When Data Is Missing

Extraction prompts often ask for something the text simply does not contain. A small model then fills the gap somehow: an empty string, a placeholder, or worse, an invented value. A program reading the JSON cannot tell any of those from real data.

Extract the hotel name, city and nightly price from this review: "Loved our stay at Hotel Mira, great breakfast!" Return only a JSON object with the keys "hotel", "city" and "price".

{"hotel": "Hotel Mira", "city": "", "price": ""}

Add one sentence, Use null for anything the review does not say, and the same prompt returns {"hotel": "Hotel Mira", "city": null, "price": null}. In JSON, null means no value: code can check for it and ask the customer, instead of storing an empty city or a made up price.