Verified knowledge for the age of AI

Your AI found three versions of your refund policy. Nothing told it which one was true.

TruthStack finds what your business actually knows, wherever it's kept, and verifies it against source. When new laws, new policies, new products, or the way customers actually experience your business change, TruthStack updates everywhere that knowledge is used.

A single support query · traced

One policy. Three copies. The oldest one answered.

Where the refund policy lives right now
SharePoint › Archive
Refund policy 2019copy 12019
← your AI answered from this one
Outlook · email attachment
Refund Policy (final)copy 22023
sent to a client, saved "so it isn't lost"
Confluence › Support space
refund-policycopy 32 mths
newest and best — nothing marks it as such
Quoting the copies
Help centre articlequotes 1
Staff handbook §4quotes 1
Client onboarding packquotes 2
Plus the current rule as your Client Services lead would actually state it — which is in her head, and nowhere else.
What went out under your name
"What is our refund policy?"
Refunds within 14 days, at the manager's discretion.
Your published terms have said 30 days since 2023.
Sent to 41 customers this month chat · email · help page
Times anything checked it against your terms: 0
Every file above was reasonable when someone saved it. Nothing has read them together since.
3 COPIES · 3 SYSTEMS · 0 CHECKS
01
The problem

One outdated document. Ten broken answers. Nobody knows until a customer does.

That's not a one-off. Every business has lived with outdated documents for years — a person just knows which pages to distrust.

AI doesn't. It reads what's written and repeats it, confidently, to whoever asked.

One bad answer used to be one bad answer: a single customer, an awkward correction, sorted in a day. Now it's every customer who asks, answered the same wrong way, instantly.

Which bot said that?

Sales runs on ChatGPT. Support runs a HubSpot bot. Ops has a Copilot shortcut someone built last year. Ask the same question in three places and get three different answers.

Nobody told the other ten

Legal updates the T&Cs on a Tuesday. The FAQ, the refund policy and the bot script that quotes them don't get the memo, and support is still promising the old terms in March.

Sharon already knows this

Every new hire, every bot, and every "quick question" in Slack ends up back at the one person who's been there twelve years. When she's on leave, that's a problem. When she leaves, it's a crisis.

In Australia this carries weight beyond an awkward correction. Consumer law, privacy, employment terms and industry codes reach almost every business of this size, and the rules underneath your documents change without telling the documents.

02
How it works

From first call to knowledge your AI can trust, in three steps.

01
DIAGNOSTIC CALL · FREE · 30 MINUTES

We map your knowledge sprawl: how many systems, how many contradictions, how much is trapped in one person's head.

02
DISCOVERY SPRINT · 2 TO 6 WEEKS · FROM A$15,000

We find and verify your knowledge wherever it already lives, and mark it: current or superseded, what it depends on, what depends on it — so a single change never breaks something silently. You get a knowledge systemisation report: recommended actions, risks and solutions.

03
MAINTAIN AND RENDER · ONGOING

Agents check currency, flag expiring content and watch for legislative change. Verified knowledge is then rendered into help pages, guides and structured content in your own brand, and feeds your customer bots, staff assistants and internal search.

The sprint · index, verify, mark

Nothing moves. Everything gets marked.

Before · same files, no marks
Refund policy 2019SharePoint › Archive
Refund Policy (final)Outlook attachment
refund-policyConfluence › Support
To your AI these three are equals — three plausible documents, no way to tell which one to trust. So it picks by relevance, and relevance is not truth.
After · same files, same places — now marked
refund-policystill in Confluence
Current · verified v. terms 8.2
Refund policy 2019still in SharePoint
Superseded → links to current
Refund Policy (final)still in Outlook
Superseded → links to current
What each mark holds, in a form AI reads firstCurrent or superseded · the source it was verified against · what replaces it · when it expires · who owns it
Your AI reads the marks before the files. Ask the same question now and it answers from the Confluence page — because the marks leave it nothing else to cite.
What cannot be verified is triaged to a person
The Dewey Decimal system for AI — we do not move your books, we catalogue them.
NOTHING MOVED · NOTHING DELETED · EVERYTHING MARKED
After go-live · the standing watch

One clause changes. We raise everything it touches.

1 · The trigger
Consumer law guidance amended
clause 19.4(b) · effective 2023
The same change behind the query above — this time every dependent document gets caught, not just one.
→
2 · The ripple
15documents flagged
T&Cs · help article · onboarding pack · email templates + 11 more, across your systems
→
3 · The change, drafted
Old 14 days, at the manager's discretion
New 30 days from delivery, no discretion
→
4 · A person decides
Accepted by your author
One approval updates all fifteen — and every rendering layer they feed. A human decides; the system propagates.
This is what you are paying for after the sprint ends.
WATCH · RAISE · APPROVE · PROPAGATE

Costs are agreed upfront, in writing, before anything is built. Nothing renews without you knowing.

03
Straight answers

“We already have a wiki or Confluence. Isn't that the same thing?”

A wiki stores documents. It doesn't verify them against source, tell you when one is out of date, or know that changing one page breaks the accuracy of ten others. That dependency mapping and ongoing verification is the part most tools skip, and it's the part that actually prevents wrong answers.

“Will this replace our staff?”

No. TruthStack exists so your staff and your AI tools are working from the same verified answer, not to replace the people who hold that expertise. Fewer hours spent re-litigating what the correct answer is, and more confidence in it once it's found.

“How do you handle our data and security?”

Your documents never leave your own systems — we verify and mark them in place. The marks themselves (what's current, what it depends on, who owns it) are access-controlled and yours. We talk through your specific security and compliance requirements on the diagnostic call, before any commercial commitment.

“How long until we see value?”

The sprint alone, two to six weeks depending on scope, typically surfaces contradictions and outdated content your team didn't know existed. That's often valuable on its own, before any retainer begins.

“What does the ongoing watch cost, after the sprint?”

It's scoped and quoted after the sprint, once we know the size of your knowledge base and how often it actually changes, not before. Whatever that number is, it's agreed in writing and it doesn't move on you: see the point above about nothing renewing without you knowing.

Find out what your documents would tell an AI tool today.

We'll help you align your documents and get them ready for AI.

Free Truth Audit
Every business's knowledge goes stale somewhere, a T&Cs page, an FAQ, a help centre article. Send us one public URL and a work email, and we'll send back up to five specific findings: the exact passage, what it contradicts or gets wrong, and roughly when it went stale.