Ground.
Knowledge your AI can stand on. Ground brings your knowledge base and your working procedures into one place, puts each one through a review gate before it is marked verified, and publishes packs that every assistant pulls from. A person approves; AI never publishes.
Review gateVersions and regional overlaysRole-based visibilityPacks by API
An assistant is only as good as what it's told.
Most teams' knowledge lives in help-centre articles, old procedures and Slack threads, and some of it is out of date. Point an assistant at that and it will answer confidently from whatever it finds, right or wrong.
Trust has to be earned.
Of customers would prefer companies didn't use AI in customer service, and wrong answers are among their concerns, in a 2024 survey of 5,728 customers.
The bot's answer is your answer.
What one airline was ordered to pay in 2024 after its website chatbot gave a customer the wrong refund policy. The tribunal said the company is responsible for all the information on its website.
Answers are hard to find.
Nearly a fifth of the working week goes on looking for internal information or tracking down the colleague who knows, in McKinsey research.
If it isn't verified, it isn't in the pack.
Ground puts every procedure through a named person before any assistant can use it, and when a source changes it goes back for review instead of quietly changing an answer. Every assistant, inside the company and out, answers from the same verified set.
Figures from research on AI in customer service and on finding information at work in general, not studies of Ground. Gartner, 2024 · Moffatt v. Air Canada, 2024 · McKinsey Global Institute, 2012
One source for every AI you run.
INSIDE THE COMPANY
answers agents in Slack
scores conversations against procedure
knows your values and policies
FACING CUSTOMERS
on the website and in-app
answers, not just articles
the same answers for resellers
Verified by people. Used by every assistant.
Powered by Satori, our own AI engine.
Ground keeps your knowledge current and serves verified answers to every assistant you run, automatically.
A person stays in charge.
AI drafts, suggests and flags what has gone stale. People verify, and Ground only publishes what they verified.
A person verifies every procedure.
Nothing is marked verified by software. A named reviewer approves each version, and the procedure shows who and when.
Named on every versionStale is a prompt, not an alarm.
When a source changes, Ground marks the procedure stale and asks for a review. It never silently updates an answer an assistant relies on. Assistants keep using the last verified version until a person verifies the new one.
Review before it changesAssistants only see what's verified.
Packs are built only from verified procedures and pulled as a full replace. Drafts and stale items never reach an assistant.
Full replace, every timeVerified knowledge in a fortnight.
Connect your knowledge base.
Articles sync in read-only. Ground drafts them into a consistent shape; nothing is verified yet.
About an hourVerify the first set.
Your reviewers work the queue. Start with the procedures your assistants use most.
The first weekPublish a pack.
Point an assistant at the pack by API. From then on it answers only from what you verified.
An afternoonLet’s make
knowledge
easy to trust.
See any of our products on your own data, or tell us about a problem your team needs solving. We’ll come back with how we’d approach it.


