12 practical ChatGPT Work tasks for people who don't code
Twelve real, non-technical ChatGPT Work tasks - inputs, expected output, a prompt to start from, and what to protect before you run each one.
Last updated July 22, 2026
ChatGPT Work is built for real, everyday knowledge work - not just chatting. You can hand it files, point it at a folder, and ask it to produce something you would otherwise spend an afternoon on: a report, a cleaned-up spreadsheet, a client handoff, a content plan. None of it requires code. What it requires is a clear ask, the right inputs, and a habit of reviewing the output before you trust it.
Below are twelve tasks that non-technical people actually run - marketers, analysts, researchers, founders, writers. For each one you’ll get the inputs to gather, the output to expect, a prompt to start from, what to check afterwards, and which files to protect before you begin. The tasks that touch folders on your Mac are the ones where protecting the folder first genuinely matters, and they’re flagged as you go.
On this page
- Two kinds of task: uploads vs. local folders
- Research & writing tasks (1–6)
- Planning, files & templates (7–12)
Two kinds of task: uploads vs. local folders
Before the list, one distinction that changes how careful you need to be. Some of these tasks live entirely in the chat: you upload a few files or paste text, ChatGPT Work produces something, and you copy the result out. Nothing on your Mac changes. Other tasks give ChatGPT Work access to a real folder on your disk, and ask it to read, rename, move, or rewrite the files in place. That second kind is where a mistake has consequences.
When ChatGPT Work works on a local folder, it can overwrite a file, move it somewhere you didn’t expect, or delete something it judged redundant - and the summary it writes in the chat is a description of what it intended, not a dependable record of what it actually did. For the local tasks below, that’s why each one ends with a short “protect the folder first” note.
The folder safety net, in one paragraph
CoworkRestore is a Mac menu-bar app that watches a folder you choose and quietly saves versions (“snapshots”) as its files change. It works alongsideChatGPT Work - there’s no ChatGPT extension or integration to install, and no connection to OpenAI. When something looks wrong, you open the folder’s history, see exactly which files changed, and restore one file or the whole folder. It’s Apple Silicon Mac-only (macOS 13+), local only, and it can only restore versions of files it was already watching - it’s a safety net for the moment a task goes sideways, not a backup and not a way to recover something from before it was installed. See how to safely use ChatGPT Work with local folders for the full setup.
Research & writing tasks
1. Organize scattered research into a structured summary
Inputs: A pile of source material - PDFs, article links, meeting notes, a few web pages you’ve saved. Upload the files or paste the text.
Expected output: A structured summary that groups findings by theme, notes where each point came from, and flags where sources disagree.
Here are eight sources on remote-team productivity. Read them and give me a one-page summary grouped by theme (tools, meetings, async communication). Under each theme, list the key claims and note which source each came from. At the end, list any points where sources contradict each other.
Review afterwards: Spot-check two or three claims against the original sources - summaries can smooth over nuance or attribute a point to the wrong document.
Files to protect: None. This is an uploads-only task; nothing on your Mac changes.
2. Turn rough notes into a finished report
Inputs: Your raw notes - bullet points, half-sentences, numbers you jotted down - plus a sense of who the report is for and how long it should be.
Expected output: A clean report with an introduction, clearly headed sections, and a short conclusion, in the tone you specify.
Turn these notes into a two-page report for our leadership team. Keep the tone professional but plain. Structure it as: summary, what we found, what we recommend. Don’t invent any numbers I didn’t give you - if something is missing, leave a clearly marked gap.
Review afterwards: Read for invented facts. The most common failure is a confident sentence that fills a gap you left open. Confirm every figure traces back to your notes.
Files to protect: None if you paste the notes and copy the result out. If you point it at a folder and ask it to save the report there, treat it as a local task and see the note under task 12.
3. Clean up a messy spreadsheet export
Inputs: A spreadsheet or CSV export that’s inconsistent - mixed date formats, stray blank rows, names capitalized five different ways, duplicate entries.
Expected output: A tidied version with consistent formatting, duplicates removed, and a short note listing every change it made.
