# AI without the jargon: what is actually happening in the chat box?

Original 5HO practice material. Fictional example. Reviewed 5 October 2026.

## Useful result

You'll finish with a plain-English explanation of why a fluent answer still needs checking.

The chat box can sound like a knowledgeable colleague. Underneath, a model generates responses from the context it receives. Tools can add search or actions, but a confident sentence is not proof that either happened.

## Set up

Set aside 15 minutes with an AI chat. You can do the exercise by pasting the example; no search, memory or connector is required.

## Practice input

Message: Our launch is pencilled in for Thursday. Context: a five-person team. Task: draft a status update. Missing: approval, final date and launch owner.

## Do the task

1. Identify which words are instructions and which are evidence.
2. Ask for an answer using only the example, then inspect every added fact.
3. Ask what additional information would be needed to confirm the launch.

## Prompt

Explain this example using the terms model, prompt, context and tool. Then write a two-sentence update from the supplied facts. Do not say you checked a calendar or searched the web.

## Refine

Separate what the notes say from what you inferred. If a tool was not used, say that plainly.

## Expected checks

Thursday remains provisional. A model is not a database of your company facts; conversation context is not guaranteed permanent memory; describing an action is not carrying it out.

## Save

Save your corrected explanation as ai-basics.txt. Use it when an answer sounds more certain than its evidence.

## Sources

- Writing clear prompts: https://help.openai.com/en/articles/10032626-prompt-engineering-best-practices-for-chatgpt