Document and reference research
Crawl4AI was already proving useful for document searching and deeper reference research. We use it for document research now.
Case study · Crawl4AI · Trial complete, 31 October 2026
From missing prices to better research.
We found missing prices, told the founder, and retested his fixes on the same pages until they matched.
The routes we looked at
Three ways we put it to work.
Exact pricing from pricing pages
We ran Crawl4AI and Firecrawl side by side on the same pricing pages. After the fixes, it performed on a par with Firecrawl on the pages we checked.
Work with the founder
We reported the gaps from real use. He suggested changes, we retested them on the same pages and shared the results.
The results
On a par with Firecrawl, on the pages we checked.
11 HubSpot plan prices
All 11 plan prices captured in our retests.
11 Semrush figures
All 11 price figures captured in our checks.
Four UK pages match
By 21 October all four UK pages we tested matched Firecrawl.
The Make.com slip, fixed
A later default run found Make.com had slipped again. The founder fixed it, and our fresh default retest returned all four plan prices.
Our own paired runs. Results vary by page, region and setup.
Our own research route, tested on real pricing pages
The gap
We build agent workflows for our own work and wholesale services for our clients. We were evaluating Crawl4AI as one of our intelligence routes: a pay-as-you-go service that might make our research workflows more efficient and help us manage delivery costs.
It was already proving useful for document searching and deeper reference research. But our early comparisons with Firecrawl found that exact pricing extraction was less reliable. Some plan prices were missing.
The build
We planned a split. Document searches and deeper research would go through Crawl4AI. Number-based research, such as pulling exact prices from pricing pages, would stay with Firecrawl, because our first comparisons showed it was more accurate.
About a week later, the Crawl4AI founder got in touch to ask how we were getting on. We told him about the split and the gaps we had found. He came back with suggestions. We tried them, compared the results against the same pages, and shared what we found.
How we worked together
Paired runs on the same pages
Every test ran Crawl4AI and Firecrawl side by side on a defined set of pricing pages, so we compared like with like.
Report from real use
We told the founder exactly where prices went missing, from our own work, not from a synthetic test.
Retest each suggestion
We tried every change he suggested against the same pages and shared the results. There was some back and forth.
Keep validating
Results vary by page, region and setup, so we keep checking the route against the needs of each workflow.
What happened
Within roughly a day of the founder getting involved, the pricing results had improved substantially, with more fine-tuning the following day.
Crawl4AI captured all 11 HubSpot plan prices and the 11 Semrush price figures in our retests. By 21 October all four UK pages we tested matched Firecrawl.
A later default run found that Make.com had slipped again. After another exchange the founder fixed it, and our fresh default retest returned all four plan prices.
After the fixes and follow-up checks, Crawl4AI performed on a par with Firecrawl for the pricing pages we tested. We are looking at it for document and reference research, and it is now a credible option for the pricing-page research we checked too.
The founder has offered to mention us in the changelog for helping find and fix the bug. Thanks to the Crawl4AI team for listening, working through the problem with us, and improving it so quickly.
This is one page of evidence from one defined set of pricing pages. Results can vary by page, region and fetch configuration.
What this means for you
We test the tools we use, on real work.
If you want to know what a build or a training route costs, every price is published.
See the rate cardAll prices are on the rate card. Transparent pricing →