Landing page teardown: a severity-ranked read against the ad promise

Most landing page reviews are a list of opinions with no order to them, so nothing gets fixed. The useful question is narrower: does the page keep the promise the ad made, and where is it leaking. This prompt reads the page against the specific promise in your ad and returns findings ranked by severity, each written the same way (what it saw, why it matters, what to change) so you can work top down.

It covers five areas: message match against the ad, clarity of the offer, evidence and trust, friction in the path, and the form itself. It ends by naming the three changes to make first, so a long list becomes a short plan. It is analysis only; it edits nothing and builds nothing.

The prompt

You are a conversion analyst reviewing a landing page that receives paid
traffic. Judge the page only against the ad promise I give you and the copy I
provide. Do not invent page elements you cannot see, and do not assume traffic,
conversion rates or test results that I have not shared. If you are inferring
rather than observing, say so. Write in British English.

INPUT
- The ad promise or offer (what the click was sold on):
- The landing page copy or URL:
- Primary conversion action (what a visitor is meant to do):

TASK
Review the page across these five areas:
1. Message match: does the page deliver the exact promise the ad made?
2. Clarity: is the offer and its value understandable in one read?
3. Evidence: are claims backed by proof (reviews, logos, data, guarantees)?
4. Friction: what slows or distracts from the conversion action?
5. Forms: is the form asking for the right fields, in the right order?

For every finding, write three lines:
- Observation: what you see on the page.
- Why it matters: the effect on conversion, tied to the ad promise.
- Suggested fix: a specific, doable change.

Rank all findings by severity, most damaging first. Severity is about impact on
conversion, not tidiness.

Finish with a section headed "Fix these three first" listing the three changes
that would move the needle most, and one short "Assumptions" list covering
anything you inferred or could not see.

How to use it

Paste the page copy directly, or a URL if your model can browse (many cannot, in which case copy the text yourself so it is judging what is really there, not a hallucinated version). Fill in the ad promise precisely, the exact words or offer, because message match is the highest-value part of the review and it is only as good as the promise you give it. A mid-tier model is enough. Take the "Fix these three first" list as your sprint, and keep the rest as backlog. This finds copy and structure problems, not speed or technical ones; pair it with a real page-speed tool for the load-time side.

Failure modes

• It reviews a page it cannot actually see. Given only a URL, a non-browsing model will confidently describe a generic page. Countermeasure: paste the real copy, and if you give a URL, confirm the observations match the live page before acting.
• It invents proof or friction that is not there. Asked for evidence gaps, a model may list trust elements as missing when they exist lower down, or vice versa. Countermeasure: the "do not invent page elements" rule and the assumptions list catch most of it; verify each observation against the page.
• Severity ranking drifts to tidiness. Models sometimes rank a cosmetic nit above a broken message match. Countermeasure: the prompt defines severity as conversion impact; if the top item is trivial, tell it to re-rank by impact on the conversion action.