Most conversion audits produce a list. Lists don't change anything, because a list has no order and no owner. This audit is designed to end with one fix, chosen deliberately, and a way to tell whether it worked.
Work the steps in sequence. Don't skip to the pages.
Step 1 — Establish what "converting" means here
Write down the single action that matters, per page type. A blog post's job is to move a reader onward; judging it on purchases will teach you nothing. A checkout's job is orders.
If you can't name the action, you cannot audit the page. You're just reviewing design.
Step 2 — Confirm your traffic contains candidates
Segment by source. If bounce is high and uniform, sessions short across every page, and the conversion rate immovable regardless of what you change — you have a traffic problem, not a page problem, and every hour spent on the pages is wasted.
Do this before anything else. It's the step most audits omit and the one that most often changes the answer.
Step 3 — Map the real path, not the intended one
Write your assumed funnel. Then check whether visitors actually take it. Many don't: they detour through docs, pricing, and back again.
An audit against a funnel nobody walks reports leaks that don't exist and misses the ones that do.
Step 4 — Find the cliff
For each transition, what fraction continues? One will lose far more than its neighbours. That's the location. Everything before this step was preparation; everything after is fixing the wrong thing.
Run it again for mobile separately. A large device gap at one step isolates interface friction, since offer, copy and price were identical.
Step 5 — Walk the failing step as a stranger
On a phone, on cellular, not logged in. Do the thing you want visitors to do, and note every moment that costs you a beat of hesitation.
You cannot see your own site. Every piece of knowledge you have — that shipping is calculated later, that the second button goes elsewhere — is scaffolding your visitor doesn't have.
Step 6 — Match the leak to its cause
Early leak: the promise didn't match the click, or the first screen didn't answer "is this for me." Decision-page leak: an unanswered objection — price, proof, fit, or what happens if they're wrong. Form leak: a field that's expensive, unexplained, hostile, or premature. Payment leak: a surprise cost, an absent trust cue beside the card, a missing payment method.
Step 7 — Choose one fix
Rank candidates by exposure times severity, then favour the later-stage one, then favour the one you can ship this week. Ship it. Watch the step you changed — not your total.
Then repeat from step 4. That loop, run six times honestly, will teach you more about your site than any forty-item report.
Defrixa's free scan compresses steps 5 and 6 for a single page: it reads structure, ranks what it finds, and names the biggest issue rather than listing all of them.
Common questions
Continuously and shallowly, rather than annually and exhaustively. A site changes; so does its friction.
An outside pair of eyes is genuinely useful for step 5, because they can see what you can't. The rest is arithmetic you can do yourself.
One fix, one reason, one metric to watch. If it produces forty items, it has deferred the decision back to you.