Hi,
Well it depends of the source, for expert reviews (TechRadar, RTINGS, PCMag, etc.), I use Brave Search to discover relevant sources for each product, then extract the content from those pages programmatically. Same for YouTube, where I pull transcripts with groq-whisper. For Reddit, I use the Reddit API to discovers and extracts relevant threads and comments automatically too.
Once all the sources are collected, the pipeline analyzes each one (sentiment, ratings, bias etc.) using moonshot and claude, then cross-references everything to find where reviewers agree and disagree. That consensus is what becomes the scores and summaries you see on the site.
So it’s fully automated, from discovery to the final page. I just review the output before publishing to make sure all make sense.