This CSV of event registrations is messy. Standardize all dates to YYYY-MM-DD, make names Title Case, remove exact duplicate rows, and drop fully blank rows. Don’t change any email addresses. Give me the cleaned file plus a list of exactly what you changed and how many rows were affected.
Review afterwards: Check the row count before and after. If it deduplicated more aggressively than you expected, you want to know before you send the file on. Confirm no email or ID column was altered.
Protect the folder first
If you let ChatGPT Work overwrite the spreadsheet in place instead of producing a new copy, the original is gone the moment it saves. Keep a folder safety net running so the pre-clean version stays one click away, or simply ask for the cleaned data as a new file and keep the original untouched.
4. Assemble a client handoff package
Inputs: The deliverables you’re handing over - documents, a summary of the work, next steps - plus any template your team uses for handoffs.
Expected output: A single, well-organized handoff document: what was delivered, how to use it, open items, and contacts, in a consistent voice.
Using these three documents and my notes, assemble a client handoff. Include: what we delivered, how to use each item, what’s still open, and who to contact for what. Match the tone of the sample handoff I uploaded. Keep it to two pages.
Review afterwards: Confirm nothing internal leaked in - pricing notes, private commentary, unfinished sections. A handoff is client-facing, so read it as the client would.
Files to protect: None if you assemble from uploads. If it’s pulling from and writing into a live client folder, protect that folder first - see task 12.
5. Review a set of documents for inconsistencies
Inputs: Several related documents that should agree - a proposal and a contract, a set of policy pages, three versions of the same pitch - uploaded together.
Expected output: A list of contradictions and mismatches: numbers that don’t line up, dates that conflict, claims made in one place and contradicted in another.
These four documents should tell a consistent story. Compare them and list every inconsistency you find - conflicting figures, dates, names, or claims. For each one, quote the exact wording from each document so I can check it myself. Don’t fix anything; just report.
Review afterwards: Verify each flagged inconsistency against the quoted wording. Asking it to quote sources (rather than paraphrase) makes this far easier to trust.
Files to protect: None - a review that only reports and doesn’t edit is safe to run on uploads.
6. Build a project brief from a loose conversation
Inputs: The messy raw material of a new project - an email thread, a call transcript, a few Slack messages, whatever captures what someone wants.
Expected output: A structured brief: goal, scope, audience, deliverables, timeline, open questions - with the open questions clearly separated from the decisions.
From this email thread and call transcript, write a project brief. Sections: goal, audience, scope, deliverables, timeline, and open questions. Put anything that wasn’t clearly decided under open questions rather than guessing. Keep it to one page.
Review afterwards: Read the “open questions” section first - it tells you what the AI wasn’t sure about, which is exactly what you need to chase down.
Files to protect: None. Uploads and paste only.
Planning, files & templates
7. Create a content calendar
Inputs: Your topics or themes, the channels you post to, how often you publish, and any key dates (launches, holidays, events).
Expected output: A dated calendar mapping topics to channels and publish dates, usually as a table you can paste into a sheet.
Build a 6-week content calendar for LinkedIn and our newsletter. We publish twice a week on LinkedIn and once a week by email. Here are the ten themes I want to cover and two launch dates to work around. Give me a table with date, channel, theme, and a one-line hook for each post.
Review afterwards: Check the cadence and the dates against your real calendar - it may schedule around the wrong week or double-book a launch day.
Files to protect: None. This produces a plan you copy out; nothing on your Mac changes.
8. Classify a folder of files
Inputs: A folder of mixed files - invoices, contracts, images, drafts - and the categories you want them sorted into.
Expected output: Either a proposed classification you approve first, or (if you let it act) the files moved into category subfolders, with a report of what went where.
Look at the files in this folder and classify each one as Invoice, Contract, Image, or Draft based on its contents and name. First show me the proposed classification as a list - don’t move anything yet. After I approve, move each file into a subfolder named for its category.
Review afterwards: Approve the proposed list before anything moves. After the move, confirm the folder count matches and nothing landed in an “uncategorized” limbo.
