The prompt I give to GPT doesn’t fully meet my expectations. I want to provide the projects of the companies I have and get the most suitable option among the relevant fields. However, as you can imagine, sometimes the companies don’t write very good project summaries, so even if there’s nothing explicitly about AI development, it still gives me outputs like “it’s doing OCR” or similar. How can I prevent this?
1 Like
When project summaries don’t have much detail, GPT tends to fill in gaps and make a best guess. You can start by being really explicit in your prompt about not guessing. For example, instead of asking “what is this company doing?”, give GPT a fixed list of fields to choose from and tell it to answer “Unclear” if the summary doesn’t provide enough evidence.
You could also break it into two steps: first ask GPT to pull out any specific evidence (keywords, phrases, context) and then have it pick a category based only on that evidence.
To prevent GPT from making guesses, you need to make your instructions very explicit:
-
Provide a fixed list of options and tell GPT to pick only from that list.
-
Tell it to respond with “Unclear” if the project summary doesn’t provide enough evidence for a confident answer.
-
Use a two-step approach:
-
First, ask GPT to extract any keywords, phrases, or context from the project description.
-
Then, in a second step, have it choose the category strictly based on that extracted evidence.
-
This way, GPT won’t hallucinate or fill in gaps with assumptions.
It might be easier to understand the issue, if you show the prompt or a closer-to-the-real prompt version. I can’t quite understand what you’re trying to get out of the prompt vs response?