18 Prompts to Audit Your Landing Page With AI

An audit is not an optimisation. Optimising means changing things; auditing means producing an evidenced, prioritised list of what is wrong and how much each problem costs you. Do the second one first, because a change list without evidence is just opinion with a deadline.

The 18 prompts below produce that list. They work on any current model that accepts a screenshot or pasted HTML, they are organised by the six layers a real audit covers, and the last two turn every finding into a scored fix queue. Once you have the queue, our prompts for improving landing page conversion rates are the ones that do the fixing.

What a Real Audit Covers

Most "AI audits" return a tidy list of generic advice: add social proof, strengthen your CTA, improve mobile. That happens because the prompt asked for an opinion rather than an inspection. An inspection has a fixed scope. These are the six layers worth inspecting, in the order a visitor experiences them:

Run all six in order. Skipping to friction when the message is broken produces a faster path to a page nobody understood anyway.

Before You Start: What to Give the Model

Audit quality depends almost entirely on what you paste in. Give the model three things and the findings become specific rather than generic:

1. The page itself — a full-page screenshot, or the rendered HTML. A URL alone makes most models guess.
2. The traffic source — the exact ad, email or post that sends people there. Half of all landing page problems are message-match problems, and they are invisible without this.
3. One number — your current conversion rate and monthly visitors. It converts vague findings into prioritised ones, and it is what prompt 18 needs.

Everything in square brackets below is yours to replace. If a prompt returns something generic, the usual cause is a missing input rather than a weak prompt.

One practical note before the prompts. An audit is only worth running if you can act on what it finds, and on most teams the bottleneck is not the list — it is that changing the page means a ticket, a sprint or a developer who is busy. If that is your situation, moving the page somewhere you can edit it yourself is the highest-leverage fix on the list. Onepage is the pragmatic version of that: a drag-and-drop builder whose layouts do not break when you edit them, with a free plan covering one site and three pages, paid site plans from $19.90 a month, and a separate Forms and Funnels add-on (free up to 100 responses a month, then from $39.90) for multi-step forms and quiz funnels. The audit is the diagnosis; being able to ship the fix the same afternoon is the cure.

Message and Clarity (1–4)

1. The five-second inspection

Establishes whether the page communicates at all.

You are seeing this page for the first time and you get five seconds
before it disappears: [PASTE SCREENSHOT OR ABOVE-THE-FOLD HTML].

Answer only from what you actually saw, not from inference:
1. What is this product or service?
2. Who is it for?
3. What does it want me to do next?
4. What would it cost me?

For each question you could not answer, name the specific element that
should have answered it and say what it says instead.
Do not suggest improvements yet. Report only.

2. Message-match inspection

Finds the gap between the promise clicked and the page delivered.

If the source is an email sequence rather than an ad, audit the sequence alongside it — the patterns in our cold email prompt templates set the expectation this page has to confirm.

Here is the ad, email or post that sends traffic here: [PASTE].
Here is the landing page it points to: [PASTE].

Produce a table with four columns: promise made in the source,
where the page confirms it, how long it takes a reader to find that
confirmation, and a match score out of 5.

List every promise made in the source that the page never confirms.
List every prominent page element that the source never set up.

3. Specificity audit of the value proposition

Tests whether the copy could describe a competitor.

Here is my headline, subhead and primary value proposition: [PASTE].

Rewrite them as a checklist of factual claims. For each claim, state
whether it is specific to this product or could be copied onto a
competitor's page without changing a word.

Count how many claims are transferable. That count is the score.
Quote the three weakest phrases verbatim.

4. Jargon and comprehension pass

Surfaces the knowledge the page assumes.

Read this page as a competent person who has never heard of this
product or its category: [PASTE].

List every term, abbreviation, product name or benefit that assumes
knowledge I do not have. For each: quote it, say what it assumes,
and rate the damage as high, medium or low.

Flag any sentence that requires reading twice. Do not rewrite anything.

Structure and Hierarchy (5–8)

5. Section inventory and order

Compares the page's order against a buyer's question order.

