ChatGPT Converts Long Pastes to Attachments: Update Your Evidence Workflow

Name and version the source

Before pasting or uploading, give the source a stable file name and preserve its original format. Record the title, owner, retrieval URL, retrieval date, publication or revision date, checksum where appropriate, license or access boundary, and the reason it is being used.

If the source arrived as a paste and ChatGPT created the attachment, download or otherwise preserve the exact source outside the conversation when policy permits. The chat should not be the only archive.

Check ingestion before analysis

1. Ask for the exact opening and ending text, then compare it with the source.
2. Sample a middle section and every material table or appendix.
3. Check page order, headings, columns, footnotes, symbols, and non-English characters.
4. Record missing pages, inaccessible images, or formatting that changed meaning.
5. Split the document when the task needs precise section-level reasoning.

For a scanned PDF, image-only file, or complex layout, the attachment may require optical character recognition or visual inspection. Do not treat a fluent summary as proof that the source was parsed completely.

Prompt with source boundaries

Name the attachment and limit the task: which sections may be used, which claims need extraction, how uncertainty should be reported, and whether outside knowledge is allowed. Ask for references by heading, page, paragraph, or line marker, then open those locations before accepting the answer.

Separate quotation, paraphrase, interpretation, and recommendation. A document can directly support a product fact while the proposed implementation remains the model’s interpretation. Keep the claim ledger outside the conversation so another reviewer can reproduce it.

When comparing documents, assign stable labels such as Source A and Source B and require the answer to distinguish conflicts rather than blending them into a synthetic rule.

Protect sensitive and licensed material

Confirm that the account, workspace, data controls, contractual terms, and organizational policy permit the upload. Remove credentials, personal data, customer records, unpublished financial information, privileged material, and unnecessary confidential sections before sending the file.

Do not upload a full licensed book, report, dataset, or paywalled publication merely because the interface accepts it. Use the minimum necessary excerpt where rights and policy permit, and preserve attribution.

For collaborative work, assign one person to prepare and version the attachment and another to verify the claims taken from it. Record whether the conversation used the original file, a converted paste, or a later revision. When an answer changes after the file is replaced, treat it as a new run rather than silently updating the old evidence record. Delete or retain uploaded material according to the organization’s documented retention rule, not according to convenience.

The source-bounded AI brief workflow provides the evidence ledger, while the publishing guide supplies the final skeptical review.

Primary documentation

• OpenAI: ChatGPT release notes