of 72 — 94.4%
HTTPS	65 of 72 — 90.3%
Current build	66 of 72 — 91.7%
Content	71 of 72 — 98.6%
On-page search basics	32 of 72 — 44.4%
Upkeep	49 of 72 — 68.1%
AI readability	56 of 72 — 77.8%
Domain reachable	72 of 88 — 81.8%

First seven rows are out of the 72 websites that loaded and were scored on every criterion. Domain reachable is out of all 88 surveyed, since a domain that does not resolve cannot be scored on anything else.

Where the three-in-four figure comes from

Each website was scored out of ten against a fixed rubric. Twenty-three scored full marks. The other sixty-five dropped at least one point somewhere, which is where the 73.9% comes from.

What makes me comfortable putting that number in a headline is that it does not depend on where I drew the line. Three different definitions land on exactly the same 65 businesses:

• Websites that did not score full marks: 65.
• Websites that dropped a point on any criterion, plus those with no working site at all: 65.
• Websites failing at least one of the three things owners never see — search basics, upkeep, or AI readability — plus those with no working site: 65.

However you define "not doing them justice", it is the same three-quarters of the sample.

More than half failed the same basic check

Seventy-two of the 88 websites loaded and were scored on every criterion. Of those, 40 (55.6%) scored zero on the criterion covering the on-page basics: a sensible page title, exactly one h1 heading, a meta description, a canonical tag, and a reachable robots.txt or sitemap.

These are not advanced techniques. They are the fields a website builder leaves blank until somebody fills them in. Looking at what those 40 failures actually cite:

• 39 had a problem with the h1, the main heading element: missing, empty, duplicated, or several on one page.
• 38 had a meta description problem.
• 36 had a page title problem, frequently still set to a template default.
• 28 had no canonical tag.
• 25 had a missing or unreachable robots.txt or sitemap.

None of this is visible to a visitor. The page looks completely normal. The heading is the right size in the right place; it simply is not marked up as the heading, so the software deciding whether to show your business is working with less than it could be. That is exactly why these persist. Nothing about using your own website tells you they are there.

The usual story about small business websites did not hold

I expected to find plenty of ancient websites that do not work on phones. That is the picture the web industry has been selling against for a decade. Among the 72 that loaded:

• 68 (94.4%) took full marks on mobile. Three scored zero, one partial.
• 65 (90.3%) took full marks on HTTPS. Three scored zero, four partial.
• 66 (91.7%) were on a current build. Six were genuinely dated.
• 71 (98.6%) had enough content to say what the business does. One did not.

That is a real change, and the credit belongs to the website builders. Squarespace, Wix, Shopify and modern WordPress themes are responsive and secure by default now. Anyone who built a website on one of them in the last few years got the structural decisions right without having to know they were decisions.

Two axes were the exception, and they are the two nobody checks. 23 (31.9%) failed on upkeep: a copyright year years out of date, a theme demo page still live, a duplicate of the website sitting on a builder subdomain. And 16 (22.2%) failed on being readable by AI assistants, which in practice almost always meant no structured data on the page at all.

Sixteen web addresses did not work at all

Sixteen of the 88, 18.2%, failed the first check: the domain did not serve a working website. Under the scoring rules that zeroes the whole rating, because a website nobody can reach cannot earn points for anything else.

How they failed: 10 domains no longer resolve at all, the registration or DNS having lapsed. 2 resolve to a parking page carrying adverts. 2 are a bare redirect with no website of their own. 1 has a host answering but reporting the site as not published. One other.

This is the number most likely to be quoted and the easiest to overstate, so here it is narrowed down honestly:

• 16 had a domain that failed the check.
• 13 of those had no other web presence I could find. For the other three the business is online elsewhere: two redirect to a marketplace shop, and one has a live, maintained website on their web developer's subdomain while the domain they own has lapsed.
• 8 of those 13 additionally had positive evidence of still trading: an active social account, recent local press, or dated reviews. For four I found no evidence either way, and one appears to have stopped trading, its published phone number confirmed disconnected.

So the figure I would stand behind is the tightest one: 8 of 88, or 9.1%, are businesses with evidence of still trading whose only web address no longer works. The 18.2% is a true statement about domains. The 9.1% is a true statement about businesses. Both are in the data and you can check either.

There is almost no middle ground

I expected a spread: some excellent websites, a long tail of tired ones, most somewhere between. What I found was two clusters and very little in between.

