How seriously you use the host's technology is an explicit scoring factor.
Look at the Bright Data hackathon winners: multi-product builds dominate, including one that used five of the company's products. At ADK and GitLab events, deep platform usage was effectively table stakes. Projects that ping one sponsor API superficially do not place.
This is rational: companies run hackathons to showcase their technology, and their engineers sit on the jury. What they praise is the project that shows them a use of their own product they hadn't imagined.
Takeaway: after registering, read the sponsor's docs first and aim for either multi-feature combination or an unexpected use case. Browse past winners of the same event (searchable on this site) to calibrate how deep the bar is.
Across five years of winners, polish and reliability beat flashy ideas.
Recent winners are built as teams of specialized agents, not single chatbots.
Winning projects solve problems someone is suffering from today.