4
10 Comments

I used my own AI critic on my landing page — and it changed the product

I built Landing Page Critic to help founders understand why a landing page may look polished but still fail to communicate its value clearly.

Before asking anyone else to trust it, I tested it on Tryvorax itself.

The first reports were uncomfortable but useful. They repeatedly pointed out three issues:

• The homepage tried to explain too many features at once.

• Several calls to action competed for attention.

• Terms such as Brand Score and Brand Brief made sense to me, but not necessarily to a first-time visitor.

I used that feedback to simplify the main promise, unify the primary call to action and explain the product’s decision signals in plain language. The clarity score improved from roughly 65–70 to 85/100.

The more interesting lesson was not the score. It was seeing whether the tool could identify specific changes I was actually willing to make.

That is now the standard I’m using while developing it: the report should not merely sound intelligent; it should help a founder decide what to change next.

There is a free snapshot and a €6.99 full report with deeper diagnosis, suggested copy, experiments and a 7-day action plan:

https://tryvorax.com/tools/landing-page-critic

For founders who have tested their own landing pages: what feedback made you change something immediately, rather than just agree with it?

posted toAvatar for product Tryvorax
Tryvorax
  1. 1

    "The report should not merely sound intelligent; it should help a founder decide what to change next" is the whole game, and most AI output tools miss it.

    We ran into the same problem building a strategy-analysis tool: early versions produced write-ups that read as confident and polished but didn't point at a specific, checkable next action. The fix wasn't better writing, it was forcing every claim to either cite what it was based on or say plainly there wasn't enough basis to make the claim at all. Founders trusted the confident version less once they'd seen the honest "not enough data" version — same as your clarity score going up once you cut clutter rather than added polish.

    The feedback that changed something for me immediately is always the one with a name and a location — "this sentence, replace it with X" — never a general score. Sounds like your critic already leans that way.

    1. 1
      Exactly — “a name and a location” is a very useful standard. The critic already tries to ground findings in text extracted from the page, but there is a meaningful difference between quoting general evidence and saying: “In the hero, replace this exact sentence with this candidate version.” I also agree that “not enough evidence” should be a valid output. A confident diagnosis without sufficient page evidence damages trust more than an honest limitation. The next iteration should therefore make every priority recommendation include: • The page section or location • The exact evidence it is based on • One candidate replacement • The behavior or metric the change is intended to influence • An explicit “insufficient evidence” state when appropriate That would make the report less like an AI opinion and more like a checkable editing decision. Thanks — this is genuinely useful product feedback.
      1. 1

        This lines up well with the separation we talked about elsewhere in this thread — a recommendation with a location, cited evidence, and a candidate replacement is something a human can actually check against the page, rather than take on faith. The one thing I'd watch for once this ships: make sure "insufficient evidence" isn't just a softer wording of "confident, but wrong" — i.e. the critic should be able to fail closed into that state even when it has some weak evidence, not only when it has none. Otherwise it'll drift back toward always finding a candidate.

  2. 1

    The filter that matters for a critic like yours is the 'so what' test. A finding becomes decision-useful when it has an owner, one candidate edit, and a metric the edit should move, not just a score. The review that made me change something immediately was one that named a single swap: replace a feature-list subhead with the outcome a first-time visitor wants. A full report that ends with exactly one candidate edit is easier to trust than one that ends with a diagnosis map.

    1. 1
      That’s a very useful distinction — especially the difference between a diagnosis and something a founder can act on immediately. The free snapshot currently ends with one priority action, but your framework makes the next improvement clearer: that action should identify an owner, propose one concrete edit, and state the metric or behavior it is intended to influence. I also like the idea that the full report should still culminate in one recommended first move, even if it contains a broader diagnosis. A long list can feel comprehensive while making the actual decision harder. I’m going to test this structure in the next iteration. Thanks for making the “so what?” test so practical.
  3. 1

    Using an 'AI critic' to audit your own landing page is a meta move! It's so hard to see your own product objectively.

    I might try this for my new project, Muzegen (a French AI music generator). We've been struggling to explain the technical synthesis part to non-musicians. Did your AI critic give you advice on the technical copywriting or more on the emotional appeal? This seems like a great way to optimize conversion rates.

    1. 1
      Thanks — that was exactly the challenge I wanted to explore. The critic looks at both sides: whether the technical explanation is understandable to a non-expert, and whether the page makes the customer outcome and emotional benefit clear enough. For Muzegen, I’d expect it to check whether visitors first understand what they can create and why it matters before encountering the details of the synthesis technology. The technical explanation can still be there, but it should support the main promise rather than compete with it. I’d be genuinely curious to see how it evaluates Muzegen. You can run a free snapshot at https://tryvorax.com/tools/landing-page-critic — if you try it, I’d really value hearing what it gets right or wrong.
  4. 1

    The strongest part is that you tested the critic on your own product and then actually changed the landing page from the findings. That makes the tool more credible than a report that simply sounds insightful.

    1. 1
      Thanks, Aryan — that was important to me. I didn’t want the critic to produce feedback that merely sounded intelligent. It needed to identify something concrete enough to change, and then help me evaluate whether that change improved the page. Testing it on Tryvorax also revealed a few false positives and inconsistencies, which helped me improve how the critic uses evidence from the actual page. I’m continuing to refine it through real landing pages and honest user feedback.
      1. 1
        That’s useful context. The fact that using it on your own product exposed false positives as well makes the validation more credible.