ChatGPT for Work
Teaches teams to convert ad-hoc ChatGPT use into shared, reviewable workflows using custom GPTs, projects, and data analysis.
- Who it's for
- Knowledge workers across operations, marketing, sales, finance, HR, and research
- Format
- 1 day core, optional half-day for team rollout and governance
- Prerequisites
- None. A ChatGPT account for hands-on exercises; a team or enterprise plan for the rollout module.
By the end
What your team walks out with.
- Turn a personal prompting habit into a custom GPT the rest of the team can use
- Use data analysis on real spreadsheets, including recognising when its output should not be trusted
- Apply an appropriate verification standard to each class of task
- Set organisational rules for what company data may be entered, under which plan
- Avoid the shadow-IT pattern where valuable workflows live in individual accounts
4 modules
How the programme runs.
01From prompts to reusable assets
2 hoursA working custom GPT built from a real recurring team task.
02Data analysis on real spreadsheets
2 hoursAn analysis workflow plus a checklist for when results need independent verification.
03Verification standards by task class
1.5 hoursA task inventory with a defined review requirement attached to each category.
04Rollout, data policy, and avoiding shadow IT
Half day (optional)A written data policy and a plan for moving workflows out of personal accounts.
ChatGPT is usually already in an organisation before any decision is made about it. People sign up individually, work out what it is good for on their own, and build habits nobody else can see. Some colleagues get substantial value; others conclude it is not useful for their work. Nothing is written down, and a meaningful amount of company information has been entered into accounts the organisation does not control.
This curriculum is about converting that situation into something deliberate: shared assets rather than individual habits, defined verification standards rather than personal judgement, and an explicit data policy rather than an unexamined one.
From personal prompts to shared assets
Module one addresses the waste at the centre of most organisations' usage. Someone works out, over several weeks, exactly how to get a good first draft of a particular recurring document. That knowledge lives in their head and their chat history. A colleague doing the same task starts from nothing.
Participants take a real recurring task and build a custom GPT for it — instructions, reference material, examples of good output, explicit constraints. The acceptance test is that a colleague who was not involved in building it can produce comparable output on their first attempt. First versions routinely fail this test, because the builder's assumptions are invisible to them and obvious in their absence. Fixing that is the exercise.
Data analysis, and its confident errors
Module two works with real spreadsheets. The feature is genuinely strong — it writes and executes actual code, so the arithmetic is correct and it will happily do work that would take an afternoon manually.
The risk is not bad maths; it is confident misinterpretation. It can take an ambiguous column header to mean something other than what it means, silently drop rows with inconsistent formatting, treat a mixed-type column as text, or make a defensible-but-wrong assumption about what a blank cell represents — and then produce a clean, confident chart based on it.
Participants learn a short verification routine that catches most of this: confirm the row count matches expectations, ask explicitly how each relevant column was interpreted, check whether anything was dropped during loading, and sanity-check a total against a known figure. It takes a couple of minutes and is the difference between a useful tool and a liability.
Verification standards by task class
Module three produces a task inventory with review requirements attached. The organising principle is that verification effort should scale with consequence, not be applied uniformly.
Internal brainstorming needs essentially none. A first draft that a human will substantially rewrite needs light review. Anything reaching a customer, a regulator, or a financial record needs full independent verification of every factual claim. Writing this down as a shared standard removes it from individual discretion, which is where inconsistency comes from.
Rollout and the shadow-IT problem
The optional half-day deals with the situation most organisations are actually in: valuable workflows sitting in personal accounts, and confidential data already entered under consumer terms.
Teams work through the practical differences between plan tiers on training and retention, produce a written policy on what may be entered and where, and plan the migration of useful individual workflows into shared team assets. The migration matters more than the policy — a policy that makes people's existing workflows harder without offering a replacement is ignored rather than followed.
Related curricula
Equivalent programmes exist for other assistants: Claude for work, Microsoft Copilot for work, and Google Gemini for work. Engineering teams should see loop engineering.
Related reading
- How to actually work with AI agents: a communication guide
- ChatGPT for academic researchers
- How to read AI benchmarks
- AI alignment for product teams
Sessions are delivered by explainx.ai and adapted to the organisation's plan tier, data policy, and functions.
Common questions
- Most of our team already uses ChatGPT. Is this still useful?
- Usually more useful. Widespread individual use with no shared structure is the specific problem this curriculum addresses — valuable workflows trapped in personal accounts, inconsistent quality between colleagues, no agreed verification standard, and confidential data entered under personal plans nobody has reviewed.
- What is a custom GPT and why does it matter for teams?
- A reusable configuration — instructions, reference files, and behaviour — saved and shared rather than retyped. It matters because it converts one person's hard-won prompting knowledge into something a colleague can use directly, which is the difference between a team capability and an individual skill.
- How reliable is the data analysis feature?
- Genuinely useful and genuinely capable of confident errors. It writes and runs real code, so arithmetic is sound, but it can misinterpret a column, silently drop rows, or misread an ambiguous header. Module two teaches a verification routine — checking row counts, confirming column interpretation, and sanity-checking totals — that catches most of it.
- What about entering confidential company data?
- A dedicated module topic, because it is where most organisations have real unmanaged exposure. Teams work through the differences between personal, team, and enterprise plans on training and retention, and produce a written policy rather than leaving it to individual judgement.
- How does this compare to your Claude or Copilot curricula?
- The structural approach is similar and the emphases differ: this one leans on custom GPTs and data analysis, the Claude curriculum on projects and long-document work, and the Copilot curriculum on tenant grounding inside Office. Organisations standardising on one tool should take that one; organisations running several often take two.
Make it fit your team
Shape this curriculum around your work.
Every session is adapted before delivery — to your tools, your data constraints, and the tasks your team actually does. Tell us the context and we will come back with a scoped outline.
Our practitioners’ training experience
Platforms our practitioners teach on
Udemy
Coursera
CodecademyOther curricula
Agent Harness Engineering
Teaches teams to evaluate and operate agent harnesses as production infrastructure, covering tool design, permissions, sandboxing, and observability.
AI Research Methods for Nonprofits
Teaches nonprofit staff to evaluate AI tools and claims rigorously on a constrained budget, rather than to build models.
AI Safety and Guardrails for Healthcare Teams
Teaches healthcare teams to find and contain AI failure modes before deployment, with escalation paths designed around clinical risk rather than model accuracy.
Claude for Work
Teaches non-engineering teams to use Claude for repeatable work — projects, long documents, and shared workflows — rather than one-off chat prompts.