What a free CRO audit should actually check (and what most of them skip)
Search for a free CRO audit and you will mostly find tools that screenshot your homepage, score it out of 100, and email you a PDF about button contrast and social proof. None of that is wrong. It is the wrong layer to start on.
A conversion rate is a ratio, and in the great majority of the sites we audit the problem sits in the plumbing behind that ratio rather than in the design in front of it. Below is the order a serious audit works in, and why the page itself comes fourth.
The fastest way through this page is not to read it. A free scan runs all four layers against your live site, prices each finding in dollars, and ships the fix as a pull request you review and merge.
Layer 1: is the measurement real?
Before anything else, an audit has to establish whether the numbers you are optimizing against exist at all.
Across 936 sites we audited recently, the measurement layer failed this often:
| What's broken | How often |
|---|---|
| No working analytics on the site at all | 37% of sites |
| Consent setup dropping conversion data | 28% of sites |
| Duplicate tracking corrupting the counts | 20% of sites |
| Two or more properties collecting at once | 20% of sites that have analytics |
Sit with the first row. More than a third of live businesses actively spending on traffic have no functioning analytics. If that is you, every conversion conclusion you drew this year came from noise, and no amount of headline testing will fix it.
This is also the layer that invalidates everything downstream. A test needs a trustworthy measurement to declare a winner. If the measurement is wrong, the test is not inconclusive, it is actively misleading, and you will ship the loser believing it won.
Layer 2: does the buying path work for everyone?
The second layer is whether a real person can actually complete a purchase on the device, browser and connection they have.
This is where the expensive, invisible failures live. A checkout that throws an error on one mobile browser. A discount field that silently rejects valid codes. A shipping calculator that times out and renders nothing. A required field that sits off-screen at narrow widths.
None of these announce themselves. They surface as a conversion rate that is quietly worse than it should be, forever, or as a channel that appears to have "stopped working" when nothing about the channel changed. A broken checkout looks exactly like a dead ad channel, which is why so many teams spend weeks rewriting creative to fix a bug in their own funnel.
Layer 3: is it fast on the connection people actually have?
Speed is a conversion problem long before it is an SEO problem, and it has to be measured on a throttled mobile connection rather than on your laptop on office wifi.
What matters is not the score. It is how long until the buying action is usable, whether the layout shifts under someone's thumb while they reach for it, and what is blocking the main thread while they wait. A 95 on desktop and an unusable four seconds on a mid-range Android are entirely compatible findings.
Layer 4: now the page itself
Once the first three are clean, design and copy work is worth real investment, and this is where a traditional CRO audit begins: clarity of the offer, friction in the form, strength of the proof, mobile hierarchy.
We are not dismissing this layer. We are saying the order is not optional. Testing a headline on a funnel with broken tracking produces a result you cannot read. Testing anything on a checkout that fails for a slice of your traffic optimizes the visitors who were already fine and never touches the ones you are losing.
Two popular approaches that start at layer four
A new category of tools promises the site will improve itself. They are worth understanding, because the mechanism has a specific blind spot.
The autonomous variant shipper. Install a script, and an agent watches visitors, drafts new versions of your pages, runs them as live A/B tests, and deploys whatever wins. It is a genuinely impressive loop, and it inherits every problem in layer one. The agent declares winners using your measurement, so if consent is dropping a quarter of your conversions or two properties are collecting at once, the agent is not neutral about the error. It optimizes into it, confidently, at speed. And because it only ever generates variants, it can rewrite a headline a hundred ways without ever reporting that the checkout throws an error on one browser. That is not a variant problem. No amount of variant generation finds it.
The simulated shopper. Other tools model synthetic visitors, walk them through your store, and predict where real people would hesitate. Useful for exploring a design space cheaply. But a simulated shopper does not hit your real payment gateway, does not load your real third-party scripts, does not encounter your consent banner, and does not experience your checkout on a real mid-range phone on real cellular. The failures in layers one through three are precisely the ones a simulation cannot reproduce, because they are properties of your live environment rather than of your design.
Both approaches optimize the site you meant to build. The first three layers are about the site you actually shipped.
What to ask anyone offering a free CRO audit
The offers in this category look similar from the outside. These five questions separate them quickly:
- Does it verify my analytics, or trust it? If the audit reads your conversion rate without checking whether that number is real, everything it recommends inherits the error.
- Does it attempt a real transaction, or read the homepage? Layer two only exists for tools that actually try to buy something.
- Does it test mobile on a throttled connection, or report a desktop score?
- Does it price findings, or rank them with a severity label it invented? "High priority" is not a decision input. A dollar figure is.
- Does it hand me a PDF, or a fix? This is the one that separates a lead magnet from an audit. A list of problems is homework. Working code you can review and merge is a result.
Most free audits in this category answer "no" to the first two, which is why they can be free and instant: reading a homepage is cheap, and checking whether a business can actually take money is not.
Where this leaves you
If you only take one thing: fix the order. Verify measurement, verify that people can buy, verify it is fast, then optimize the page. Teams that run it in the reverse order spend a quarter testing copy on a funnel that was broken the whole time, and the test results tell them nothing was wrong.
If you want the first three layers checked against your live site rather than described in the abstract, that is exactly what our free scan does, with each finding priced and the fix shipped as a pull request.