Analyzing Documents and Files with ChatGPT

1Document ingestion and targeted extraction in ChatGPT

Document ingestion and targeted extraction in ChatGPT turns hours of manual document review into structured analytical dialogues. When a file is uploaded, the platform parses text layers, indexes core passages, and injects relevant snippets directly into the attention window.

With extensive records, the system may exhibit reduced focus on clauses located in the middle sections. Enforcing mandatory verbatim citations with page references compels the language model to locate the physical source before generating findings.

Tap to enlarge

2Auditing an annual PDF report to extract critical ratios

Auditing an annual PDF report to extract critical ratios eliminates manual data entry mistakes while preserving full auditability back to official financial notes.

An ESG analyst assesses a contractor's annual report to evaluate greenhouse gas disclosures.

Weak prompt.

Review the attached PDF and tell me the company's carbon emissions.

The model yields an aggregate figure without clarifying Scope boundaries or citing document locations.

Strong prompt (Turn 1).

Context: ESG audit of the attached annual PDF. Task: Extract the 2024 greenhouse gas inventory into a 4-column table: Emissions Scope (Scope 1, 2, 3), Emissions (Metric Tons CO2e), Year-over-Year Delta, and Source Page Number. Absolute rule: Quote the exact sentence verifying each reported metric.

The model extracts reported data cleanly with verifiable page references.

Turn 2: Capital expenditure audit follow-up.

To uncover underlying capital commitments without losing context, the analyst follows up:

From the same document, list in a three-column table the capital expenditure planned for decarbonizing the transport fleet: Initiative, Allocated Budget 2025-2026, Page. If an amount is not stated, write "not specified".

ChatGPT extracts the three budget lines without speculating, referencing the exact financial disclosure notes.

What changes. The initial prompt enforces strict accounting boundaries. The follow-up deepens the inquiry into actual investments without starting from scratch.

3Extract contractual obligations from a business agreement

Extract contractual obligations from a business agreement to practice employing ChatGPT as a dependable legal pre-review assistant.

To practice without handling sensitive records, here is an anonymized master services excerpt (fictional company NovaTech Ltd), ready to copy or save into logistics_agreement.docx:

[Logistics Services Agreement Excerpt - NovaTech Ltd (Fictional Training Sample)]
Article 3 - Service Level Agreements (SLA): The service provider guarantees monthly order platform availability of 99.5%.
Article 4 - Dispatch Timelines: All orders validated before 2:00 PM must ship on the same business day. In the event of delays exceeding 48 hours not caused by force majeure, a fixed delay penalty of €150 excl. VAT per calendar day is credited against the monthly invoice.
Article 5 - Data Residency & Security: Order tracking logs must reside exclusively within the European Union. (Audit Note: this contract contains no stipulations governing end-of-contract data reversibility or transfer protocols).

The "Load exercise" button in the composer pre-fills the objective with the weak starting prompt:

"Analyze this attached contract and list the provider's obligations."

From this starting point, complete the prompt with the composer. Define the legal counsel role and require a four-column table: article, obligation, verbatim quote, and delay penalty.

Enable the "not found" fallback for unstated terms, particularly data reversibility. Apply the end-instruction technique (arXiv:2307.03172) by placing your questions at the end to prevent attention loss across documents.

Self-evaluation rubric: (a) document ingestion proceeds without error; (b) contractual articles and delay penalties are accurately identified with verbatim citations; (c) the absent reversibility clause correctly returns "not found".

Open the prompt composer