How do I make sure Claude gets up-to-date information about my site?

Up front: Anthropic pulls from a great many sources and any of this is liable to change. Take everything below as “this worked for me in mid-2026.”

Claude is both trained on the web and performs searches on demand. I’m talking about the on-demand searches – the ones it runs when you ask it about something current.

A bit of context. pyStrich is a Python barcode library I’ve maintained since 2015, when I ported it from the Python 2-only huBarcode. It wasn’t just a port, some features were expanded, but by 2016 it had hit the stable, useful, leave-it-alone phase. The only changes since then have been the bare minimum to keep it working through Python version bumps, dependency churn, change in CI etcetera. A couple of months ago, I decided to spend some time improving it and getting it to a more polished, complete and credible state: comprehensive docs; a fresh look at the feature set and old “won’t fix” calls – the lot. After the better part of a decade of quiet, there’s been a recent flurry of activity around the project.

In 2026, where so much day-to-day development happens via agents, the audience for project work isn’t only human. If a developer (or their agent) asks Claude about pyStrich, I’d quite like Claude to know what’s true today rather than answering from a stale page or training data.

So when I noticed Claude was answering with information from the hiatus, I went looking for why.

I started by trying the same questions on Gemini and ChatGPT. Both had the current information. I’d already made sure the docs were in Google Search Console and Bing Webmaster Tools, so the main search engines were covered. That narrowed the gap to Claude specifically.

Which raises an obvious question: where does Claude get its information when it searches the web? It appears to be Brave. I confirmed this by comparing Claude’s citations and answers with Brave’s results for the same queries – the matches were too similar for it to be a coincidence.

The stale information was weeks or more out of date across several pyStrich-related URLs. Not a huge surprise: pyStrich is niche, doesn’t get much traffic, and the relevant pages had sat unchanged for years before this push.

Brave Search builds its index partly through the Web Discovery Project, which uses anonymised browsing data from Brave users who’ve opted in. That’s presumably supplemented by more conventional crawling, but it does mean low-traffic sites without much of a Brave-using audience can lag behind.

Brave Search has no Search Console equivalent. What it does have is a URL submission form. I filled in the key pages that were out of date or missing, and for good measure browsed around them in Brave.

A few hours later Brave’s results were current. I asked Claude the same question I’d asked before and the answer was clearly drawing from the fresh pages – titles and quoted content gave it away.

One side note: Brave’s results don’t have quite the same authoritative-source weighting we’ve come to take for granted from Google or Bing, and that seemed to mildly confuse Claude in places. Having some diversity in how a niche topic is described – without literally duplicating content – may help Claude pull the right thread.