Shipped an interactive AI loop to fix landing pages (Need feedback!)
Hey everyone,
I just pushed a major upgrade live to RoastMyLanding to solve a big issue: users get a website audit, but then they get stuck trying to actually implement the changes.
To fix this, I built RoastBot—a live chat assistant right inside the report that knows the exact context of your audit so you can ask it how to rewrite copy or fix design bugs on your specific tech stack.
I just wrote a deep dive on how I built this product loop and updated our $19 premium pricing.
Check out the full breakdown and share your thoughts here: https://www.indiehackers.com/post/how-i-turned-my-static-audit-tool-into-an-interactive-ai-loop-meet-roastbot-3965baba7d
Trending on Indie Hackers
This is a cool concept! I love the idea of an interactive loop for landing page fixes.
One thing I've noticed with these AI-driven audit tools (I've been playing around with SiteScoper lately) is that the 'brutal honesty' part is what founders actually value most. Most generic AI just says 'improve your SEO,' but the real value is in the 'Your hero headline is generic — visitors bounce in 4s' type of feedback.
How are you handling the 'emotional' side of the audit? E.g., does the page feel 'trustworthy' or 'premium'? That's usually the hardest part for AI to nail compared to technical SEO.
Would love to see how your tool compares to a deep-crawl auditor like SiteScoper on a complex SaaS page.
You hit the nail on the head. Most AI auditors just scrape metadata and say "add alt text to images." It’s useless because a page can have perfect technical SEO and still look like a scam.
Because my background is in graphic design, I trained the logic loop to look at structural trust signals—things like visual hierarchy, cognitive load (clutter), and messaging alignment. If a hero section has three different font sizes and a generic "Optimize your workflow" headline, it triggers a flag for low trust and high bounce probability.
It’s definitely a challenge making AI understand if a page feels "premium," but the trick is teaching it to spot intentionality in the layout. I’d actually love to run a complex SaaS page through it and stack it against SiteScoper to see where the blind spots are. Drop a link if you have a specific page in mind!