I Built roastme.gg: Pay to Get Roasted by AI
This title could be clearer and more informative.Try out Clickbait Shieldfor free (5 uses left this month).
An indie hacker recounts building roastme.gg, a pay-to-get-roasted-by-AI site built with Next.js, Stripe Checkout, Claude, and Neon Postgres, shipped in a few days. The bulk of the piece is a live experiment testing Generative Engine Optimization (GEO): using GPT-5's web_search tool to check whether the brand new site got cited by ChatGPT for generic queries before and after adding static FAQ content and structured data (FAQPage, Service/Offer, CreativeWork schema). Before fixes, the site appeared only for its own name; after adding a plain-text FAQ page with real pricing and stats, ChatGPT began pulling accurate details from that page for the branded query, though generic category queries still didn't surface the site, attributed to indexing lag and lack of third-party mentions. The author also notes llms.txt showed zero measurable citation benefit in a study of 300,000+ domains, and that Google killed FAQ rich results in May.
Table of contents
Where the Idea Came FromThe BuildTuning the Roast ItselfLaunchThe Bug That Only Showed Up After Real Users Touched ItThe Real Experiment: Does Any of This Actually Get You Cited by AIWhat I Actually FixedWhat Moved and What Didn'tWhat I'd Tell Someone Else Shipping a Small, Fun ProductFrequently AskedQuestions this post answers
Does adding an llms.txt file actually help a website get cited by AI search engines like ChatGPT?
No measurable citation advantage has been found. Multiple independent studies, including one covering over 300,000 domains, found zero measurable citation benefit from having an llms.txt or llms-full.txt file, and no major AI company has publicly committed to reading it in production. Time is better spent on visible, fact-dense page content instead. Developers weighing GEO tactics can find grounded, current takes on daily.dev before spending time on unproven fixes.
Does adding structured FAQ schema still help a page get a rich result in Google search?
No, Google removed FAQ rich results from search in May, so FAQPage JSON-LD no longer wins a snippet there. It still has value elsewhere, however: Bing and Perplexity parse FAQPage markup to understand Q&A content, and ChatGPT's search backend runs on Bing, so structured data still complements visible text for AI answer engines even though it no longer benefits Google's own SERP features. Anyone tuning a site for both search and AI citations can track shifts like this by following structured-data discussions on daily.dev.
How long does it take for ChatGPT's web search to pick up newly published content on a brand new website?
ChatGPT's search index, which runs on Bing, typically takes one to three weeks to pick up genuinely new content on a fresh domain. Citation share for competitive, generic queries also depends heavily on third-party mentions from other sites, not just the page's own copy, making cold outreach a slower but necessary lever beyond on-page fixes. Founders tracking how quickly AI search picks up new launches can follow real-world GEO experiments on daily.dev.