Create intermediate Works with: Any AI

Information Extraction From Messy Text Prompt

Json output Has example

Pull structured data out of emails, notes, or documents.

The Open buttons send the prompt text as currently shown below to that site only when you click. Fill in the placeholders first.

Write a prompt to extract structured information from messy text. The input type and an anonymized example: [INPUT_EXAMPLE]. The fields I need: [FIELDS]. How the results will be used: [USE].

Produce:
1. The prompt text: task statement; field definitions with examples and formats (dates as YYYY-MM-DD, currency amounts as numbers, names as written); rules for missing information ("null" rather than guess); rules for ambiguity (list alternatives in a notes field); a rule to include a short quote that supports each extracted value; and a rule to ignore text that is quoted from earlier messages or signatures.
2. The output format as JSON with an example record, and an instruction to return a list when there are multiple items.
3. Three test inputs: a clean one, a messy one with abbreviations and typos, and a tricky one with conflicting values; for each, the expected output.
4. A review process: sample 20 outputs, compare with the source, and track error types by field.
5. Safety notes: remove or mask personal data when possible, avoid sending sensitive information to services that are not approved, and keep a record of what was extracted for audit.

Explain the trade-off between a strict schema and flexibility, and how to handle documents longer than the model can read (split by section and merge). Never invent values.

Fill in before use

Optional: type your details and the prompt above updates. Nothing is saved or sent.

Replace each bracketed placeholder with your own details, then paste into Any AI or another assistant.

Input: "Hi, pls send 40 boxes of the blue mugs to Dana at 12 Dock Rd by Fri 18th, PO 7731, thx"
Output
{"quantity": 40, "item": "blue mugs", "recipient": "Dana", "address": "12 Dock Rd", "due_date": "Friday 18th (month not stated)", "po_number": "7731", "support_quote": "by Fri 18th", "notes": ["month missing"]}
Rules: null for missing values; quote support for each field; ignore signatures and quoted earlier messages. Review 20 samples for errors by field and keep a table of error types (wrong date, wrong quantity, missed item). Remove or mask personal data when possible, and send sensitive text only to approved tools. For documents longer than the model can read, split by section and merge the records, deduplicating by PO number.

• extraction