Protect the folder first
This is a local task - ChatGPT Work is moving real files on your disk. A misread name can send a file into the wrong bucket, and the chat summary won’t always show it. Run a folder safety net so the before-sorting layout stays recoverable, and see how to let AI organize your files without losing control for the full playbook (test small, approve the plan, snapshot, roll back if needed).
9. Prepare a meeting pack
Inputs: The agenda, background documents, last meeting’s notes, and any data the group needs to decide on.
Expected output: A single pre-read: agenda with time boxes, a short brief per item, decisions needed, and links or references to the supporting material.
Prepare a meeting pack from these documents. Start with the agenda and a suggested time for each item. Under each item, give a three-sentence brief and state the decision we need to make. End with a list of open action items carried over from last time. Keep the whole thing to two pages.
Review afterwards: Confirm each “decision needed” is actually a decision and not a restatement of the topic - a common way these packs go soft.
Files to protect: None if built from uploads.
10. Consolidate feedback from many people
Inputs: All the feedback you’ve collected on something - comments, emails, survey responses, tracked-change notes - pasted or uploaded together.
Expected output: A de-duplicated, grouped summary: common themes, how many people raised each, contradictory requests surfaced side by side, and a suggested priority order.
Here is feedback from twelve people on our draft. Group it into themes, tell me how many people raised each theme, and put any contradictory requests next to each other so I can see the tension. Don’t resolve the contradictions - just make them visible. Suggest a priority order at the end.
Review afterwards: Sanity-check the counts and make sure a strong minority opinion wasn’t averaged away. The value here is seeing the disagreement, not smoothing it over.
Files to protect: None. Uploads and paste only.
11. Create reusable templates
Inputs: A few real examples of a document you produce often - three past proposals, several status updates, a handful of onboarding emails.
Expected output: A reusable template that captures the common structure, with clearly marked fill-in-the-blank placeholders and short notes on what goes in each.
Look at these three proposals I’ve sent before and build a reusable template from them. Keep the structure and tone they share, replace the specifics with clearly marked placeholders like [CLIENT NAME] and [SCOPE], and add a one-line note under each section explaining what belongs there.
Review afterwards: Make sure no real client’s details survived into the template as a placeholder’s default. Run it once on a fresh example to see if it holds up.
Files to protect: None if you build from uploads and save the template yourself.
12. Update a collection of related documents at once
Inputs: A folder of documents that all need the same change - a renamed product, an updated address, a new policy line - and a precise description of the change.
Expected output: Each document updated consistently, plus a report listing every file it touched and every change it made in each one.
Every document in this folder still uses our old company name, “Northwind Co.” Replace it with “Northwind Studio” everywhere it appears, including in headers and footers. Don’t change anything else. When you’re done, give me a report listing each file you edited and how many replacements you made in it.
Review afterwards: Open two or three of the edited files yourself - a bulk find-and-replace can catch the name inside a URL or an unrelated phrase. Compare the report’s file count to the folder’s.
Protect the folder first
This is the riskiest task on the list, because ChatGPT Work is editing many real files in place at once. A single over-eager replacement, repeated across a dozen documents, is a lot to fix by hand - and the chat report is a description, not a guaranteed record. With a folder safety net running, you can see exactly which files changed and restore any one of them (or all of them) to the pre-change version. Set it up on the folder before the run, since it can only restore versions it was already watching.
The habit that makes all of this safe
The pattern across every task is the same. Give clear inputs, ask for a report of what changed, and review the output before you rely on it. For the tasks that only read uploads and hand back text, that’s the whole job. For the tasks that reach into a real folder on your Mac - cleaning spreadsheets, classifying files, updating a collection - add one more step: protect the folder first, so a change you didn’t intend stays reviewable and one click from being undone.
If you’re setting this up for the first time, start with the calm setup for ChatGPT Work and local folders, and see how CoworkRestore works alongside ChatGPT Work for the folder-watching side. ChatGPT Work itself is documented at openai.com.