List every section of this page in the order it appears, with a
one-line summary of each: [PASTE].

Then list the order in which a sceptical buyer forms questions:
what is it, is it for me, does it work, what does it cost, what happens
if it goes wrong, how do I start.

Show both orders side by side and name the single worst-placed section.

6. Visual weight versus importance

Finds what the design emphasises that the business does not.

Rank every element above the fold by visual weight, judging size,
contrast, position, colour and surrounding whitespace: [PASTE SCREENSHOT].

Now rank the same elements by their importance to the conversion
decision.

Present both rankings in one table and mark every row where the two
rankings differ by more than two places.

7. Above-the-fold inspection

Audits the only part of the page most visitors see.

Considering only what is visible before any scrolling, at 1440px wide
and at 390px wide: [PASTE SCREENSHOTS OF BOTH].

For each width, list what is visible, what is cut off, and whether the
primary call to action is reachable without scrolling.

Name anything occupying prime space that does not help someone decide,
including cookie banners, navigation and decorative imagery.

8. The scan test

Checks the page for people who never read a full sentence.

Extract only the elements a scanning reader actually registers from
this page: headlines, subheads, bold text, button labels, image
captions and list items: [PASTE].

Assemble them into a single document in page order.

Read that document back and answer: does it on its own explain what
this is, who it is for and what to do next? Quote the gaps.

Evidence and Trust (9–11)

9. Claim-to-proof inventory

The single most useful output of any audit.

Extract every factual or implied claim this page makes: [PASTE].

For each claim produce a row with: the claim quoted, the type of
evidence that would satisfy a sceptic (number, named customer, demo,
guarantee, third-party citation), whether the page provides it, and
how close that evidence sits to the claim on the page.

Sort the table so unsupported claims appear first.

10. Social proof credibility check

Distinguishes real proof from decoration.

Here is all the social proof on my page: [PASTE TESTIMONIALS, LOGOS,
COUNTS, RATINGS].

Score each item from 1 to 5 on credibility, judging specificity, whether
a real identifiable person or company is named, whether it mentions a
concrete outcome, and whether it addresses a real objection.

Flag anything that is unverifiable as written. State plainly which
items are doing no work and would cost nothing to remove.

11. Objection coverage audit

Finds the doubts the page never answers.

Product: [DESCRIPTION]. Price: [PRICE]. Audience: [AUDIENCE].

List the 12 reasons a qualified visitor who understands the offer still
does not convert, ranked by how often each is the real reason rather
than the stated one.

For each: say where on the page it is addressed, quote the text that
addresses it, or mark it UNADDRESSED. Do not write replacement copy.

Friction and Action (12–14)

12. Form field inspection

Every field is a cost; this prices them.

My form asks for these fields: [LIST, MARKING WHICH ARE REQUIRED].

For each field produce: who inside the business uses this data, at what
point in the process, what breaks if it is missing, and whether it could
be collected after conversion or enriched automatically from an email
address.

Classify each as necessary, deferrable or removable, and count each group.

13. Call-to-action inventory

Audits what the page asks for and how often.

List every call to action on this page in order, including buttons,
text links, forms and chat prompts: [PASTE].

For each: quote the label, state what a visitor reasonably expects to
happen when they click it, and name the intent level of someone who has
read that far.

Flag duplicates, competing actions on the same screen, and any label
that describes our internal process rather than the visitor's outcome.

14. Conversion path count

Counts the work between arrival and done.

Walk through every step from landing on this page to completing the
conversion: [DESCRIBE OR PASTE THE FLOW].

Count clicks, form fields, page loads, decisions and any moment where a
visitor must find information that is not on screen.

Produce the total for each category, then list every step that exists
for our convenience rather than theirs.

Technical and Accessibility (15–16)

15. Mobile and performance inspection

Most traffic is mobile; most audits are done on a laptop.

Audit this page for mobile conversion specifically, not just layout:
[PASTE MOBILE SCREENSHOTS AND HTML].

