AI agents are changing how people discover and use websites. LucidVia helps you understand how well your site works for them by auditing crawlability, AI readability, structured data, navigation, accessibility, AI commerce readiness, and GEO. Every recommendation is backed by deterministic evidence, so you get actionable insights instead of AI guesswork.
No reviews yetBe the first to leave a review for LucidVia
Maker
📌
Hey Product Hunt! 👋 I'm the solo founder behind LucidVia.
The idea came from a simple question I couldn't answer: **Can AI agents actually use my website?**
There are plenty of SEO tools, and Google recently added basic agentic-browsing checks to Lighthouse, but I wanted something that explains *why* AI agents struggle with a site and how to fix it.
The biggest challenge was the scoring model. I didn't want an LLM to simply "rate" a website. Instead, LucidVia combines deterministic signals with AI-generated analysis. Hard evidence constrains every score, and categories without enough evidence are marked as **low confidence** instead of guessed.
A few things I learned while building it:
* 47% of the 129 sites I scanned had no JSON-LD structured data.
* Of the sites that did, 21% were missing required properties.
* 62% of content sites had no author or publication date.
* 17% blocked at least one major AI crawler in `robots.txt` (sometimes intentionally, often not).
I'd genuinely love feedback on the scoring methodology, the reports, and what signals you'd want to see added. Thanks for checking it out! 🙏
Report
Curious how you actually pull the deterministic evidence for AI readability — are you simulating how GPT-based crawlers parse the DOM, or is it more like rule-based checks against schema and semantic markup?
Report
Maker
@gocemennur76590 Good question. It's closer to deterministic rule-based extraction than trying to simulate a specific GPT parser.
For AI readability, I extract structural signals from the HTML/DOM (things like heading hierarchy, semantic markup, content structure, JSON-LD coverage, JS-render heuristics, etc.) and use those as the evidence layer. I deliberately don't try to emulate the internals of any particular model because those behaviors are proprietary and constantly evolving.
The LLM is used to synthesize the findings into recommendations, but the scores themselves are evidence-capped. For example, if a site has no JSON-LD, the structured data score is capped regardless of what the model suggests. And if there isn't enough evidence for a category, it's marked low confidence instead of guessing.
So the goal is to produce scores that are reproducible and grounded in observable signals rather than trying to reverse-engineer how a specific AI crawler parses the DOM.
Curious how you actually pull the deterministic evidence for AI readability — are you simulating how GPT-based crawlers parse the DOM, or is it more like rule-based checks against schema and semantic markup?
@gocemennur76590 Good question. It's closer to deterministic rule-based extraction than trying to simulate a specific GPT parser.
For AI readability, I extract structural signals from the HTML/DOM (things like heading hierarchy, semantic markup, content structure, JSON-LD coverage, JS-render heuristics, etc.) and use those as the evidence layer. I deliberately don't try to emulate the internals of any particular model because those behaviors are proprietary and constantly evolving.
The LLM is used to synthesize the findings into recommendations, but the scores themselves are evidence-capped. For example, if a site has no JSON-LD, the structured data score is capped regardless of what the model suggests. And if there isn't enough evidence for a category, it's marked low confidence instead of guessing.
So the goal is to produce scores that are reproducible and grounded in observable signals rather than trying to reverse-engineer how a specific AI crawler parses the DOM.