The audit tells you what's wrong. Someone still has to fix it.

The uncomfortable part of any audit arrives on the last page. You now have a prioritised list of everything wrong with your site, ordered by impact, with the fixes spelled out. It is genuinely useful and it is also, at that exact moment, more work than your team has room for. Audits fail at implementation far more often than they fail at analysis.

So it's worth being direct about the arrangement. AI Audit is built by Metamode, a design and development studio working with Australian small businesses. The free audit is a sample. A paid audit is the product. And when clients would rather not work the list themselves, we do it on a monthly plan. About a third of them ask. There's no obligation attached to either, and the list is yours regardless.

Why the audit came first

Most studios open with a discovery phase: weeks of workshops producing a document that describes your business back to you. We inverted it. By the time a paid audit is delivered we have already worked through your site, read your competitors, scored every page, and identified the things standing between your traffic and your enquiries. That is discovery, done with evidence instead of sticky notes, and it's already paid for.

Which means the work starts from a shared, specific picture of the problem. No re-litigating scope in week three. The backlog already exists, already prioritised, already agreed, because you watched the walkthrough and told us which findings you disagreed with.

What we actually do

• Design: interface and web design, and the unglamorous work of collapsing fourteen greys and nine button variants back into something maintainable.
• Build: shipped front-ends and full websites, not handoff files. The audit engine you just used is our own work.
• AI, applied: the tooling that makes a slow audit fast is the same capability we build into clients' sites: what to automate, what must stay human, and where a model will quietly embarrass you in front of a customer.
• Ongoing work: a monthly plan working the list, with a re-score at thirty days so the improvement is measured rather than asserted.

That last one matters more than it sounds. We re-run the engine on your site thirty days after delivery and send the delta. Same rubric, same rules. Either the score moved or it didn't, and both of us can see which.

Design and AI in the same room

The usual arrangement is a design studio that treats AI as a rendering trick and an AI shop that treats the interface as decoration. The results are predictable in both directions: something that looks fine but does nothing, or something capable that nobody can work out how to use.

We built a scoring engine that reads pages, a pipeline that captures them, and a craft lens whose entire job is spotting generic, templated work. Being able to recognise that failure mode in a machine's output is not a small qualification for building with machines.

If you're somewhere in that sequence and want to talk about the last step, book a call. Bring the audit. It's a much better briefing document than an agenda.