Upload files with a job
Files only help when ChatGPT knows how to use them
After this, you'll be able to upload a file, tell ChatGPT exactly what to extract or compare, and prevent it from inventing missing details.
Before you start
Complete Search when the answer can change first.
The idea
Uploading a file is not the same as giving ChatGPT a job. The file is source material. The request still has to say what to do with it.
- The output is a table tiedThe output is a table tied to the uploaded file
- Missing owners or dates areMissing owners or dates are marked instead of invented
- Evidence points back to theEvidence points back to the file
- Upload one file, extract aUpload one file, extract a table from it, and confirm at least one row marks missing
Here is the before and after: before, ChatGPT is answering a loose request. After, you can upload a file, tell ChatGPT exactly what to extract or compare, and prevent it from inventing missing details.
Now try it use the exercise prompt on one real task from today. Keep the answer small enough to check before you reuse it.
You are ready when the task, source, surface, and proof are all visible.
Try it (12 min)
Watch out for
- Uploading multiple files and asking a vague question across all of them.
- Letting ChatGPT summarize when you needed extraction, comparison, or decisions.
- Forgetting to tell ChatGPT what to do when the answer is missing.
- Using a normal chat for a file that should not be retained or reused.
Paste this into Claude
I uploaded a file. Use only that file. Job: Extract every deadline, owner, open question, and client request. Output: A table with columns Item, Type, Owner, Due date, Evidence, Missing information. Rules: - If the file does not name an owner, write "owner unknown." - If the date is implied but not explicit, write "date unclear" and explain why. - Do not add advice until after the table. - After the table, list the three highest-risk missing details.
What a good response looks like
Item: Draft campaign landing page. Type: client request. Owner: owner unknown. Due date: date unclear, the file says before launch but does not give the launch date. Evidence: Proposal section Deliverables. Missing information: owner and launch date.
What good looks like
- The output is a table tied to the uploaded file
- Missing owners or dates are marked instead of invented
- Evidence points back to the file
- Advice appears after extraction, not mixed into it
When this breaks
- Breaks when the file is complex and the request asks for a broad summary because important details disappear into the overview.
- Breaks when missing information is not allowed as an answer because ChatGPT may fill the table to make it look complete.
AI can help with this
Use ChatGPT to help you you can upload a file, tell ChatGPT exactly what to extract or compare, and prevent it from inventing missing details. Start with the exercise prompt and your real input. Ask for one draft, then check it against this proof: The output is a table tied to the uploaded file. Accept only the version you can verify yourself.
You can now
You can give an uploaded file a specific job
- You can ask for evidence from the file
- You can make unknown an allowed answer
- You can decide whether the file should persist
Key takeaways
Files become useful when the job is specific. Tell ChatGPT what to extract, how to show evidence, and how to mark gaps.
- A file is source material, not a task.
- Use tables when you need extraction or comparison.
- Ask ChatGPT to mark unknowns before it fills gaps.
- Private files require account and retention awareness.