Why managing source truth in your business matters in an AI world

Why is wrong information more dangerous once AI is involved?

Because AI cannot tell which of your documents to trust, and it answers at scale.

Every business has always carried outdated documents: the 2019 refund policy in an archive folder, the price list attached to an old email, the onboarding pack nobody updated after the last restructure. It rarely mattered, because people knew which pages to ignore. The experienced staff member just knew.

AI tools do not know. A support chatbot, a staff assistant built on Copilot or ChatGPT, or an AI search engine describing your business to a prospective customer will all read whatever they can find. Faced with three versions of the same policy, they pick the one that looks most relevant, and relevant is not the same as true. Then they state it plainly, with no hint of doubt.

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

What goes wrong when your business information is wrong?

In our work with Australian businesses, the damage shows up in seven predictable ways.

1. Customers are told the wrong thing, over and over

Picture a refund policy that changed from 14 days to 30 days two years ago. The current version sits in your support wiki. The old one still sits in a shared drive archive, and a PDF copy lives in a client email. Your chatbot finds the old one first. Every customer who asks gets the old answer, and nobody inside the business sees it happen, because nobody reads every chatbot conversation.

2. You are held to what your AI said

In 2024, a Canadian tribunal ordered Air Canada to compensate a customer after its website chatbot wrongly told him he could claim a bereavement fare discount after travelling. The airline argued the correct policy was published on another page. The tribunal rejected that: the chatbot was part of the airline's website, the airline was responsible for all of it, and customers should not have to cross-check one part of a website against another.

The lesson travels. If your AI tool says it, your business said it.

3. You mislead customers without meaning to

Under the Australian Consumer Law, it does not matter whether a misleading statement came from a person, a brochure or a piece of software. The ACCC is clear that claims must be accurate and based on reasonable grounds, and that fine print does not fix an overall misleading message.

Wrong data at scale is expensive. In 2024 the Federal Court ordered Qantas to pay a $100 million penalty, plus around $20 million in payments to customers, after it kept selling seats on more than 70,000 flights it had already decided to cancel. Tickets stayed on sale for an average of 11 days after cancellation, and in some cases for up to 62 days. At its heart, that is a source truth failure: the decision was made in one place and the systems customers relied on did not reflect it.

4. Your AI tools contradict each other

Sales uses ChatGPT. Support runs a chatbot in the help desk. Operations has a Copilot shortcut someone built last year. Each one was pointed at a different set of documents. Ask the same question in three places and you get three different answers, and the customer notices before you do.

5. AI search describes your business wrongly

Customers increasingly meet your business through an AI-generated summary before they ever visit your website. Those summaries are assembled from everything the AI can find: your current pages, your old pages, PDFs still sitting on your server, directory listings and other sites that mention you. If your own sources disagree, the summary may confidently quote the wrong price, an old service area or a product you discontinued. You will not see it, and the customer may never call to check.

6. Staff stop trusting the tools and go back to asking a person

Once staff catch an internal AI assistant giving a wrong answer, they stop using it, and every question goes back to the one person who has been there twelve years and actually knows. The investment in AI is wasted, and you are left with a single point of failure. When that person is on leave it is a problem. When they leave, it is a crisis.

7. One change silently breaks ten documents

Laws change, prices change, policies change. The rule underneath your documents moves, and the documents are not told. Update the terms and conditions and forget the help article, the staff handbook and the sales deck that quote them, and you now have four sources disagreeing. In a world of AI, every one of them is still being read.

The same failure can happen at a much bigger scale. In 2024 an investigation by The Markup found that New York City's official AI chatbot for business owners was giving advice that broke the city's own laws, including on tenant discrimination and workers' tips. An official source, built to help, confidently telling people the wrong thing.

Why don't businesses notice until a customer does?

Because nothing reads all of your documents together. Every file was reasonable when someone saved it. The 2019 policy was right in 2019. The PDF was attached to an email "so it isn't lost". The wiki page is current, but nothing marks it as the one to trust.

People update the document in front of them. Nobody checks what else says the same thing. And the AI tools reading your business treat every copy as equally valid, because nothing tells them otherwise.

How do you stop wrong information reaching customers?

Not by writing more documents, and usually not by moving everything into a new system. The businesses that get this right do five things:

1. Find every copy. Know where each important fact lives, across your website, shared drives, wikis, email attachments and help desk.
2. Verify it against source. Check each version against the thing that actually governs it: your current terms, the legislation, the price book.
3. Mark what is current and what is superseded. Old copies can stay where they are, as long as they are clearly marked and point to the current version. AI tools then have nothing ambiguous to quote.
4. Map what depends on what. When the refund policy changes, you should already know which fifteen documents quote it.
5. Watch for change. Prices, policies and laws keep moving. Someone, or something, needs to notice when a source changes and raise every document it touches.

This is the problem we built TruthStack to solve. It finds what your business actually knows, wherever it is kept, verifies it against source and marks each version as current or superseded, without moving or deleting your files. When a law, policy or product changes, it flags every document that depends on it, so a single change never breaks ten answers silently.

How can you check your own business today?

Try this in fifteen minutes. Ask your own chatbot, and then ChatGPT or Google, five questions a customer would ask: your refund policy, your prices, your opening hours, what you do not offer, and who to contact about a complaint. Compare each answer with what your current terms actually say.

If even one answer is wrong, the cause is almost always in your own sources. To see where your website disagrees with itself, run a free Consistency Audit: it reads every page of your site and flags the places where your prices, hours, policies or claims contradict each other.