When a Gen AI-based bot doesn't respond correctly or provides incomplete answers (e.g., without including the technical data sheet or corresponding URL), one of the most frequent causes is an improper configuration of the training files or a non-specific prompt.

Before uploading training documents:

✅ Only upload relevant files: avoid technical data sheets of products that are not mapped or are not part of the expected responses.

✅ Check that all current models are included in the file.

✅ Unify the technical data sheets into a single document, instead of dispersing them across multiple PDFs or separate files.

✅ Verify that the model name matches exactly the one used in the bot mapping or the query source.

✅ Avoid contradictory or outdated training.

Include clear examples and restrictions to guide the response.

For example, in the [Tasks] section of the prompt you can use something like:

Uploading too many documents, especially if they are not correctly structured or contain models that are not part of the conversational flow, can:

Explicitly indicate what information the model should consult (e.g., official database, FAQs, validated documents).

Example: "Only consult the official database to answer this question."

Specify what the model should do when it doesn't have information available: respond with an error message, ask the user to consult with an agent, or simply indicate that it doesn't know.

Example: "If there is no confirmed information, respond: 'I don't have that information available, would you like me to connect you with an agent?'"

Ask it to avoid generating speculative or invented answers. Example: "Do not make assumptions or invent data."

Incorporate examples of correct answers and also of how to respond if the information is not available. This helps the model understand the limits and the expected format.

Add phrases that remind the model of the priority of giving accurate and reliable answers.