Score sites worth redesigning
If you sell web design, finding businesses is the easy part. What eats my day is working out which 40 of the 300 on a list are worth an email, and what to put in the first line so it does not read like every other agency pitch. This does that pass for you, and it tells you where its own numbers stop being trustworthy.
The questionHow do I find which businesses need a website redesign, and what do I say to them?
How the working version is put together
Load the site the way a visitor does
A real Chromium per site, no clicking and no banner dismissed. That matters for what comes next: everything I measure here is what the page did on its own, before anyone chose anything.
Work out how old it looks
The newest lastmod in the sitemap, the newest date in the page's structured data and time elements, the footer copyright year, the generator tag and the jQuery version. I give the weight to dates the site publishes about its own content. A footer year is usually printed by a script and rolls over every January on its own, so it only counts when there is nothing else.
Watch what fires before consent
Every network request is recorded from the moment the page starts loading and for three seconds after. Whatever fired in that window fired before a visitor had a say. I name the well-known tracking services, count the cookies already set as first-party and third-party, and look for a privacy link and for a consent tool from a list of the usual vendors.
Run axe-core on the rendered page
I inject axe-core into the rendered page and run it against the WCAG 2.1 A and AA rules. You get failing elements rather than failing rules, split by impact, plus the five rules that failed on the most elements. An unmeasurable signal never scores at all, so a site whose accessibility could not be tested does not quietly collect zero points as though it had passed.
Write the three plainest reasons as sentences
Those top reasons come back as one flat sentence each, with the real numbers already in them, ready to paste into an opening line. I keep them unexcited on purpose, because the specific detail earns the reply and adjectives get in its way. Every number in one of them is read back out of the row it describes, so nothing there is generated and nothing can be invented.
What it found on public sites
I scanned these public websites to test the build, and they are nothing more than that. A farm site came back at 68 and WARM: the newest page in its sitemap was from 2020, the footer still read 2014, it was serving jQuery 1.4.4, and a Meta pixel fired before any consent choice was made. A bed retailer scored 52 with no mobile viewport tag at all and six critical accessibility failures. Another site was served over plain http with four trackers firing before consent.
Then a large retailer scored 2, COLD, with a consent platform detected and zero trackers before a choice. That is the modern case behaving exactly as it should, and it is the row I point at when someone asks whether the scoring just punishes everything it looks at.
It scores my own site COLD
I ran it on fractionalhq.uk. It came back at 25, COLD, and that is the correct answer. I would rather show you that than a screenshot of it savaging a stranger, because a scoring tool you cannot see aimed at its author is a scoring tool you have no reason to trust.
It also sets your expectation properly. A site built this year with a consent tool and a viewport tag should score near the floor, so if everything on your list comes back HOT, I would be reading the list as the finding rather than the tool.
What the score is made of
Six lines add to 100: staleness up to 35, accessibility up to 25, consent and tracking up to 15, a missing mobile viewport tag at 8, https problems up to 7, and an old platform up to 10. I band those at 70 and above for HOT, 40 to 69 for WARM, and under 40 for COLD.
A line only fires when it has something to say, and turning a check off in the input removes its lines too, so only compare scores across runs with the same checks switched on. I collect no personal data: a phone is a yes or a no, social links are hostnames only, and no email address is ever read or stored.
A defect I shipped and then fixed
An early version I published scored every single site COLD. The staleness check was reading the Last-Modified header, and that header updates itself, so it can never be a staleness signal. Every site on earth looked freshly touched and the whole rubric collapsed to nothing.
Content dates decide it now: the sitemap, the structured data, the time elements, then the footer year as a last resort. Last-Modified still appears in the row because it is occasionally useful to see, and it is never scored. I am writing this down because a scoring tool that is confidently wrong in one direction is worse than no scoring tool, and mine was, for a while.
When I would tell you not to bother
I read the homepage only unless you raise the page limit, and that is deliberate. For prospecting, the homepage is what the buyer will look at and what you will talk about. If you need a full site audit before quoting, this is the wrong end of the job.
The consent findings describe what loaded before a choice was made and they stop there. Whether any of it breaks a law is a question for the prospect and their lawyer, not for me and not for the row. I also find it makes a much better opening line than an accusation does.
Accessibility numbers here are axe-core on one page against WCAG 2.1 A and AA. Automated testing catches a portion of the real barriers, so I would treat the number as a reason to open a conversation and never sell it as a WCAG audit.
Some sites refuse a bot outright. Those come back as an error row with the status rather than a score, because I think scoring a page you could not open would be worse than useless, and it is how a lead list fills up with confident nonsense.
What lands on your desk
- One scored site per row: 0 to 100, with a HOT, WARM or COLD band
- Every reason that fired, with the numbers behind it
- The top three reasons rewritten as pasteable sentences, numbers included
- Freshness, consent and tracking, accessibility and the technical basics as separate field groups
- An error row with the real status for any site that refused, never an invented score
- If you want the HOT rows flowing into your CRM and an outreach draft per prospect, that is the work I do
Open the proof
Published case studies of systems that are running. Each links to the thing itself.
Questions this page answers
- What does the score actually measure?
- Six things add to 100: how old the site looks, axe-core accessibility failures, what tracking fired before any consent choice, whether there is a mobile viewport tag, https problems, and an old platform such as jQuery 1.x. I band it at 70 and above for HOT, 40 to 69 for WARM, and under 40 for COLD.
- What do I say to the prospect?
- I give you the top three reasons as one plain sentence each with the real numbers already in them, so they paste straight into an opening line. Every number in those sentences is read back out of the row, so nothing is generated.
- What does it score your own site?
- 25, COLD. I ran it on fractionalhq.uk and that is the correct answer for a site built this year with a consent tool and a viewport tag.
- Are the consent findings a legal opinion?
- They are not. They describe what the page loaded on its own before anyone chose anything, and they stop there. Whether any of it breaks a law is a question for the prospect and their lawyer.
- What happens when a site blocks it?
- You get an error row carrying the real status rather than a score. Scoring a page that could not be opened is how a lead list fills up with confident nonsense.
- Does it read more than the homepage?
- Only if you raise the page limit. By default it is the homepage, because that is what the buyer will look at and what you will talk about.
Want this built and handed over working?
One line is enough to start. You get an honest answer on fit, and a number rather than a discovery call.
Start a brief