A file type tells software how information is organised. It helps you choose how to edit, read or exchange that information. It does not tell you whether the contents are correct, safe or approved to share.
If somebody asks an AI to “read the file”, I would first ask which file, what kind, and whether it is the copy we meant to use. The existing files and folders guide explains names, locations and copies. This lesson is about choosing the format and checking what survives a transfer.
A few types you will meet
The ending of a filename, such as .csv, is its extension. A system may also describe the format with a media type, sometimes called a MIME type. MDN lists common types. Neither label is a security certificate.
| Type | Common use | Question to ask |
|---|---|---|
| TXT or Markdown | Plain text and lightweight notes | Is the structure clear enough? |
| DOCX | An editable document | Which version is approved? |
| A document for reading and sharing | Can the needed text be extracted accurately? | |
| XLSX | A spreadsheet workbook | Which sheets, formulas and hidden content matter? |
| CSV | Rows and fields exchanged as text | Were IDs, dates and numbers imported correctly? |
| JSON | Structured data exchanged between systems | What do the named fields and nested records mean? |
| PNG or JPEG | Images, including screenshots | Is the text an image rather than selectable text? |
The table describes typical uses, not a promise that every AI tool supports every format. Check the tool's actual support and its approved use with your information.
CSV is not the whole spreadsheet
CSV's documented format describes text records and fields. It does not carry the full experience of a workbook. Microsoft explains that saving to text or CSV can lose formatting and features.
Suppose an invented customer ID is 0017. It is an identifier, not a number to add up. Some spreadsheet imports can remove the leading zeros unless the field is handled as text. Microsoft's import guidance covers that issue.
The same care applies to dates. Write the intended interpretation down rather than asking the AI to guess whether 03/04/2026 means March or April. Keep the original; use a separate practice copy for conversions.
JSON can express named fields and nested structures, as RFC 8259 explains. It still needs a definition of what those fields mean. A neat structure does not settle a business ambiguity.
A familiar extension does not make a file safe
Renaming an unexpected attachment to end in .pdf does not convert its contents or establish that it is safe. OWASP's file-upload guidance warns that claimed types can be spoofed and recommends several checks for upload services, not reliance on one label.
For a business team, my proposed routine is to use approved receiving and scanning tools, confirm an unexpected file through a known contact route, and follow IT's instructions about active content. Do not upload a confidential attachment to a public converter or scanner just to see what it contains.
An AI may miss text in an image, misunderstand a table or read only part of a document. For an important result, ask which pages or records it actually inspected and check a sample against the original. “It accepted the upload” is not the same as “it understood the whole file”.
Ask your AI to teach you
Use an approved chat with no file connections for this exercise. A prompt is not a malware scanner or a restriction on the tools an agent really has.
Check that you can use it
You receive a CSV exported from a workbook. Can you assume it includes every sheet, formula and format? No. Can you safely change customer ID 0017 to 17 without knowing the receiving system's rules? No. Decide what must be preserved and check a few records before relying on a transfer.
Continue to keeping business data secure, or revisit SQL to see how a data extract is selected.
Sources and boundaries
Sources were checked on 21 October 2026. MDN and the RFCs describe formats; Microsoft explains its import/export behaviour; OWASP addresses application file-upload controls. The scenarios and team routine are proposed practice. No particular attachment, provider's parser or conversion service has been tested or approved by this lesson.