Skip to main content
Define the fields you need, attach a document, and Anara reads it and returns a structured output using only what the document says. Run the same prompt against every document in the stack. One chat per document, consistent output every time. For professionals who process the same document type repeatedly and need a reliable, documented extraction they can trust.

​
1. Describe the task

You receive stacks of documents that all have the same structure but different content. Infraction tickets. Construction tender specifications. Service contracts. LEVI procurement documents. The fields you need from each one are the same. Reading them end to end to find those fields is slow, inconsistent, and scales badly as the volume grows. Anara can run a fixed extraction prompt against each document and return only what the document actually says, in the fields you define. You write the prompt once, refine it until it handles your edge cases, and then run it against every new document in the stack. One chat per document. Structured output every time. The most effective approach, developed over months of real use: treat your extraction prompt as a versioned protocol. Start with a first draft, test it against three or four documents, note where it produces wrong or missing cells, and revise. By version five or six, the prompt handles the exceptions automatically. One professional in the painting trade is currently on version ten of their extraction protocol for construction tender specifications. Here a regulatory compliance analyst is processing a backlog of Brazilian administrative infraction tickets and needs to extract the same six fields from each document in a consistent, auditable format. 
Extract the following fields from the attached document. Use only what the document explicitly states. If a field is not present in the document, write "Not found" for that field. Do not infer or estimate any value.

Fields to extract:
- Case number
- Date of infraction
- Entity cited
- Violation type (quote the specific article or clause cited)
- Penalty amount (exact figure as stated)
- Deadline for response or appeal

After the extraction, quote the exact passage from the document where the penalty amount appears.

​
2. Give Anara context

Required context
  • The document to extract from, uploaded to your library or attached to the chat.
Optional context
  • A completed example extraction from a prior document of the same type. Anara uses it to match your format and level of detail.
  • The document type and jurisdiction, if the extraction depends on reading legal or regulatory conventions that are not obvious from the document text alone.

​
3. What Anara creates

A structured output for this document: each field filled with the value the document explicitly states, or flagged as “Not found” where the document does not contain it. The required source passage at the bottom gives you an audit trail for the most important field in the extraction. Paste the output into your tracking spreadsheet or database and open a new chat for the next document.

​
4. Follow-up prompts

​
Add a verification step before extracting

When you want Anara to confirm what type of document it is reading before proceeding with the extraction. 
Before extracting any fields, identify the document type and confirm whether this document is a formal infraction notice, an administrative warning, or a pre-infraction notification. Then proceed with the extraction only if it is a formal infraction notice. If it is not, stop and tell me what it is instead.

​
Generate a structured output from the extracted scope

When the extraction is complete and you need to produce a downstream document from it. 
Using the extracted fields from this infraction ticket, draft a formal response letter from the cited entity acknowledging receipt of the notice, stating the grounds for appeal, and requesting the thirty-day extension permitted under Article 74.

​
Compare two documents for consistency

When two documents cover the same case or contract and you need to check that the terms align. 
I have two documents for the same case. Extract the penalty amounts and response deadlines from both and compare them. Flag any discrepancy between what the two documents state.

​
5. Tricks, tips, and troubleshooting

​
How you word your prompt shapes what you get

Add a “CRITICAL INITIAL VERIFICATION” step at the start of your extraction prompt once you know your edge cases. For example, if your document type sometimes includes factory-applied specifications that should not be extracted alongside field-applied ones, add a verification step that checks for this before extracting. This single addition prevents the class of errors that most often invalidate an extraction. Ask for a source quote on the most critical field in every run. If the extracted value does not appear in the quoted passage, the extraction has failed and you need to run it again.

​
Check the output against your own understanding

You know your document type better than Anara does. Infraction tickets, tender specifications, and service contracts all have idiomatic language that Anara may interpret differently than a trained specialist would. Check the first five or ten extractions against the originals before you trust the protocol on new documents. If Anara consistently misclassifies a field or quotes the wrong passage for a specific document structure, add a correction to your extraction prompt as a named exception. Version the prompt: keep a record of each change so you can trace why the protocol produces what it produces.

​
What to do with the output next

Paste the structured output into a spreadsheet or database alongside the document reference. Over time, sorting by violation type, penalty amount, or deadline gives you patterns that are invisible when reading documents one at a time. If you maintain a separate folder for background context you want Anara to draw on but not cite as primary evidence, keep it separate from the documents you are actively extracting from so the extraction stays clean.