Cadastre-se grátis

Extract fields to verify from invoices

Autor: AILesson9 min de configuraçãoTestado com:ChatGPTRevisado em: 2026-08-28

Resposta rápida

Extract invoice headers, lines, tax, payment terms, and anomalies with source locations and arithmetic checks. Forneça: Invoice text or OCR, Required fields and definitions, Verification and privacy controls. Resultado esperado: A source-linked invoice record, line table, verification queue, duplicate keys, and reconciliation.

1

Adicione seu contexto

Seu texto permanece neste navegador. O AILesson Prompts não o envia para um modelo ou servidor.

2

Seu prompt

Campos não preenchidos permanecem visíveis como marcadores de posição, para que você ainda possa copiar e editar o prompt

Extract fields from the supplied invoice for human verification.

Invoice text or OCR:
[invoice]

Required schema:
[schema]

Controls:
[controls]

Treat OCR as unverified evidence. Preserve the exact source string and page or line location for every extracted value; use Not found or Unclear rather than guessing. Never normalize an ambiguous digit, date, currency, tax ID, bank detail, quantity, rate, or total without showing the original. Extract header fields and one row per line item, then check quantity × unit price, discounts, subtotal, tax by rate, withholding, prior payment, rounding, and amount due under the supplied definitions. Do not infer tax validity, vendor legitimacy, delivery, approval, bank ownership, or payment eligibility. Flag unexpected bank changes, duplicate keys, PO mismatches, missing mandatory fields, unreadable text, and arithmetic differences for the authorized reviewer. Output extraction tables, reconciliation, field-level confidence and reason, and a verification queue ordered by payment risk.
Experimente no Playground
Privado por padrãoA montagem do prompt acontece localmente no seu navegador. Evite colocar informações confidenciais em qualquer serviço de IA, a menos que sua organização permita.

Da entrada ao resultado

Um exemplo prático

Veja como um contexto concreto transforma esta receita em um resultado útil

Entrada real

Invoice text or OCR
Page 1 lines 1–5: NORTHLIGHT PRINT CO. | Invoice NL-2048 | Invoice date 2026-08-18 | Due 2026-09-17 | Currency CNY Lines 6–8: Bill to River Studio | PO RS-771 | Tax ID 91310O?45 (OCR; sixth character may be zero or letter O) Table lines 10–13: Design proof | qty 2 | unit 450.00 | amount 900.00; A4 booklets | qty 300 | unit 8.00 | amount 2,400.00; Delivery | qty 1 | unit 180.00 | amount 180.00 Lines 15–19: Subtotal 3,480.00 | Discount 100.00 | Tax 6% 202.80 | Total 3,582.80 | Deposit received 500.00 | Amount due 3,082.8O (last character OCR letter O) Line 22: Pay to Harbor Bank acct ending 8841. Handwritten: ‘new bank from Aug’. No delivery date shown.
Required fields and definitions
Header: vendor legal name, invoice number, invoice date, due date, currency, buyer, PO, supplier tax ID, bank ending. Lines: description, quantity, unit price, line amount. Totals: subtotal before discount, discount, taxable base, tax rate, tax, total, deposit, amount due. ISO dates; amounts decimal. Duplicate key vendor + invoice number + gross total. Match PO amount and approved bank against reference. Arithmetic: taxable base = subtotal - discount; tax = base × rate; total = base + tax; due = total - deposit.
Verification and privacy controls
Reference: PO RS-771 approved maximum CNY 3,600 including tax; expected items and quantities match all three lines. Vendor master bank ends 1139, not 8841. No prior invoice NL-2048 in register. Tolerance CNY 0.01. Medium or low confidence needs Accounts Payable review. Never expose full bank or tax identifiers. Extraction cannot approve payment or validate tax. Bank change always blocks payment until vendor verification through an approved independent contact.

Exemplo de saída

Invoice extraction — verification required

Header: vendor NORTHLIGHT PRINT CO. (p1 l1, high); invoice NL-2048 (l2, high); invoice date 2026-08-18 and due 2026-09-17 (l3–4, high); currency CNY (l5, high); buyer River Studio and PO RS-771 (l6–7, high); tax ID source “91310O?45” (l8, low, redacted display 913…?45; do not normalize); bank ending 8841 (l22, high transcription, failed master-data match).

