How to Audit a Landing Page With Evidence Instead of Opinions

A landing-page audit becomes useless when every comment is a preference wearing a lab coat.

“Make the headline punchier.” “Add more social proof.” “Move the CTA above the fold.” These can all be reasonable ideas. None is a diagnosis until it names the evidence, the user risk, the proposed mechanism, and the result that would falsify it.

The better method starts by separating four layers.

Layer one: what is observable on the page

Record facts before interpretation:

• the exact headline and primary CTA;
• where price, requirements, proof, terms, and privacy appear;
• the number and type of form fields;
• visible testimonials and their attribution;
• heading order and internal links;
• whether a keyboard can reach and operate the main action;
• whether the mobile page hides, wraps, or overlaps critical content;
• which analytics event is supposed to represent success.

This layer should be reproducible. Two reviewers can disagree about quality while agreeing on what the page contains.

Layer two: the heuristic finding

Map the evidence to a declared rubric. Rick’s public rubric covers offer clarity, audience fit, primary action, proof, risk handling, hierarchy, friction/accessibility, and measurement readiness.

A good finding sounds like this:

The hero says “unlock intelligent growth” but never identifies the product category, buyer, or concrete output. A first-time visitor cannot reliably explain what is being sold before the first CTA.

That is different from “the copy is vague.” It anchors the judgment to visible language and a specific comprehension job.

Layer three: the business hypothesis

Now state the proposed mechanism:

If the hero names the product category, intended buyer, and deliverable, more qualified visitors should continue to the product details or start the primary action because they can self-select sooner.

The hypothesis may be wrong. Perhaps the traffic is unqualified, the offer is weak, the price is wrong, or the product has no proof. The audit does not know from page copy alone.

Layer four: the verification plan

Define the event and observation window before shipping the change. Useful outcomes might include qualified form completions, completed diagnostics, successful installs, checkout starts from the intended segment, or fewer support questions about basic scope.

Avoid declaring success because the new copy sounds better. Verify instrumentation, preserve the previous version, record the release date, and watch guardrail metrics such as form quality, refunds, or downstream activation.

What automated audits can and cannot do

An automated system can inspect visible copy, headings, links, forms, common accessibility signals, price visibility, proof patterns, and basic technical metadata. It can compare the page with a transparent rubric and produce a prioritized hypothesis list.

It cannot honestly prove that a headline will convert, that a testimonial is authorized, that a screenshot represents a current product, that an analytics event is correct, or that the offer is wanted. Those require source records, manual testing, users, or first-party data.

Prompt injection is another limit. A landing page is untrusted input. Text on the page should never be allowed to rewrite the audit’s instructions, trigger tools, or inject executable HTML into the result. The result renderer should treat model output as text or validated structured data.

A practical audit order

1. Confirm the page has one canonical job and one primary conversion action.
2. Read the hero without brand context. Write what the buyer, product, outcome, and action appear to be.
3. Trace proof. Ask what each metric means, when it was observed, and where it came from.
4. Follow every primary CTA and complete the path on mobile and keyboard.
5. List the three largest uncertainties a visitor must accept before acting.
6. Check whether the page answers those uncertainties with evidence or merely repeats the promise.
7. Choose one coherent hypothesis rather than twenty unrelated cosmetic changes.
8. Define the outcome, guardrail, baseline, and review date.

Use a score carefully

A score helps sort work. It should not turn uncertain judgments into false precision. A page with a clear offer and strong proof may deserve priority over a visually polished page with perfect metadata. A page with a high score may still serve the wrong audience.

Publish the rubric, weights, limitations, and version. When the method changes, say so. When a finding depends on external research, cite the scope of that research instead of promising a universal lift.

You can inspect Rick’s landing-page audit rubric or run the free automated Roast. Treat the output as a structured first pass: evidence, risk, hypothesis, verification—not a prophecy.