A Lighthouse score is not a decision about your site

The audit lands as a 60-page PDF. Performance 47. SEO 82. Accessibility 91. There are 340 issues, colour-coded, sorted by severity, and a page near the front with a speedometer graphic on it.

Nobody in your business will ever finish that list. Nobody in your business can even tell you which four of the 340 would make the phone ring. The report is accurate, thorough, and completely undecidable — which is why, eleven months later, it is still in the same folder with nothing ticked off.

TL;DR: Automated site audits produce scores and long issue lists, both of which measure a site without deciding anything about it. What an owner needs from an audit is a short, ordered queue of changes tied to a business outcome, with the false positives already removed by someone who looked.

The SharpOS version, and how it actually works

Audits sits inside SharpOS, the private workspace included with every SharpHaw subscription, and its mechanics are worth describing precisely because they are unusual.

An audit target represents a website belonging to an organisation. Trusted agents crawl and analyse that site from the outside, then push what they found back into SharpOS through the integration API: the pages, the findings, artifacts, screenshots, AI-written insight, and the copy for a shareable report. Members browse targets and read the latest retained run. Platform admins can edit the findings and share a public prospect-facing version of the report.

The important word in that paragraph is edit. An audit whose findings can be edited by a human before anyone acts on them is making a specific admission about how automated auditing works, and it is an admission the category mostly avoids making.

Every automated audit is wrong about something

This is the part that gets left out of audit marketing, and leaving it out is how audits become shelf-ware.

Crawlers generate false positives. They flag a missing alt attribute on a decorative image that is correctly marked decorative. They report a render-blocking resource that is render-blocking on purpose. They ding a page for thin content when the page is a form. They surface 200 “issues” from a pagination pattern that is working exactly as designed. Anyone who has read a few hundred audit findings knows the rough hit rate, and it is not near 100%.

None of this means the tools are bad. Semrush is a serious platform, Lighthouse is a genuinely excellent piece of engineering, and both are more rigorous than any manual review. It means the raw output is an input to judgement, not a substitute for it, and that the step everybody skips is the triage.

So the honest shape of a useful audit is: machine finds candidates, human discards the wrong ones, human orders what is left by business impact, and the survivors become work. Skip the middle two steps and you have handed a business owner 340 undifferentiated tasks and called it an audit.

Scores are for comparison. Queues are for deciding.

A score is a measurement of a site against a rubric. It is genuinely useful for one thing: comparing the same site against itself over time, or against a competitor on the same rubric.

It is useless for deciding what to do on Monday, because it collapses every dimension into one number and discards the only information that matters — which specific change, on which specific page, produces which specific outcome.

Watch the difference on a real page. An audit says: Performance 47. Largest Contentful Paint 4.1s. A decision says: the hero image on the services page is a 2.3MB PNG, it is the LCP element, converting and resizing it takes twenty minutes, and the services page is where 40% of your enquiries start. The first is a fact about the site. The second is a job, with an owner, a size and a reason.

The gap between those two sentences is where most audit value evaporates. And it is not a tooling gap. It is that the tool has no idea which of your pages makes money, and nobody told it, because nobody can.

What “the findings become work” has to mean

An audit that is a report ends when it is read. An audit that is connected to the work ends when the last item ships.

Because audits sit in the same workspace as the boards in SharpOS, the natural next step after triage is that surviving findings become cards — with an owner, a status and a visible position in a queue the client can look at. Audit findings can also trigger automations, so a finding of a particular kind can raise work without anyone reading the report first.

The reason to care about that plumbing is not elegance. It is that it changes what happens in month two. A PDF audit’s month-two outcome is that the PDF is still in the folder. A queued audit’s month-two outcome is that eleven items are closed and the rest have a reason they are not.

And the shareable public report has a specific job that is worth naming honestly: it is a sales artefact. It is how a prospect sees a real read of their own site before any money changes hands. That is a legitimate use, and it is more useful to say so than to dress it as pure diagnostics.

A short test for any audit you are handed

Whoever produced it, run these four questions over it. They take five minutes and they will tell you whether you were given analysis or output.

• How many findings are there, and how many are you expected to do? If those numbers are the same, nobody triaged.
• Which three items are first, and why those three? “Severity: high” is not a reason. “It is on the page where your enquiries start” is.
• Which findings did they discard as false positives? A good auditor can name several. An auditor who says the tool found no false positives has not read their own report.
• What is the expected effect of item one, and how would you know? If nobody can name the number that should move, the item is housekeeping. Housekeeping is fine — it should just not be at the top of the list.

Any audit that survives those four questions is worth the money, regardless of who ran it or which crawler they used.

What a good audit ends with

An audit is not a document, a score or a list. It is a decision about what to do next, made by someone who looked at the output and knows which of your pages matters.

Want a read of your own site that ends in an ordered queue rather than a speedometer? Book a 30-minute call. Bring the last audit anyone did for you, and we will go through how many of its findings were real, how many were noise, and which three should have been at the top. More on the workspace those findings live in is on the SharpOS page, and the site work itself is on Conversion-First Websites.