We tested five AI research tools. Only two saved real time.
A controlled field test reveals where research agents help—and where they create more checking work.
REPORT
We gave five research assistants the same brief: map a fast-moving software category, identify the strongest primary sources and produce a decision-ready summary in thirty minutes.
Every product produced a polished answer. That was the least useful way to judge them. We scored source quality, traceability, omitted evidence and the time required to verify each claim.
Two tools consistently led us back to primary material and made uncertainty visible. The others were fast, but their confident summaries created a second research task: checking what had been flattened or overstated.
Signal & Syntax will continue to update this report as products, pricing and practical evidence change.