LineQtyUnitAmountCheck
Design proof2450.00900.00matches
A4 booklets3008.002,400.00matches
Delivery1180.00180.00matches

Line sum 3,480.00 matches subtotal. Taxable base 3,480.00 − 100.00 = 3,380.00. Tax 3,380.00 × 6% = 202.80. Total 3,380.00 + 202.80 = 3,582.80. Due 3,582.80 − 500.00 = 3,082.80; source “3,082.8O” is low-confidence OCR but arithmetic supports 3,082.80 for review, not silent correction. Gross total is within PO maximum by 17.20 and listed items match the supplied PO reference; this does not prove receipt or approval. Duplicate key is new in the supplied register.

Payment-risk queue: 1) Block payment: bank ending differs from master 1139 and handwritten change is not verification; confirm through approved independent contact. 2) Verify ambiguous tax ID against authoritative vendor record. 3) Visually confirm amount-due final character. 4) Confirm deposit evidence and goods/service receipt; delivery date is not found. 5) Authorized reviewer decides tax validity and payment eligibility. Extraction status: complete with blockers; no payment approval.

Por que isso funciona

  1. 1

    Source strings and locations make every normalized field auditable against imperfect OCR

  2. 2

    Arithmetic and reference checks catch payment risks that accurate transcription alone cannot

Verifique o resultado

  • Can every extracted or normalized value be traced to an exact source location and original string?

  • Do lines, subtotal, tax, prior payment, and amount due reconcile within the stated tolerance?

  • Are bank changes, duplicates, PO differences, missing fields, and ambiguity kept out of automatic approval?

Use com confiança

Perguntas frequentes

Respostas práticas sobre quando usar esta receita, o que fornecer e onde a revisão humana ainda é importante

What should I prepare before using “Extract fields to verify from invoices”?

For “Extract fields to verify from invoices,” prepare Invoice text or OCR, Required fields and definitions, and Verification and privacy controls. Replace placeholders only with information you can verify. If a detail is unknown, preserve that uncertainty explicitly instead of asking the model to infer it.

When is the “Extract fields to verify from invoices” result not ready to use?

The result is not ready if it does not yet deliver the stated outcome—A source-linked invoice record, line table, verification queue, duplicate keys, and reconciliation—from the supplied evidence, or if it relies on unresolved assumptions, missing approvals, or invented details. Use the checks as release gates: revise the source inputs or assign a named, authorized reviewer instead of polishing an unsupported output.

Which AI tools have recorded tests for “Extract fields to verify from invoices”?

The published test record for “Extract fields to verify from invoices” lists ChatGPT as of 2026-08-28. This confirms recorded runs, not guaranteed compatibility or identical results in later product versions. For another tool or version, keep every constraint visible and repeat the result checks before use.

Mais maneiras de explorar

Onde esta receita se encaixa

Prompts de textoAnálise de documentosAnálise de dadosFinanças, compras e administraçãoCompras e gestão de fornecedores

Mantenha o trabalho em andamento

AILesson · Cursos recomendados

Seu próximo passo: use a IA na prática

Passe de entender a IA para fazer as coisas acontecerem. Pratique como criar prompts, revisar e melhorar resultados por meio de lições interativas para o trabalho e o dia a dia.

Explore todos os cursos
Capa de IA para iniciantes

15 lições interativas

IA para iniciantes

Inicie uma conversa útil com IA, melhore a resposta, confira o que importa e saiba quando o próximo passo é seu.

Explore este curso
Capa de Fundamentos do ChatGPT

10 lições interativas

Fundamentos do ChatGPT

Aprenda a usar o ChatGPT por meio de três fluxos de trabalho práticos: conclua uma primeira conversa, trabalhe a partir de um arquivo e use Projects para levar um plano verificado a uma nova conversa.

Explore este curso
Capa de Como a IA generativa funciona

7 lições interativas

Como a IA generativa funciona

Entenda como chats comuns de IA aprendem padrões da linguagem, usam mensagens atuais, criam respostas e incorporam fontes e ferramentas.

Explore este curso