You open a page that isn’t getting enquiries. You move some things around. You change the button colour. It still doesn’t convert. The problem wasn’t the button.

The real mistake is redesigning before you look at the data

Moving pixels feels productive. It isnt. Without funnel data you’re guessing which leak to fix, and you’ll almost always pick the one that’s most visible, not the one that’s most damaging.

This is from a real week of conversion work on a design web app with Stripe payments. The same mistakes kept surfacing. Once they were named, the fixes were fast. The workflow that came out of it transfers to any site.

The framing problem with AI conversion advice

Ask an AI to “make this better” and it will return visual polish: tighten the spacing, round the corners, soften the palette. Useful if your problem is aesthetics. Useless if your problem is that people are leaving without buying.

Ask the same model, with the same screenshot: “what’s failing here for conversion?” and you get a completely different answer. It names the sales problems. Thats the whole trick. Same tool, different frame.

Three real examples from one week

1. A Save/Cancel dialog was the highest-intent moment in the app

The analytics showed users hitting a plain Save/Cancel confirmation box at a critical point in the flow. The AI flagged it immediately when asked what was failing: this was the moment a user had just done the work, they were invested, and the app greeted them with a checkbox interaction.

That’s the highest-intent moment in the whole product being squandered. It became a sales surface instead: a short prompt, a clear value statement, a single action. One screen, no redesign of anything else.

2. The paywall screen had the worst drop-off in the funnel

The analytics event data made this unambiguous. The paywall had the steepest exit rate. So that’s where the work went, not a hunch about the homepage.

Four specific fixes came out of asking what was failing:

  • The header was 180px tall on mobile. The offer wasn’t visible above the fold. Compressed to 80px.
  • The headline described features. It was rewritten around the outcome the user actually wanted.
  • The real product logo wasn’t on the screen. Adding it created visual continuity straight into the Stripe checkout, so it didn’t feel like leaving.
  • Social proof existed but sat in 11px grey text at the bottom. Moved it next to the value proposition, readable size. The number was real: 1,500+ creators already using it.

None of this required a redesign. It required knowing where people were leaving and asking the right question about why.

3. Google Analytics plus Search Console surfaced ten new landing pages in one session

Connecting both tools and feeding the data to the AI produced a ranked list of gaps: countries sending organic traffic with no dedicated landing page, pages with 50% engagement rates while sibling pages sat at 80% (fix those first, don’t build new ones), and content tags pulling search traffic with no page behind them at all.

One session. Ten specific, prioritised things to build or fix. That list came from evidence, not from someone’s instinct about what the site “needed.”

The four-step workflow

This is the transferable version. Run it on your own site before you touch a single element.

  • Pull the data first. Google Analytics engagement rate by page, exit events, funnel drop-off points. Search Console for queries you rank for but have no page targeting. You need evidence of where the leak is before you fix anything.
  • Feed evidence, not vibes. Give the AI the actual numbers alongside the screenshot. “This page has a 48% exit rate. Here’s what it looks like.” Context changes the output entirely.
  • Frame the question for conversion. Not “how do I improve this?” or “make this better.” Ask: “what is failing here for conversion, and why?” You’ll get problems named, not aesthetics adjusted.
  • Iterate, don’t accept the first answer. Push back. Ask which fix would move the needle most. Ask what the user is thinking at this exact moment. Ask what’s missing that a buyer would need. The first response is a starting point.

What this looks like in practice

A rough prompt structure that works:

I’m looking at [page name]. Analytics shows [exit rate / drop-off event / engagement %].
Here is a screenshot. The user at this point has already [done X].
What is failing here for conversion? What would a buyer need to see
that they currently can’t, or what is making them hesitate?

Swap in your real numbers. The more specific the evidence, the more useful the answer.

For the Search Console gap analysis, pull your top queries and pages into a spreadsheet and ask:

Here are my top 40 search queries and the pages currently ranking for them.
Which queries have no dedicated landing page? Which pages have low engagement
compared to similar ones? Rank the gaps by likely conversion impact.

You’ll get a prioritised to-do list, not a general recommendation to “create more content.”

The short version

Dont redesign. Diagnose. Find the leak with data, name it precisely, then ask AI what’s failing at that specific point. The answer will be different, and better, than anything you’d get from a vague brief about improvement.

If your site is getting traffic but not enquiries and you want a second pair of eyes on the funnel, let’s have a conversation.