Sixty-two of 88 scored 8 or above. Sixteen scored zero. Only ten landed anywhere in the range between, and exactly one business scored 6. The tired-but-working website that everyone pictures as typical for a small business was the rarest thing in the survey.

Which is the encouraging part of all this. Three in four websites were not doing their business justice, but for most of them the gap between where they are and where they should be is a handful of settings, not a rebuild.

Everything here came from public sources

Worth stating plainly, because a survey of other people's websites invites the question. Every piece of information behind this came from what those businesses publish about themselves, freely available to anyone with a browser.

• One page per business, fetched as any visitor would. The scoring script requests the public homepage over the ordinary web, reads what is returned, and stops. No logins, no paywalls, no member areas, no customer portals.
• No access control was bypassed. Nothing was guessed, forced, or worked around. A handful of websites refused the script's default identity and were re-fetched with a normal browser identity, which returns the same page those websites serve to every visitor anyway.
• Directory, review and social listings were read as published. They are public pages, used only to confirm whether a business is still trading.
• Domain checks are public records. Asking the internet's own name servers whether a domain resolves is the same query your browser makes before loading any page.
• No personal data beyond what a business publishes about itself. Nothing about customers, staff or anyone's private information was collected, and none of it appears in the released dataset in any case.

The published data goes further than that: business names, web addresses, telephone numbers, street addresses and the names of proprietors are all stripped out, settlement names are dropped in favour of a distance band, and the smallest sector groups are suppressed. What is left describes a market, not any identifiable business. No business is named anywhere in this post or in the data, and none will be.

How this was done

Each business was checked two ways: a sweep of directories and aggregators, cross-checked against a direct search for the business name plus its settlement. Nothing was recorded as having no website on a directory check alone.

Every domain that failed to resolve was re-checked against an external resolver with a known-good control domain in the same query, so a genuinely lapsed registration is distinguished from a local network fault. Websites that blocked the scoring script were re-fetched with an ordinary browser identity and their stylesheets read by hand, so being blocked never counted against a site. Where a signal was simply absent rather than broken, the point was awarded rather than withheld.

Scoring is against a fixed ten-point rubric, published in full in the README. Blocking AI training crawlers was not treated as a fault, since that is a legitimate choice a business is entitled to make; the AI criterion scores only whether anything prevents the website being used as a source.

On tooling, because it affects how much weight the numbers carry: the measured signals are computed by a script from the fetched page — HTTPS behaviour, viewport, media query counts, platform and library detection, markup structure, text length. Discovery and the written assessment notes were produced with AI assistance and reviewed. Every percentage here is a count of recorded criterion scores, not a count of statements in those notes. That distinction matters: an earlier pass of this analysis derived the rates from the written summaries instead, and it undercounted the search failures by half, because a summary only mentions what its author chose to write about.

What this cannot tell you

I would rather set these out than have them found.

• One area. Rings out to roughly eight miles from Newcastle-under-Lyme, taking in parts of Stoke-on-Trent and the surrounding Staffordshire and Cheshire villages. Not a national picture, and not necessarily true anywhere else.
• One snapshot, July to August 2026. No trend data, so nothing here supports a claim about conditions improving or worsening.
• Not a random sample. Independent businesses only, and later sweeps deliberately excluded food and hospitality, computer and IT, convenience and grocery, ceramics manufacturers, and professional services.
• Businesses with no web presence anywhere are under-represented, because the method finds businesses through listings. That biases the dead-domain figure downward, not upward.
• n=88. That supports coarse proportions. It does not support ranking trades against one another, which is why the smallest sector cells are suppressed in the published data.
• Homepage-only for the content and search criteria. The script reads one page, so a website could have a correct h1 on inner pages and still score zero here.
• A technical score, not a design review. A website can score 10 here and still be unpleasant to use, and one can score badly while serving its customers perfectly well.
• The trading evidence is not uniform. Nine of the sixteen had it recorded explicitly; for the rest I have said so rather than assuming.

If you want to check any of it, the data and the rubric are both published and the individual criterion scores re-add to the totals. If something in it is wrong I would rather know.

If you would rather find out where your own website stands than work from someone else's average, that is what my SEO Management service reports on each month. The checks behind the search criterion are covered in what AI search engines see on your website, and the consistency checks in why your business details need to match everywhere, if you would rather run them yourself.