Start with the conversion, not the screenshot.
A CRO audit finds what might stop a visitor from converting. The useful output is a fix order: what to change first, why, and how you'll know it helped.
Start with one sentence: For this audience, arriving from this source, success means this completed action.
“More engagement” gives you nothing to audit. “New visitors from our search campaign complete a demo booking” gives you an audience and an endpoint. The click matters only if the booking works.
Record the URL, device, source, conversion event, and reporting window. Define the denominator, such as eligible visitors, and keep it consistent. Check for internal traffic and duplicate completions.
Walk the route in an approved test flow. Check the button, error recovery, and success event. If the event fires before completion or fires twice, fix measurement before trusting the numbers.
Give every finding an evidence label.
“People don't trust this page” is a guess until you can show why.
Write each finding in three lines:
- Observation: the form requires a phone number without explaining why.
- Evidence: a screenshot confirms the field; a usability participant asks whether they'll receive a sales call.
- Hypothesis: explaining the field may reduce hesitation and increase completed bookings.
Add the source, date, and segment. Analytics shows where people leave. User research helps you investigate why. One participant's concern does not tell you how common it is.
Contentsquare's audit guide combines behaviour data and feedback. CXL's research framework treats page-review findings as areas to investigate. Use both to keep your audit grounded.
No analytics? Run a structural review and label the gaps. An AI estimate is not measured abandonment.
The five pillars to check.
Review the page through these five lenses. Keep its job in view: a checkout needs a different fix order from a campaign landing page.
1. Clarity
Can a new visitor explain the offer and what happens next? Compare the ad with the headline and button. Mark the words that don't match. See what to fix in your hero copy.
2. Visual hierarchy
Is the primary action obvious on desktop and mobile? Look for buttons competing in size, contrast, and position. Use the three-CTA guide to check the hierarchy before removing a useful secondary path.
3. Friction
Try the action. Check unnecessary fields, repeated information, surprise requirements, and errors. Test keyboard access and relevant browsers. Reproduce a broken control before adding it to the fix list.
4. Trust
Does the page answer the doubt that appears before commitment? Check the terms and proof beside the action. Keep claims accurate and testimonials attributed. See which trust signals belong where.
5. Cognitive load
How much must the visitor decode before acting? Check jargon, repeated choices, and conditions scattered across sections. Ask a representative user to explain the next step. If they can't, find what made it hard.
Save a concrete observation for each pillar. “No issue found” is a valid answer.
Pick three. Put them in order.
Repair confirmed blockers first. A form that cannot submit needs a fix, not an A/B test that leaves half your visitors stuck.
Rank the remaining issues using four questions:
- Consequence: how directly does it affect the conversion?
- Reach: who encounters it? Use actual counts or mark it unknown.
- Evidence: is it reproduced, supported by research, or still a guess?
- Effort: what does it take to ship and reverse the change?
Choose three and write one sentence explaining their order. Skip invented impact scores. If two issues are close, start with the one you can test cleanly. Put the rest in the backlog.
A worked example: the demo page.
Illustrative example only. These invented numbers and research notes are not Levri customer results or benchmarks.
A software team audits a mobile demo page for one search campaign. In four weeks, 1,000 eligible visitors start 80 forms and complete 20 bookings: 2% visitor-to-booking conversion, 25% form completion. The events are verified. The reasons for leaving are still unknown.
The form works. After reviewing the five pillars, the team ranks three issues:
- Phone-field uncertainty. Three of four usability participants ask whether submitting means receiving a sales call. The sample is small, but the concern is specific and a copy change is easy to test.
- Three equal buttons. “Book a demo,” “Start a trial,” and “Watch a video” compete visually. Review click paths before deciding which alternative to demote.
- Generic headline. “Work smarter, together” says little about the offer. Check visitors' understanding against the campaign promise before rewriting it.
The phone field comes first because it has specific user evidence, sits near completion, and needs little work to test. If few visitors reach the form, revisit the order.
Brief the first experiment.
Continue the example. The team confirms the phone number is used for the scheduled demo call. Now it can write truthful helper text.
The brief:
- Audience: new mobile visitors from the campaign, through confirmed booking.
- Hypothesis: explaining the phone requirement will reduce uncertainty and increase bookings.
- Change: add “We'll use this number for your scheduled demo call” beside the field. Keep the other copy, fields, and buttons unchanged.
- Primary metric: unique confirmed bookings divided by eligible visitors assigned to each variant, using the same conversion window.
- Diagnostic: form completion among starters. It explains the step; it doesn't replace the primary metric.
- Guardrails: submission errors and qualified-demo rate, measured after the qualification window closes.
- Test plan: random, persistent assignment to the original or helper-text version. Use the verified baseline and minimum worthwhile effect to calculate sample size and duration. Agree the uncertainty threshold and stopping rule before launch.
- Owner and decision: the growth lead measures; the developer ships and checks both versions. Roll out only when the agreed primary-metric criterion is met without unacceptable guardrail harm.
The example's 1,000 visitors are not a test sample recommendation. If traffic is too low, use more task-based research or monitor a documented usability change. Before-and-after results cannot establish causal lift. An inconclusive test stays inconclusive.
AI Analysis → Funnels → Variant Studio.
AI Analysis surfaces page issues to investigate. Add your audience context and evidence to the findings. The score assesses the page; it doesn't measure conversion.
Funnels extends the review across steps. Diagnose estimates weaknesses from public URLs. Explain uses funnel data you provide. Keep estimates separate from observed counts.
Variant Studio turns the chosen issue into a page concept. Keep the brief beside it so the design stays focused. Your team still implements the change and runs the experiment.
One page. Three issues. One measurable next step.
Fix these first.
Before handing over the audit:
- Define the audience, completed conversion, and denominator.
- Verify the events and reporting window.
- Walk the route through completion and error recovery.
- Check clarity, hierarchy, friction, trust, and cognitive load.
- Separate each observation from its evidence and hypothesis.
- Rank three issues and explain the order. Repair confirmed blockers.
- Brief one change with a metric, guardrails, owner, and decision rule.
Can't explain why issue one comes first? Gather the missing evidence. Can explain it? Hand over the brief and get the first change moving.
Frequently asked.
What should a CRO audit include?
A defined audience and conversion, working measurement, a review of the page and its next step, and findings backed by evidence. Finish with three ranked issues and one experiment brief.
Can I run a CRO audit without analytics?
Yes. Walk the page and ask representative users to complete a task. Label what you observe and what you suspect. A page review can identify risks, but it cannot measure conversion or drop-off.
How do I choose what to fix first?
Repair confirmed blockers and broken measurement first. Then compare consequence, reach, evidence strength, and effort. Give a supported finding priority over a visual preference.
Does a higher CRO score prove that conversion improved?
No. A score assesses the page. Measure outcomes with verified conversion events and an appropriate test. AI recommendations and estimated drop-off still need validation.