How to do website analysis that wins clients

An audit is something a client pays for; a prospecting analysis is something a stranger has to find worth reading. Most agencies run website analysis the first way and send the result the second way, which is why a 40-point Lighthouse export turns into an email that says the site is slow and gets nothing back. Website analysis for outbound has one job: find the single, verifiable finding on a prospect's site that a non-technical owner would agree is costing them money, and find it in minutes rather than hours.

That changes what gets checked, how long it takes and what happens to the findings afterwards. A comprehensive audit treats a missing favicon and a 9.2 second mobile booking page as two rows in the same table. A prospecting analysis treats the first as noise and the second as the whole email. The rest of this piece is the working version: the criteria that earn replies, a routine that runs in under ten minutes, the thresholds that decide what leads, and how to turn a finding into a sentence an owner will act on.

What website analysis is for when the site belongs to a stranger

An SEO consultant reads the phrase as crawl errors and thin content, and a performance engineer reads it as Time to First Byte and render-blocking scripts. For a web design agency doing outbound, it means examining a prospect's site for the design, speed, SEO and mobile problems that justify a conversation about fixing them. The output is not a report. It is one to three findings, ranked by how much damage they are likely doing to the business, with the strongest one chosen to open the email.

The categories worth checking barely change from one prospect to the next, and that is what makes the process repeatable. The findings are unique to each site, and that uniqueness is what makes the outreach feel personal. Templates go wrong when they quote the criterion instead of the finding: "your site may have speed issues" is a criterion, while "the booking page takes 8.1 seconds to become usable on a mid-range Android phone" is a finding the owner can check in thirty seconds.

The test every finding has to pass

Ask whether a business owner with no technical background would immediately understand what the issue is costing them. A slow page where the sale is requested passes, and missing alt text on a three-year-old blog post does not. A contact form that fails to submit passes, and an H3 used where an H2 belongs in a sidebar widget does not. Everything that fails the test can still go in a proposal later, but it has no place in a first email.

The six criteria that produce outreach-grade findings

After stripping out everything that belongs in a paid deliverable, six categories consistently produce findings that are specific, measurable with free tools and tied to something the owner already cares about. Each one can be checked in under three minutes. The goal is not six findings per prospect; it is enough breadth that almost any site with real problems yields at least one strong one.

Mobile load time on the page that asks for the sale

Generic speed tests run against the homepage, but the number that matters is the load time of the booking, quote, checkout or enquiry page on a phone. Test that page specifically and the result is usually worse than the owner expects: a physiotherapy practice whose appointment page takes 11.3 seconds to respond is losing patients who gave up before the form appeared. Treat roughly four seconds to interactive on a mid-tier mobile connection as the line; above it, load time is usually the lead finding, because it is both damaging and fixable. A free website analysis of the prospect's URL returns the mobile speed figure alongside SEO and mobile issues, which is a faster first pass than opening developer tools for every site.

Core Web Vitals that fail Google's own thresholds

Pass or fail against Google's published limits carries weight a third-party score never will, because the owner can confirm it in their own Search Console. A Largest Contentful Paint of 7.2 seconds on a services page, against a 2.5 second threshold, is not an agency opinion. Note the specific metric and value: LCP above four seconds, Total Blocking Time above 600 milliseconds (usually a chat widget or tag manager added without review) and Cumulative Layout Shift above 0.25, which is the technical name for the page jumping as someone tries to tap a button.

Mobile usability on the pages that matter

Open Chrome DevTools at 390px and 360px wide and the common failures show up in seconds: a hamburger icon that receives the tap but never opens, hero text sized at 11px in fixed pixels, a phone number that is an image rather than a tel: link, a primary button sitting at 28px when 48px is the usual minimum. These land well because owners have felt them on their own phones without having a name for them. Naming the exact element validates something they already suspected, which is far more persuasive than introducing a problem they have never noticed.

Search listings, conversion paths and visible inconsistency

Search the prospect's main service and city, then read their listing: "Home | JB Plumbing Ltd" next to competitors who name the service and the town is a finding the owner can see on their own phone in ten seconds. Then click through the conversion path as a buyer would and note friction no crawler catches, such as a quote form demanding eleven fields or a pricing link that returns a 404. Finish with design signals that are observable rather than opinion: three different typefaces across three pages, a 2019 copyright line, a hero image that a reverse image search finds on thousands of other sites. A prospect can argue with "looks dated"; nobody argues with a footer that has not changed in five years.

A ten-minute routine for auditing prospect sites fast

The fastest analyses follow the same sequence on every site, so the process becomes muscle memory instead of a fresh research project. Start with a 90-second disqualification pass, then the measurable checks, then the judgement calls. Consistency matters more than depth at this stage, because the depth comes later if the prospect replies.

