Usability audit: what it finds and what it's worth

A usability audit checks whether visitors can complete the intended task, with findings tied to a page, device, evidence and proposed repair.

A usability audit asks whether a visitor can find, understand, and finish the job they came for: send a request, book a service, buy. It is not “is it pretty,” and it is not a meta-tag SEO pass.

If ads land on a page where the button has slid off a phone screen, the form hangs on a phone field, and delivery terms are in the footer, the budget is already burning. The audit names those barriers as tasks: what is broken, on which screen, what to change.

Buying the PDF does not raise conversion. A list of faults is a list. Evaluate any business effect after implementation with an appropriate observation window; a report alone does not establish conversion improvement.

ISO without the seminar

ISO 9241-11 defines usability as effectiveness, efficiency, and satisfaction for specified people, goals, and context.

Effectiveness — did they finish (form sent, paid).
Efficiency — how many extra steps and minutes.
Satisfaction — did the process make them close the tab.

ISO 9241-110:2020 (self-descriptiveness, conformity with expectation, error tolerance) is common sense with a standard number, not a hundred-point checklist.

Do not buy the neighbouring product by accident:

SEO audit — indexation, duplicates, titles. Robots need clean code; people need a path.
Technical audit — server time, script errors, load.
Analytics setup — events and goals. A counter records; it does not interpret.
CRO / A/B — an ongoing process on a real traffic stream.

A usability audit is a snapshot: where a person stumbles now.

Nielsen’s ten are not a hundred-item sheet

The working base for an expert pass remains Jakob Nielsen’s ten heuristics (1990–1994; NN/g restated them in 2020 for modern UI). They are logic, not a scorecard:

Visibility of status — a loader, “order received.”
Match to the real world — the customer’s words, not warehouse jargon.
User control — undo, dismiss the modal.
Consistency — the same control behaves the same.
Error prevention — a phone mask beats a scolding after submit.
Recognition over recall — keep choices in view.
Flexibility — reorder, saved details, for people who already know the path.
Minimal distraction — kill the banners that compete with the job.
Plain errors — what broke and how to fix it.
Help in place — a hint at the field, not a PDF in the footer.

The expert walks the buyer’s path against those rules.

The myth of five users and 85 percent

“Five users find 85 percent of problems” is a compression of Nielsen, 2000, itself from a 1993 model that put the chance of one person hitting a given fault at about 31 percent.

The rest of the sentence matters:

In Nielsen’s model, five people find about 85 percent in one iteration; this is a model result, not a guarantee.
The model suggests about fifteen for nearly complete discovery in a single pass; real studies depend on the audience and task.
Nielsen’s actual advice: three rounds of five (find, patch, retest), not one heavy fifteen.
Separate audiences (wholesale and retail) are separate tests.

Laura Faulkner’s 2003 work (see the HFI summary): five users averaged 85 percent, but some groups hit 55 percent. Ten users averaged 95 percent, with a floor of 82 percent. Perfetti and Landesman (2001) saw five users on a music site find 35 percent.

Five is not a guarantee. An expert walkthrough, behaviour-data review and participant study answer different questions. Name which methods the engagement includes and which user behaviour remains untested.

Search: convenience is not a ranking lever you can buy

No search engine will promote a page because a button was repainted. The link from comfort to rank is indirect.

Google’s page experience docs are explicit: there is no single page-experience signal. Relevance stays primary. What Google does use are Core Web Vitals: LCP, CLS, and INP (INP replaced FID in the set on 12 March 2024; Chrome dropped FID from PSI and CrUX later that year). Good CWV do not buy first place. They remove technical stall. An awkward page can still win if it answers the query better.

A “fix the button, climb the SERP” formula does not exist. Do not treat an audit as a guarantee of a ranking position.

What session replay sees, and what it will not say

GA4 and Plausible record events you asked them to record. Session replay (whatever vendor you have) shows scroll, field fill, the mobile keyboard, idle stretches.

Limits to keep honest:

Sampling. Many tools do not write every visit. Short bounces often never become a video.
Blockers. Ad blockers and corporate networks drop trackers. Your “heatmap” is a biased subset.
Bots. Some sessions are not people.
Motive. A cluster of clicks on a non-clickable heading is a fact. It does not explain why they left.

Replay shows what happened on the glass. It does not show why. Watching hundreds of videos without a walkthrough protocol is a way to spend the week.

If there is no analytics, the brief says so. Inventing a replay you do not have is not an audit.

The market, without a tariff card

The market splits by what you actually receive:

Checklist review — confirm whether it includes a walkthrough, evidence and implementation; the supplier category alone does not establish scope.
A working studio pass — key scenarios, analytics if you grant it, a task list a developer can execute.
A broader journey review — clarify the covered steps, available behaviour data and whether implementation is included.
A UX lab — live participants. That is not a PDF “by eye,” and it is not the same product.

A claimed conversion uplift needs a baseline, sample, observation window and implementation details. Do not infer the effect from the audit format alone.

A useless audit is forty slides: “dated design,” “not enough air,” “dull buttons.” A developer cannot schedule that.

A useful audit can include a cognitive walkthrough. Each fault is a ticket:

Finding — page, block, device.
Evidence — annotated screenshot or a session id.
Patch — phone mask, remove an unnecessary required field, put the request button in the first screen.
Priority — blocks the job / adds friction / cosmetic.

The developer can use those fields to estimate and implement a specific repair.

What an owner actually chooses

New site, little traffic. Skip empty charts. An expert pass on Nielsen: terms readable, cart or request usable, forms on a phone.

Traffic exists, requests do not. Open the funnel in GA4 or Plausible. See the step that dies. Check whether the leak is the interface or the delivery chain behind the form.

Match the report to the decision and implementation work. Contrast plugins find contrast. Why a person did not send a request is a different object.

Budget the patches. An audit with no implementation slot is a file.

This studio’s formats match the SEO-and-growth service: slice (one landing or a small site, expert pass, no participant study), funnel (landing to request, behaviour data required), fixes (funnel plus the first-wave repairs — forms, offer, buttons, mobile hits, obvious stalls; not a redesign). Agree the scope and available evidence in the written brief; no conversion result is guaranteed.

The point of the audit is not the thickness of the report. It is fewer obstacles between the offer and the decision.