Browse AI: our review

Browse AI starts with a simple action: you browse the page you want to track, click the fields that interest you — a price, a date, a title — and the tool turns those clicks into a "robot" that repeats the same extraction on its own, at the time or frequency you choose.

What it actually does

The robot memorises the clicked locations as selectors on the page. Once launched, it replays the path, reads the same areas and returns a structured table, exportable or queryable through the API. The piece that justifies the tool's name isn't the extraction itself — mechanical, based on the saved selectors — but the automatic detection of layout changes: when a site redesigns itself, the so-called robot adjusts instead of returning an empty page. That's a useful distinction to know before buying: the AI keeps the tool working, it doesn't understand the content in your place.

Why the question is legal first, not technical

Where Browse AI stands clearly apart from its competitors is on personal data: its own terms forbid using the service to "access, collect or store" personal information. A contact database, a list of named email addresses, an export of profiles: all of that falls within the scope of this ban, whether the data is public or not. That's a firmer stance than most tools in the sector take, which generally just shift responsibility onto the user without banning the use itself.

That doesn't exempt anyone from any other check. The terms explicitly make the user responsible for not violating the targeted site's terms of use, and that site's robots.txt still applies no matter how easy the point-and-click process is. The fact that a robot can be built in three minutes without writing a line of code says nothing about the right to collect the targeted page: that question remains open, and it comes before clicking, not after.

Where it falls short

On model training, the privacy policy only takes a position on one particular case — data obtained through Google Workspace APIs isn't used to train a generative model — with no general commitment beyond that. Data hosting is only partly documented: the privacy policy mentions access from Canada and the United States without specifying where the servers themselves are located, which rules out any European guarantee. The cost per run also varies with how hard the targeted site is to scrape — from two to ten credits on the most protected sites — which makes estimating a bill unreliable until you've tested the actual target.

Our recommendation

Browse AI suits a team with no developer that wants to monitor public prices or run recurring content watches, within the boundary the vendor itself draws: no personal data. For a use case close to that line — a contact list, a professional directory — it's better to check beforehand than to test the limit, especially in a real estate or legal sector where the origin of the data matters as much as its accuracy. See also our automation service to scope a collection project before choosing the tool.