A practitioner’s guide to ChatGPT.
Not a prompt list. This is how people who use ChatGPT for real work structure it: what to put in front of the model, what to reuse, and what to check before anything leaves the building. Independent, and not affiliated with OpenAI.
Start from the task, not the tool
The people who get the most out of ChatGPT did not learn features first. They picked one weekly task that eats an afternoon — a client update, a QBR deck, a batch of support replies, a data pull — and rebuilt it with the model in the loop until the output was good enough to send. Then they moved to the next one. Feature tours produce enthusiasm; task rebuilds produce hours back.
Projects and custom GPTs
A plain chat forgets. A project remembers — the files, the tone, the standing instructions — which is why serious use looks like a handful of well-kept projects rather than an endless scroll of one-off threads. When a workflow is stable and other people need it, package it as a custom GPT so the context travels with the tool instead of living in one person’s head.
- One project per recurring output, with its reference documents attached.
- Write the standing instructions as if briefing a new hire: audience, format, what to never do.
- Promote to a custom GPT once a second person needs the same result.
- Prune quarterly. Stale reference files are how good workflows quietly go wrong.
Four habits that separate serious practice
The difference between a mediocre answer and a usable one is almost always what you gave it to read, not how you worded the ask.
Anything you have prompted twice belongs in a project or a custom GPT with the reference files attached.
Numbers, names, dates, citations. Verify the parts that would embarrass you, and stop re-reading the parts that never break.
Save the prompts and outputs that worked. It is the only way practice compounds instead of resetting every Monday.
Rolling it out to a team
Licences are not adoption. The rollouts that stick pick three workflows, name an owner for each, agree in writing what data is allowed near the tool, and publish the shared projects on day one so nobody starts from a blank box. Everything else — policy documents, training decks, dashboards — can wait until people have used it on something that mattered.
Frequently asked
What can ChatGPT actually do at work?
Beyond answering questions: it reads documents you upload, holds context across a project, writes and runs code on data you give it, searches the web, and drafts in your voice once you have taught it what that sounds like. Most teams use maybe a tenth of it.
What is the difference between a project, a custom GPT, and a plain chat?
A plain chat is a one-off. A project keeps files, instructions, and history together so the same context is reused every time. A custom GPT packages instructions and reference files so other people on your team can run the same workflow without rebuilding the prompt.
Is there an official ChatGPT certification?
No credential from OpenAI decides who is good at this, and we do not issue one either. Judge a practitioner by what they shipped, who used it, and what it replaced.
How do I get better at ChatGPT?
Pick one task you do every week, rebuild it as a reusable project or custom GPT, and measure whether the output is good enough to send without rewriting. Then do the next one. A coach shortens that loop by reviewing your actual work instead of teaching prompt tricks.