Best prompts for ChatGPT using structured data-extraction templates

Extract the product details from the following text and return a single JSON object with exactly these keys: sku (string), title (string), short_description (string, max 200 chars), price (number, USD), currency (string, 3-letter), categories (array of strings), availability (boolean), weight_kg (number or null), dimensions_cm (object with width, height, depth numbers or null), images (array of URL strings). If a field is not present, use null or an empty array as appropriate. Do not include any extra keys. Input: "<<<PRODUCT_TEXT>>>"

Read the resume text inside <<RESUME>> and output one JSON object with these keys: name (string), contact {email, phone, location}, summary (string, max 160 chars), education (array of objects with degree, institution, start_year, end_year or null), experience (array of objects with title, company, start_month_year, end_month_year or 'Present', responsibilities (array of short strings)), skills (array of strings), total_years_experience (number, round to one decimal). Only populate fields you can infer. Use null for unknown values. Input: "<<RESUME>>"

Analyze the meeting notes in <<NOTES>> and output a CSV (header row included) with columns: action_id (A1, A2...), description, owner, due_date (YYYY-MM-DD or empty), priority (Low/Medium/High), status (Todo/In Progress/Done). Only include clear actionable items. Do not add commentary. Input: "<<NOTES>>"

Given the customer email text between <<EMAIL_START>> and <<EMAIL_END>>, return a JSON object with keys: from {name,email}, subject, received_date (YYYY-MM-DD if present), intent (one of: Support, Sales, Billing, Feedback, Other), urgency (Low/Medium/High), required_action (short instruction for agent), tags (array). Infer values conservatively and use null when not available. Input: <<EMAIL_START>> <<EMAIL_TEXT>> <<EMAIL_END>>

From the contract text between <<CONTRACT>> extract a single JSON object with these keys: parties (array of objects with name and role), effective_date (YYYY-MM-DD or null), termination_date (YYYY-MM-DD or null), term_length (string or null), payment_terms (string or null), liability_limit (string or null), termination_for_cause (boolean), governing_law (string or null), auto_renewal (boolean). Provide null for unknowns. Input: "<<CONTRACT>>"

Parse the clinical note provided in <<NOTE>> and output one JSON object with: patient_age (number or null), sex (Male/Female/Other/null), visit_date (YYYY-MM-DD or null), chief_complaint (string), diagnosis (array of strings), medications (array of objects: name, dose, route, frequency), vitals (object: bp, hr, rr, temp_c or null), plan (array of short strings). Use null for missing fields. Do not output PHI beyond age and sex. Input: "<<NOTE>>"