I built “Lighthouse, but for AI” would love feedback
Hey everyone đź‘‹
I recently shipped a small tool called AI Lighthouse:
https://getlighthouse.dev/
The idea came from a question I kept running into while building web products:
How does my website actually look to AI systems?
We spend a lot of time optimising for users and search engines, but LLMs, AI search, and voice assistants are increasingly reading and summarising our sites, often very differently from how browsers do.
AI Lighthouse tries to make that layer visible.
Right now it focuses on things like:
- content extractability
- structural clarity
- where context gets lost for AI
- potential AI misunderstanding/ hallucination triggers
It’s not an “AI SEO” tool, and it doesn’t magically optimise anything.
It’s more of a debugging/ inspection tool, so you can see what’s going on first.
A few things I’m unsure about and would love input on:
Is this a real problem you’ve run into?
Who do you think this is most useful for (devs, founders, marketers)?
What would make this genuinely valuable for you?
It’s still early and a bit rough around the edges, but usable.
Any feedback, good or bad, would be super helpful.
The idea of “AI misunderstanding / hallucination triggers” caught my eye, that’s a layer most people probablly aren’t thinking about yet. I’m wondering how obvious those issues look in practice vs only after you’ve already been burned once.
I think that’s mostly true. A lot of these issues only become obvious after you’ve been burned once wrong summaries, missing context, AI confidently saying the wrong thing.
The goal here isn’t to predict every edge case, but to shorten that feedback loop so people can catch issues earlier instead of learning the hard way.
This looks like a thoughtful approach to making AI interpretation visible, which is a real blind spot for a lot of founders and teams right now. One thing that might help clarify the problem you’re solving is tying it to the decision it leads to.
For example, does running a URL through AI Lighthouse help someone decide whether to restructure content before publishing, change how they write pages for AI summarization, or adjust marketing copy when targeting AI-driven channels? If there’s a clear “do this because the AI can’t extract X,” that helps separate curiosity from urgency.
Curious if you’ve seen cases where someone ran the tool, saw something unexpected in the report, and then made a concrete decision (like redesigning a section or rewriting content) based on that output. Those stories help clarify who this tool is most useful for.
Yeah, so it does give you actionable things that you can do to improve your score, and it shows the score improvement that you expect.
I think the tool is useful for devs in general, since I am planning to release a CLI too for this and mostly aim to help people fix things that might matter to AI but wont to humans.
That makes sense, especially the part about things that matter to AI but not to humans.
I’d be curious how the score improvement ties to a real trigger for devs. Does this become something they run as part of a deploy or CI check, or only when they already suspect AI interpretation might be an issue? The CLI angle feels strong if it naturally fits into an existing workflow.
Have you noticed whether users come in with a specific problem in mind, or whether the tool surfaces issues they didn’t realize could matter yet?
Yeah, that is the intention with building a CLI and then eventually a CI check, Ideally I think the best time to use it would be when the copy or structure of the product page changes.
From the people I have talked to who have tried it, they have found issues when they used it, which they didn't even know about.
But I am still experimenting and working on fine tunning the logic and scoring so that it feels rewarding to actually implement the fixes suggested by the tool.