Short answer: Colleagues get different answers from AI because the tools read different material: personal GPTs, old copies or several versions of the same document. The fix is not better prompts but one shared source where every piece of knowledge has an owner, a date and an approved version.
A common example
Two salespeople prepare quotes for the same customer. One asks their Custom GPT about the discount rules and hears that 15% is fine. The other asks Copilot, which finds a draft from last spring where the limit is 10%. Both answers sound confident. The customer receives two quotes with different terms, and nobody in the company knows which is right.
Neither AI tool made a mistake. Each answered correctly based on what it had. The problem is that what they had was different.
Four reasons the same question gets different answers
- Different material. Every GPT, project and Gem has its own copy of the documents, uploaded at different times.
- Several versions of the same document. "Price_list_2025_final" and "Price_list_2025_final_v2" sit side by side, and the AI cannot know which applies.
- Unwritten knowledge is interpreted differently. Exceptions and practices that were never written down get filled in differently by each person explaining them to the AI.
- The AI fills gaps. When the answer is missing from the material, the model guesses, and the guess differs every time.
Why it stays invisible until it is too late
A wrong AI answer looks exactly like a right one. It is well written, confident and without caveats. Because colleagues rarely compare their answers, the contradiction is only discovered when it reaches a customer, a contract negotiation or an audit.
The problem is not that the AI guesses. The problem is that nobody decided what it should work from.
Who decides what is true?
It cannot be the AI, and it cannot be whoever happens to ask. It has to be a person responsible for the area. The key is to set ownership per knowledge area, not per document:
| Knowledge area | Owner | Example questions |
|---|---|---|
| Pricing and discounts | Head of sales | Which price list applies? How much discount can I give? |
| Suppliers and logistics | Operations | Which supplier handles express deliveries? |
| People and benefits | HR lead | How does the wellness allowance work? |
| Tone of voice and brand | Head of marketing | How do we write to customers? |
The owner does not have to write everything. Their job is to approve what applies and to decide when the material says different things.
Three rules for one shared truth
- One approved version per question. Drafts and history can exist, but only one version is used as the answer.
- Every piece of knowledge has an owner and a date. Then it is visible when something is old and who should update it.
- Contradictions are settled by the owner, not the AI. When two sources disagree, it is flagged and the owner chooses.
How to handle conflicting documents
- Find the contradiction. Start with the questions that most often get different answers, or let a system flag when documents disagree.
- Let the owner choose. Whoever owns the area decides which version applies.
- Mark the old one as replaced. Do not delete the history, but make sure it is not used as an answer.
- Update every tool at once. If knowledge sits in one shared source, changing it in one place is enough.
How Company Brain does it
A Company Brain is built on these rules. It reads from SharePoint, Google Drive and uploaded documents, flags when material disagrees and lets an accountable person choose what applies. Every answer shows source, date and owner, and the same approved answer is available in Copilot, ChatGPT and Claude. What is not written down anywhere is captured by an interview agent and approved the same way.
Want to understand why the built-in tools are not enough? Read the comparison of Projects, Custom GPTs, Gems and Copilot. For knowledge management more broadly, see What is knowledge management?
Get started in one afternoon
- List the twenty questions your team asks most often.
- Ask them in the AI tools you use and compare the answers.
- Mark every question where the answers differ.
- Name an owner for each area and let them decide what applies.
That exercise alone usually shows where the biggest risks are.
Frequently asked questions
Why does ChatGPT give different answers to the same question?
Partly because language models phrase things slightly differently each time, and, more importantly, because different people give the AI different material. The latter is what leads to real contradictions.
Do better prompts help?
Better prompts give clearer answers, but they do not fix material that differs. Two perfect prompts against two different price lists still give two different answers.
Who should own company knowledge?
Whoever already owns the area in the business, for example the head of sales for pricing and HR for people questions. Ownership is set per area, not per document.
Can AI decide which version applies?
AI can find and flag contradictions, but it should not settle them. The decision about what applies should be made by a person responsible for the area.