Disqualify before analysing

Some sites with problems are still bad targets. Staging URLs in the source or a "coming soon" banner on key sections suggest a redesign is already under way; a domain six months old suggests a business not ready to spend; no form, no phone number and no social presence suggest web leads are not a priority. A site that scores above 80 on mobile PageSpeed, has no indexing errors on its main pages, renders cleanly at 390px and looks under three years old has no honest hook either. That gate typically removes a third of a list before a word is written, and the reply rate on what remains rises accordingly. Building the list from sites that already show signs they need a redesign means fewer prospects fall out at this stage.

Measure first, then judge

Run PageSpeed Insights on the conversion page, check robots.txt by appending it to the domain (a leftover Disallow rule from staging blocks an entire site from Google more often than it should), and note title tags on the main pages. Those checks are mechanical and can be gathered for a whole list at once. The time saved goes into what cannot be automated: clicking through the conversion path, judging which finding matters most for this particular business and deciding whether the prospect is worth an email at all.

How to rank findings so the right one leads

A site with real issues usually produces between six and fifteen findings, and the mistake is trying to use them all. Rank by commercial visibility: lead with whatever is most likely to be costing the business customers today and needs no explanation to feel. A checkout that takes eleven seconds on mobile beats a missing meta description on the blog archive, and a navigation menu that will not open on a phone beats an inconsistent heading hierarchy.

Match the finding to the business model

The same number carries different weight depending on what the site is for. A five-second LCP on a local plumber's homepage costs some organic traffic, while the same delay on an e-commerce product page fed by Shopping ads burns paid clicks on every visit. A layout shift of 0.3 on a designer's portfolio is untidy; on a B2B contact form it interrupts the one action the site exists to produce. Across sectors the pattern holds: speed on product and checkout pages leads for e-commerce, search listings lead for solicitors and accountants, and broken mobile paths lead for local trades whose sites were built cheaply.

Use thresholds so the triage can be delegated

Write the thresholds down: time to interactive above five seconds is a strong finding, three to five is weak, under three is not worth mentioning. With rules like that, a junior team member can triage a hundred prospects in the time a senior person would take to analyse twenty properly. Senior judgement then goes where it matters, on the thirty or forty sites that cleared the bar. Logging which criterion produced each reply shows within a quarter which two or three are doing most of the work for each sector, and the analysis time can shift accordingly.

Turning a website analysis into an email that gets a reply

The translation from finding to email is where most of the value gets lost. "Some issues with your site that the team could address" lands in the same mental folder as every other vendor email that week. The frame that works puts the finding first and the agency last: the page, the number or observable detail, what it likely means for the owner, and a question they can answer in one line.

One finding, in the first two sentences

A single precise observation reads as attention; a list of four reads as a pitch and makes people defensive. The finding should sit inside the preview pane, which shows roughly the first ninety characters, and the subject line can carry it outright, such as a 9.4-second load on the booking page, rather than a vague tease about something noticed. Keep the remaining findings back. They become the agenda for the first call and later the substance of the proposal, which is how the analysis keeps earning after the first email. For more on which problems tend to earn the strongest responses, see this breakdown of website issues that drive replies.

If an email cannot cite at least one number or one observable detail, it does not go out. That rule keeps the analysis honest as volume rises, because no email can be written without a real finding behind it. It also makes follow-ups easy: a note two months later saying the mobile homepage still takes 8.7 seconds shows continued attention without being pushy, and the owner can check that it is still true. The wider campaign mechanics, from list building to sequencing, are covered in the cold prospecting guide for agency teams.

Running the analysis at scale without losing the detail

Four to eight minutes per site is fine for twenty prospects and impossible for two hundred, which is where most agencies give up and go back to templates. The answer is to split the work. Data gathering (speed, Core Web Vitals, title tags, mobile errors) is mechanical and can run across a list in minutes, while choosing which finding leads stays a human call on the prospects that matter most.

Judge any tool by whether its output is quotable. "Your site is slow" is useless, while "the appointment page scores 23 on mobile and the main delay is a 4.8MB uncompressed hero image" is a finding. Many popular senders were built for volume rather than evidence, which is the gap this Instantly alternative comparison looks at, and a separate roundup covers tools for analysing prospect websites by how specific their output is. Swokei is built around this split: it analyses each prospect site for design, SEO, speed and mobile issues and turns the real findings into personalised emails sent from the agency's own mailboxes, with plans on the pricing page.

The quickest way to test the method is on the next prospect on the list. Run the URL through the free website analysis tool, which needs no signup and no email address, and see whether the mobile speed, SEO and usability findings give a number worth putting in the first sentence. If they do, write the email around that one finding and hold the rest for the call. If they do not, that prospect has just been disqualified in under a minute, which is the other half of what good analysis is for.