What makes a website speed audit worth paying for?
You should know which problems deserve attention and what the proposed repairs would do. A long list of warnings is not a plan by itself.
Use this guide when commissioning an assessment or comparing a report with the experience people describe. It doesn’t require you to become a performance engineer. It does require the report to explain its sample, findings and proposed work.
Begin with the sample, not the score
Ask which pages were selected and why. A site's homepage might contain little of the functionality used elsewhere. A product with no options doesn’t represent a subscription product; an empty search result doesn’t represent a broad query; a short article doesn’t exercise a long media-rich template.
The sample should cover the complaint and important variations without pretending that every URL needs identical testing. If the owner says a menu hesitates after opening, an audit that measures only initial navigation has left the core question unanswered.
For each example, look for a recorded URL, task, starting state and environment. Note whether the visitor is signed in, whether the visit is fresh or repeated, and which device conditions the comparison uses. These details make a finding reproducible rather than merely plausible.
Separate the two kinds of evidence
Google's PageSpeed Insights combines controlled Lighthouse analysis with available real-user information from the Chrome User Experience Report. Its field information covers a previous 28-day period. If a URL lacks sufficient samples, the report may use origin-level information; an origin can also lack enough data. Read the coverage label before treating the result as a description of one page.
Controlled testing helps investigate a cause under recorded conditions. Field information reflects collected visits but offers less detailed diagnostic information. A disagreement can therefore be a useful lead, not proof that one report is defective. Google's PageSpeed Insights documentation explains those distinctions.
An audit should state which evidence supports each conclusion. “The page is slow” is too broad if the evidence is a single run on one template. “The selected product page repeatedly delays its main image under these conditions” is a more useful claim.
Look for a chain from symptom to intervention
Imagine a fictional print studio whose portfolio is slow to reveal its first project photograph. A report identifies a late image request, explains the dependency that delays discovery and proposes changing the template so the browser can find it earlier. It also records that the remaining gallery can retain its detail-loading behavior.
That’s a testable finding. By contrast, “optimize images, install caching, improve hosting” offers several possible actions without explaining which one addresses the observed sequence. Those actions might be useful, but the report has not yet established their priority.
A finding should include the affected experience, observation, interpretation, proposed intervention and check. Where the cause remains uncertain, it should identify the next evidence needed. A clearly bounded hypothesis is more valuable than an unsupported guarantee.
Ask what the change could disturb
Performance changes can alter timing and dependency order. An app that seems unnecessary during page load might supply a later interaction. A script used by a form might be absent from the initial view. A cache rule that helps public pages may be unsuitable for a signed-in response.
The report should identify these relationships before recommending removal or deferral. That doesn’t mean every intervention needs an enormous test plan. It means the checks should follow what the changed component actually does.
Ask who can verify the business behavior. A store operator can identify an incorrect subscription selection that a general page test might miss. An editor can tell whether the revised publishing path remains practical.
Evaluate the priorities
A useful ranking considers importance, evidence, effort and risk. A major wait on a high-value journey deserves a different conversation from a minor warning on a little-used page. An easy correction can be worthwhile, but ease alone doesn’t establish that it solves the problem the business raised.
Check whether provider-controlled work is separated from changes the site team can make. The audit shouldn’t turn an external constraint into an unconditional implementation promise. It can instead provide reproduction steps, timings and questions for the provider.
Agree what happens after the report
Determine whether the engagement includes implementation, a review with the existing developer, a repeat comparison or later field-data review. Assign an owner to unresolved findings. Otherwise the audit can become a document everybody has read but nobody has turned into a decision.
Before accepting it, try answering four questions: What did we learn about the original complaint? Which change should happen first? What must stay correct? How will we know whether the change helped? If the report can’t answer them, ask for clarification before approving a larger scope.
