Using Claude Code for research: what worked and what didn’t

The job

The question was simple: what do Submagic, Captions, OpusClip, Descript and CapCut charge, and what do you get for it? Claude Code searched the web, opened pricing pages and comparison sites, and put what it found into a table, with the link for every number.

What worked

It’s fast. Pulling five pricing pages and the coverage around them is an afternoon by hand and a few minutes this way. It also keeps sources next to claims, so you can click through and check.

It’s good at structure, too. We asked for a table with the plan, the monthly price, the yearly price and what’s included, and every competitor came back in the same shape.

What to double-check

Some prices came from comparison websites, not the company’s own page, and companies change prices often. One product listed iPhone app prices, which are different from the website. Ask where each number came from, and treat any figure from a third party as a lead.

The bigger lesson came from a different task. We asked for a summary of the same open-source README twice. The two summaries disagreed on specific numbers and on one piece of font advice. Both sounded certain. When a detail matters, fetch the original file and read it.

The setup that works

Ask for sources with every number. Ask for the date it was checked. Ask what it couldn’t confirm. For anything you’ll publish, read the original page yourself. Treat what two separate runs agree on as safer, and what they disagree on as the list of things to check.

How we used the answers

We didn’t copy a competitor’s price. We used the research to see where the gap was: tools that charge for exports, and tools priced for agencies. Then we wrote our own plans around the person we’re building for. The research told us what was true. The decision was ours.

Questions

More guides