Check: whether the primary action is reachable without scrolling back
up, whether form fields trigger the correct keyboard type, whether tap
targets are large enough and far enough apart, whether a sticky header
or banner eats the fold, and how heavy the hero media is.

Rank findings by conversion impact, not by how easy they are to fix.

16. Accessibility audit

Missing focus states and absent error handling are the two failures this turns up most, and both are design-system problems rather than page problems — the state-completeness prompts in our vibe coding prompt set fix them at the source.

Accessibility failures are conversion failures with legal exposure.

Audit this page against WCAG 2.2 AA: [PASTE HTML].

Report in a table: the guideline, the element, the failure, and the
severity. Cover at minimum colour contrast ratios with the measured
values, visible focus states on every interactive element, form labels
and error associations, heading order, image alternative text, and
whether the whole conversion path can be completed by keyboard alone.

Mark anything that would block a screen reader user from converting.

Scoring and Reporting (17–18)

17. The scored audit report

Turns 16 outputs into one document you can send.

Here are the findings from my landing page audit: [PASTE THE OUTPUTS
OF THE PREVIOUS PROMPTS].

Produce a single audit report with: an overall score out of 100 with the
weighting shown, a score out of 10 for each of the six layers (message,
structure, evidence, friction, technical, reporting readiness), the
three most damaging findings stated in one sentence each, and a table
of every finding with its layer, severity and supporting evidence.

Do not include recommendations. This document is the diagnosis only.

18. The prioritised fix queue

Converts the diagnosis into an ordered plan.

Here is my scored audit: [PASTE]. My page gets [VISITORS] visitors a
month and converts at [RATE]%.

Score every finding on impact (1-10), confidence that fixing it helps
(1-10), and effort (1-10, where 10 is hardest). Compute impact times
confidence divided by effort and sort descending.

Mark which items are large enough to detect as a test at my traffic
level and which must simply be shipped on judgement.
Output as a table I can paste into a project tracker.

Running the Audit as One Pass

If you would rather not run eighteen prompts by hand, the sequence works as a single brief for an agent: give it the page, the traffic source and the numbers, then instruct it to work through the six layers in order and stop for review after each. The reason to stop between layers is the same reason to run them in order — findings in the message layer change what counts as a finding in every layer below it.

Two habits make the output worth more. Record the date and the page version with every audit, because an audit is only useful against a baseline. And keep the diagnosis and the recommendations in separate documents, so nobody reads a fix list and assumes the evidence behind it.

Pro tip: Re-run prompts 1, 2 and 9 after every significant page change. Those three catch the most common regression: a page that drifts away from the ad that feeds it while picking up claims nobody can substantiate.

Five Mistakes That Produce Useless Audits

1. Asking for opinions instead of inspections. "What do you think of my landing page?" returns advice that fits any page. Every prompt above asks the model to report what is there, not to imagine what could be.
2. Auditing without the traffic source. Message match is the largest single category of landing page failure and it cannot be seen from the page alone.
3. Letting the model fix things mid-audit. Once it starts rewriting, it stops looking. Keep diagnosis and treatment in separate passes.
4. Treating every finding as equally urgent. A 40-item list with no scoring gets ignored. Prompt 18 exists so the list arrives ordered.
5. Auditing a page you cannot change. If shipping a fix takes three weeks, the audit will be stale before it is actioned. Fix the publishing bottleneck first.

What to Do With the Results

If the queue says the page itself is the constraint, our comparison of AI landing page builders covers what it costs to move it somewhere you can edit and test freely. A finished audit gives you a score, six layer scores and an ordered queue. Work the queue from the top, ship in batches small enough to attribute, and re-run the audit a month later against the same prompts so the two scores are comparable.

Where the queue says the page needs restructuring rather than editing, the fastest route is usually to rebuild the page rather than patch it — and if the audit was triggered by a page that an AI tool generated in the first place, our prompts for fixing AI-generated pages cover that specific repair job in dependency order.

Keep Reading

Creating a landing page with AI prompts covers the build side, and funnel builder software is the comparison to read if the audit says you need checkout and upsells rather than a single page. Browse all prompts on PromptsRush.