Leave no trace
Good trampers follow the track, respect the land and leave it as they found it. We treat vendors' websites the same way: we read only what they make public, follow the rules they set, and keep a fair record of what we found.
SNOW GENTIANGentianella bellidifolia
-
1
We ask before we enter
Every website can publish a robots.txt file saying which pages may be read. We check it before every page. If a site says no, we don't read that page, and we never try to get around a block.
-
2
We follow their wishes for AI tools
Some sites publish an llms.txt file that tells AI tools how their content may be used. Where a site has one, we will read it and follow it. Being added.
-
3
We stay on the track
We read only public pages about the product: what it does, who it is for, pricing and customer stories. We don't sign in, fill in forms, download files or go behind paywalls.
-
4
We walk, we don't run
We read one page at a time, with at least three seconds between pages, and slow down further if a site asks us to. We say who we are on every visit, with a way to contact us.
-
5
We only say what the vendor says
Profiles rebuilt under these rules are written only from the pages we read, and list the pages they came from. If it isn't on the vendor's own pages, it isn't in a rebuilt profile: missing detail shows as "Not recorded yet". The exceptions are G2 ratings, which come from G2's public pages, and details the vendor's site doesn't give, such as founding year, which can come from another public source. An AI web search finds these, and each must come with its source and exact words. Some older profiles still use earlier data; the line below says how many.
How a vendor gets into the directory
-
A
Find
We search Google for AI tools at each stage of the customer journey, three searches per stage, and look at the first results. A fixed list of blocked sites (such as social networks, review sites and directories) and simple rules about articles and job pages take out results that aren't vendors. Vendors and readers can also suggest a tool.
-
B
Read
Our own crawler reads a vendor's public product pages under the five rules above, and saves each page it reads successfully, exactly as it was read, with the date.
-
C
Check it's a vendor
A small AI model running on our own computer reads the profile text built from the vendor's pages and must quote the exact words that show what the company sells and who buys it. Plain code makes the decision from that quote, not the model. Agencies, publishers and communities are left out; anything unclear is held for a person to review.
-
D
Write the profile
A rebuilt profile is made only from the saved pages. Where a quote is used, it is checked word for word against the saved page. New company details, such as headquarters and founding year, come from the vendor's own site or another public source, each with its source and exact words, and are never guessed. Details an older profile already had are kept until a person corrects them.
-
E
Approve the release
Each new version of the directory is prepared as a release that is kept, so it can be checked or rolled back. So far every release has been reviewed by a person (what is new, changed or removed) before it went live.
-
F
Refresh and correct
Vendor sites can be read again with the same crawler, so profiles can follow what vendors publish. Corrections and removals can be requested at any time, and they are always free.
Where we are now. Every profile is being rebuilt under these rules from a fresh read of each vendor's site. Still being added: following llms.txt, and linking each fact to the page it came from.
Are you a vendor?
If something on your profile is wrong, or you would rather not be listed, tell us. We'll check it against your pages and act on it.
directory@successbycs.comTo see what has changed and when, read our releases.
