Freelance, agency or AI tool: who should audit your UX?

There are three ways to get a UX audit: an automated tool, an independent practitioner, or an agency. They are not three price points for the same service. Tools check rules, independents apply judgement to one surface, and agencies bring process and research to many — and the right choice depends on which of those you actually need.

The reason this gets confusing is that all three use the same two words. A $0 AI report and a $12,000 agency engagement both arrive labelled “UX audit”, and neither is misusing the term.

What does each one actually give you?

Compared on	AI tool	Independent	Agency
Typical cost	$0–199/mo	$150–800 per surface	$3,000–20,000+
Turnaround	Minutes	1–5 days	2–8 weeks
Best at	Rule coverage, consistency, volume	Judgement on one high-stakes surface	Breadth, research, accountability
Weakest at	Knowing what the page is for	Scale	Cost per surface
Who examines it	A model against a ruleset	The person who writes the report	Varies — worth asking

What can an AI UX audit tool actually find?

More than most people assume. Automated tools reliably catch contrast failures, missing labels, tap targets under the minimum, heading structures that skip levels, layout shift, slow resources, missing alt text, and a long list of known dark patterns. On coverage and consistency they beat a human comfortably.

They are also improving quickly and honestly marketed. One of the current tools advertises 1,450+ research-backed rules across e-commerce, SaaS, landing and corporate templates, runs a browser against your live site, and returns a report in about five minutes. For $0 on a free tier that is a good trade, and anyone telling you not to run one is protecting their own invoice.

Where a rule engine genuinely wins:

• Exhaustiveness. It will not get bored on page forty and start skimming.
• Consistency. The same rule applied the same way every time.
• Speed and price. Minutes and often nothing.
• Regression checking. Re-run it after every deploy, which no human does.

If you have never audited your product, running a free tool first is the correct first move. Fix everything it finds. Then read the next section.

What can’t an automated audit find?

It cannot tell you what the page is for, so it cannot tell you when the page is failing at it. Every rule engine evaluates the interface against a standard. The expensive problems are almost never standards violations — they are a correctly-built page doing the wrong job.

Some concrete examples of findings no ruleset produces:

• “Four plans, none marked as recommended.” Every column is accessible, legible and correctly marked up. The page still hands the sorting work to a visitor who came here to be told which one is for them.
• “The headline names a category, not a benefit.” Perfect contrast, correct heading level, passes every check. It just does not say anything.
• “This form asks for company size before the visitor has seen the product work.” The field is labelled, keyboard-accessible and validated. It is also the reason people close the tab.
• “The reassurance is in the footer and the doubt is at the button.” Both elements are fine. Their relationship is the problem.

The common thread is that each one requires knowing what the visitor came to do, what the business needs them to do, and where those two diverge. That is a judgement about intent, and intent is not in the markup.

This is not a permanent property of machines — it is where the tools are in 2026. What is unlikely to change soon is that someone has to decide what the page is for, and that decision is upstream of any audit.

When is an agency worth the money?

When the product is large, when real user testing is in scope, when several teams need to be brought along, or when the organisation needs a name on the document that procurement recognises. Those are real requirements and they cost real money. A single landing page is not one of them.

Most of the gap between $400 and $6,000 for a read of the same page is overhead: an account manager, a scoping call, a kickoff, a designed deck, a presentation, a follow-up, and often a junior doing the examination that a senior reviews. None of that is dishonest — it is what makes engagements manageable at scale — but it is not analysis, and you should know that is what you are buying.

Ask one question that separates agencies quickly: who does the examination? “A senior reviews it” and “a senior does it” are different services at the same price. More on what moves the price.

When is an independent the right call?

When you have one high-stakes surface, you want judgement rather than coverage, and you would rather have a document next week than a process next quarter. A pricing page, a signup flow or a checkout carries a disproportionate share of the revenue, and one person can read it properly in an afternoon.

The trade you are making is scale for directness. There is no account layer, no scoping call and no deck — and equally, no team to absorb it if that person is unavailable, and no research capacity behind them.

What to check before paying anyone independent:

1. A finished report, redacted if necessary. Quality varies by an order of magnitude in this category and the deliverable is the only honest signal.
2. Public reasoning. Anyone doing this seriously has argued in public somewhere. A portfolio of screenshots shows they can design; a written audit shows whether they can diagnose.
3. A written method. A published one means the audit is repeatable and arguable. An unwritten one means you get whatever they noticed that day.

How do you choose?

Start with what you already know. If you have never checked, run a free tool — it costs nothing and clears the rule violations. If you have fixed everything a tool found and conversion is still wrong, the remaining problem is one of judgement, and that is what a human is for. If the product is large or the organisation needs process around it, that is an agency.

Written as a sequence rather than a menu:

1. Run a free automated audit. Fix what it finds. This is genuinely useful and costs nothing.
2. Still not converting? The problem is intent, not compliance. Get one high-stakes surface read by a person.
3. Many surfaces, or need research and buy-in? That is the agency tier, and it is worth what it costs when those are the actual requirements.

Most teams stop needing step three entirely, and a surprising number never get past step one because a tool finds enough to keep them busy for a month. That is a good outcome and it does not require hiring anybody.

---

The checklist I work through is published in full if you would rather do step two yourself, and what a UX audit is covers the ground beneath all three options.