In 2022, a Canadian man named Jake Moffatt asked Air Canada’s website chatbot about bereavement fares after his grandmother died. The chatbot told him he could buy a full-price ticket and apply for the discount afterwards, provided he did so within 90 days of travelling. Air Canada later refused the claim because its official policy said the reduced fare had to be requested before the journey.
Moffatt took the airline to Canada’s Civil Resolution Tribunal and won. The tribunal rejected Air Canada’s suggestion that the chatbot was somehow separate from the company and responsible for its own answers. Customers, it said, should not have to work out which part of a company’s website they can trust. The case is often used as an example of AI getting something wrong, although the underlying problem was more familiar: Air Canada had conflicting information in different places. The chatbot found one version while the company relied on another, and nobody had made sure the two matched.
That same problem already exists in plenty of smaller businesses. Old pricing sheets, duplicate policies, superseded contracts and several versions of the same spreadsheet can sit side by side for years. Once an AI tool has access to those files, any one of them can shape the answer it gives. It does not know which document the finance director trusts or which policy was replaced six months ago. Unless the difference is clear in the way the files are stored, an old document can look every bit as useful as the current one.
Why This Persists
Most file clutter has no single cause. It could be a project folder that was never cleared out, a document downloaded and edited locally, or an attachment sent around because it was quicker than finding the shared copy. Someone saves another version to their desktop, someone else changes the filename, and before long the business has differently named final versions sitting in different locations.
Deleting anything then feels risky because nobody is certain what may be needed later. Keeping everything feels safer, even though it makes finding the right information much harder. Before generative AI became part of everyday business software, the main cost was wasted time. Someone might spend 20 minutes hunting for the latest proposal template, but they would often spot an old date, question an unfamiliar price or ask a colleague before relying on it. AI tools do not have that background knowledge. They search the information available to the user and generate a response from what they find, so an outdated document may still be treated as a valid source. That is why Microsoft’s guidance for preparing organisations for Microsoft 365 Copilot recommends reviewing sharing permissions and identifying stale or overshared content before rollout. The less outdated or unnecessary information your AI can access, the more trustworthy its responses are likely to be.
AI Still Depends on Good Housekeeping
An AI policy is still worthwhile. Staff need to know what they can enter into a tool, which answers must be checked and when human approval is required. The policy cannot fix unreliable source material, though. If an outdated price list or superseded contract remains in an accessible folder, telling staff not to use old information will not stop AI from finding it. Current documents need to be easy to identify, older versions need to be archived properly, and important files need a clear owner. Otherwise, both people and AI are expected to work from information that nobody fully trusts.
Cleaning this up also supports an obligation businesses already have under UK GDPR. The storage limitation principle requires organisations not to keep personal data for longer than necessary. Old HR records, customer information and superseded contracts may all contain personal data, so they cannot simply be left untouched indefinitely.
The ICO expects retention to be reviewed regularly, rather than dealt with through a one-off clear-out every few years. In practice, that means knowing who owns important information, how long it should be kept and what should happen when the retention period ends. Regular reviews, proper archiving and the removal of information that no longer needs to be kept will also improve the quality of AI results. With fewer outdated or duplicate files in circulation, there is less chance of a tool pulling an answer from the wrong source.
Where This Starts
A sensible place to begin is with the folders staff use every day, including the main shared drive, SharePoint, Teams and personal storage. The initial review should identify where current documents sit alongside older copies, where naming conventions have broken down and where important files depend on one person knowing where they are.
Searches for terms such as final, copy, old, v2 and FINAL_v7_FINAL usually reveal plenty. So does checking whether key documents exist only in an inbox, on a laptop or inside a former employee’s personal folder. The aim is not to reorganise everything at once, but to establish which version is the master copy, who owns it and when it should be reviewed. Older versions can be moved into a clearly marked archive where there is a genuine reason to retain them, while files that no longer serve a business or legal purpose should be removed.
Permissions need attention as part of the same review. Restricting access to a current contract means very little if an unrestricted copy is sitting three folders away. Once duplicates have been dealt with and ownership is clearer, it becomes much easier to decide who should see what and which information an AI tool should be allowed to use.
This is the same kind of audit we run with clients before changing cloud storage or Microsoft 365 configuration, and it nearly always uncovers more mess than expected. That is normal. Most businesses are starting from years of shortcuts, staff changes and temporary fixes, not from a perfectly organised system.
Once the structure is clearer, everything else becomes easier to manage. Access rules make more sense, retention decisions are easier to defend, and AI tools have a better chance of finding the right information. Staff also spend less time searching, checking filenames and asking colleagues which version to use.
AI can make finding information faster, drafting documents easier and routine work more efficient. It can also repeat old prices, quote outdated policies and surface information that should have been archived years ago. The quality of the answer depends heavily on the quality of the material behind it.
If you’re planning to introduce AI into your business, start by making sure it can trust the information it’s using. Download our AI Policy Starter Kit or book a 30-minute AI readiness chat, and we’ll tell you what we would look at first.
