Think of the AI as a freelancer
This is the #1 advice we give users when they interact with our system.
The more precise your ask, the better the outcome. For instance, let's say you want to test if structured data can be captured from unstructured emails. A simple prompt to test this could look like this:
As you can see, the system does precisely what we ask it to do, since we are clear in our ask.
Prompting is a bit like using Google; it needs some exercise
Depending on the configuration of your system, your message is either directly processed as raw input into the AI system, or it is transformed to better fit e.g. a search query.
As an example, if you first ask "How old is Cristiano Ronaldo", and get a response from the system, you might follow-up with "Where does he currently play?".
If your system is configured to rephrase questions into search queries based on the current conversation, it will transform this question of yours into "Where does Cristiano Ronaldo currently play?". It's easy to find a result for this in the internet. On the other hand, if this configuration isn't set up, the search will have "Where does he play" as a query, which leaves open questions to the search engine: who is "he"?
You can always see the processing steps the system is doing under the hood in the progress bar:
If you click on the text (here: "Understanding the question..."), it will open the logs:
Questions are handled differently than instructions
In most cases, your solution is configured to enrich the message you send with psychological characteristics. For instance, we automatically classify if a message is information-seeking, in which case we trigger the internal search. If, on the other hand, you rather tell the system to summarize a text, it will not search in the internal documents, but just looks into the messages of your current conversation.