ChatGPT making up information?

Why AI invents things

ChatGPT doesn't have access to truth. It has access to patterns of text. When you ask something it doesn't know, it doesn't say "I don't know." It assembles the most likely answer based on what it learned — even if it's wrong.

This has a name: hallucination. The AI hallucinates facts, numbers, references, and even books that don't exist.

The anchor method

The most reliable fix is to give the AI a source document before asking your question. You paste the text and say: "answer using ONLY what's in this document."

Practical example:

1. Paste the company financial report into the chat
2. Then ask: "Which quarter had the highest growth, according to this report?"
3. Ask: "Quote the exact number from the document."

With the anchor, the AI stops guessing and starts extracting information that's already there.

Three anchors that work

• Documents: reports, contracts, presentations — paste the text before asking
• Data: spreadsheets, tables, lists — AI reads numbers better than free text
• Examples: "I want it in this format" and paste a real example

What to do when you have no document

If you don't have a source to anchor with, ask the AI to flag what's uncertain:

"If you're not sure about any data point, mark it with [?] and I'll verify it later."

This doesn't eliminate hallucinations, but it shows you exactly where to double-check.

The simple rule

For facts: always anchor. For opinion and creativity: let the AI fly. Knowing when to use each mode is what separates beginners from confident users.