How to Write a CRO Audit Prompt That Scores Conversion Blockers by Impact

Most "CRO prompts" you'll find online are a wall of 20 prompt ideas with no shared output shape. You paste one in, get three paragraphs of advice back, and still don't know which fix to ship first. The job a cro audit prompt actually has to do is narrower: look at one landing page or funnel step, find the things hurting conversion, and rank them so the highest-impact fix is line one.

That ranking is what almost no published prompt delivers. The pages that rank for this query give you copyable text but stop short of a scored output contract. So the model returns whatever shape it feels like that day, and two runs on the same page disagree on what matters.

This post shows how to build a conversion audit prompt with a locked, scored output — and how a prompt pack turns it into something you reuse on every page instead of rewriting from scratch.

What a scored CRO audit prompt does that listicles don't

A CRO audit prompt is a reusable instruction that makes a model inspect one page or step, surface conversion blockers, and score each by impact and effort. The score is the point. Without it you get advice; with it you get a queue.

Look at what currently ranks. VWO's roundup gives 11 copyable prompts organized by theme (testing ideas, pain points, visitor-focused copy), but none of them score or rank blockers by conversion impact. You get a prompt per task, not a prioritized audit. That's the gap. The reader still has to decide what to fix first, which is the hard part.

Here's the opinionated take: a flat checklist is worse than useless on a page that converts at 1.8%, because it treats a missing trust badge and a broken mobile CTA as the same weight. They aren't. The whole value of an audit is the ordering. If your prompt doesn't force a rank, you've automated the easy 20% and skipped the part that decides where the next sprint goes.

What you can use this prompt for

• Auditing a single landing page before a launch, ranked by what's likeliest to lift conversion
• Triaging one funnel step (checkout, signup, pricing) when its drop-off spikes
• Comparing two page variants and getting a scored reason one should win
• Turning a heatmap or session-replay summary into a ranked fix list
• Generating test hypotheses for an A/B tool, each tied to a blocker
• Producing a stakeholder-ready audit table instead of a Slack paragraph

The anatomy: variables, prompt, scored output

A reusable audit prompt has three parts. The variables hold the page-specific inputs. The prompt body sets the role and the scoring rules. The output contract locks the shape so every run is comparable.

Variables:
  {{page_url_or_copy}}   -> the landing page copy, structure, or URL context
  {{conversion_goal}}    -> the single action the page should drive
  {{funnel_step_data}}   -> optional: traffic + conversion for this step
  {{audience}}           -> who the page targets (intent + objections)

Prompt:
  Role: senior CRO auditor.
  Task: inspect the page against the conversion goal, list blockers,
        score each by Impact (1-5) and Effort (1-5), rank by Impact/Effort.

Output contract (locked):
  | Rank | Blocker | Why it hurts conversion | Impact | Effort | Suggested test |
  Then: top-3 fixes to ship this sprint, one line each.

That Impact/Effort ratio is doing the real work. It's why a high-effort redesign idea sinks below a one-line headline rewrite that moves the same needle. State the scale explicitly in the prompt, or the model will invent its own and the numbers won't mean the same thing twice.

Step-by-step: run the audit

1. Gather inputs

Pull the page copy (or a clean text export), the one conversion goal, and the step's numbers if you have them. Don't paste the whole site. One page, one goal. An audit that tries to cover a homepage and a checkout in one shot returns mush.

2. Fill the variables

Set {{conversion_goal}} to a single action — "start free trial," not "engage users." Set {{audience}} with the objection they arrive carrying ("skeptical, comparing three tools, worried about migration"). The objection is what the blockers get scored against.

3. Run the prompt

Run it at a low temperature for the scoring pass. Extraction and ranking want consistency, not flair. If the model returns prose instead of the table, restate the column headers on the final line and re-run. That's usually the fix.

4. Sanity-check the ranks

Read the top three. If the model ranked a cosmetic nit above a broken primary CTA, the Impact scale wasn't anchored. Add one example to the prompt ("a hidden CTA on the primary step is Impact 5"). Re-run once.

5. Turn blockers into tests

Each row already carries a suggested test. Feed those into your A/B workflow. A scored audit pairs naturally with a structured experiment write-up. See the A/B test comparison prompt for turning a blocker into a clean hypothesis and variant spec.

Prompt-craft patterns that make the audit hold

Anchor the scoring scale with one example. Don't just say "score Impact 1-5." Add: "Impact 5 = blocks the primary conversion action for most visitors; Impact 1 = minor polish." Without an anchor, the model compresses everything into 3s and 4s and the ranking flattens.

Score each blocker:
  Impact 5 = blocks the primary action for most visitors
  Impact 1 = cosmetic, unlikely to move conversion
  Effort 5 = needs design + dev + a sprint
  Effort 1 = a copy change someone ships today
Rank by Impact divided by Effort, highest first.

Force a refusal boundary on invented data. Models love to assert a page "loads slowly" with no signal. Add: "If you can't observe it from the input, say 'not assessable from input' instead of guessing." That single line is what keeps a CRO audit honest. It's also the line most listicle prompts skip, which is why their output reads confident and wrong.

Separate observation from recommendation. Make the model name what it sees before it prescribes. A blocker row that says "no pricing visible above the fold" is checkable. A row that says "improve trust" isn't. The output contract should demand the observation in column three.

Variables you'll set

Getting started

1. Copy the three-part scaffold above into ChatGPT or Claude.
2. Fill {{conversion_goal}} with one concrete action.
3. Paste one page's copy into {{page_url_or_copy}}. Just one.
4. Add the objection your audience carries into {{audience}}.
5. Run at low temperature and read the top three ranked blockers.
6. Restate the table headers on the last line if the format drifts.
7. For a maintained version with the scale anchored and the refusal boundary built in, reach for the CRO Audit Prompt Pack instead of re-deriving the scaffold each time.

A scored audit is only half the loop. The other half is diagnosing where in the funnel the leak starts before you audit the page — that's what the funnel analysis prompt covers, turning step data into ranked drop-off hypotheses. And once a page passes its audit, the user story acceptance criteria prompt keeps the fix from regressing when engineering picks it up. Browse the full catalog of conversion and marketing packs, or pair this with the Funnel Diagnostics Prompt Pack for the upstream half of the same job.