Choosing Your Tools and Organizing Your Workspace

The best prompt in the world is a problem if it is typed into the wrong place. Which tool you use, under which account and with which settings, determines what happens to the information you give it. In this lesson you will learn to tell approved tools from personal accounts, to check what really matters in a tool's terms, and to set up a clean, shared prompt library.

You now have three target tasks. Before you try them, you need a safe place to do it. This is not a technical matter reserved for the IT department: every person who uses AI at work makes choices about tools and data several times a day.

Approved tools versus personal accounts

Marc, the plumber, is his own IT department. When he started using a free chatbot on his phone, he pasted a customer's address and phone number to prepare a message. Later he realized that he had not checked where this information goes, how long it is kept, or whether it can be reused. That is the central problem of a personal account used for professional work.

In an organization, the question is even more important. Companies and public bodies usually have rules on which tools can process work information, and often have signed contracts with providers to frame what happens to the data. A personal free account is outside that framework. If you use it for work content, you may be breaking an internal rule, a confidentiality clause or data protection law, even without intending to.

So the first rule is simple: use the tools your organization has approved. If there is none, ask. If you are self-employed, choose deliberately and read the terms before you paste real work data.

Consumer versus professional offers

Providers often have several offers: a free consumer service, a paid individual plan, and business or enterprise versions. Names and conditions change often, so never assume; check the current terms. These are the points that matter most.

What to check	Why it matters
Data retention	How long are your conversations and files stored, and can you delete them?
Reuse for training	Can your inputs be used to improve the provider's models? Can you opt out, and is it off by default?
Hosting and location	Where is the data processed and stored? This matters for personal data under European rules.
Contract and role of the provider	Is there a data processing agreement? Who is responsible for what?
Access and security	Who inside the provider can read the content? Is there single sign-on, administrator control, access logs?
Features that share data	Plug-ins, connected drives, shared links, browser extensions: each one widens the circle of access.

You do not have to become a lawyer. If you work in an organization, your IT, data protection or legal contact can check these points. If you work alone, read the privacy and business terms of the tool, look for the settings above, and keep it simple: when in doubt, leave personal or confidential data out.

Privacy settings to review on day one

• Check whether conversations are used to train models, and switch this off when possible.
• Check history settings and how to delete conversations and files.
• Review connected apps, drives and plug-ins and disable those you do not need.
• Use a strong, unique password and two-factor authentication on any AI account.
• Check the default sharing settings for links to conversations or files.
• Make sure the custom instructions or "memory" of the tool contain no sensitive details.

Data traffic light

A simple rule of thumb helps colleagues decide quickly. Your organization will define its own version, but here is a model you can start from.

Level	Examples	Rule
Green	Public information, generic questions, texts you wrote with no personal data	Any approved tool
Orange	Internal documents, non-public plans, anonymized customer cases	Only approved tools with suitable terms; anonymize where possible
Red	Personal data about identifiable people (especially sensitive data), health, finance details, credentials, trade secrets, anything covered by professional secrecy	Not in a tool that has not been specifically cleared for that purpose

Organizing your workspace

Once your tool is chosen, organize your work so that your good prompts do not vanish in an endless chat history.

A prompt library

Nadia notices that she rewrites a similar prompt every week. She creates a shared document with the prompts her small HR team uses: one entry per prompt, with a title, goal, the prompt itself with its placeholders in brackets, what must never be pasted, and how to check the result. Colleagues can add entries and suggest improvements.

A useful entry looks like this:

Title: Polite reply to unsuccessful candidates
Owner: [name]            Version: 1.2          Last review: [date]
Goal: Draft a respectful, short rejection message.
Tool: [approved tool]
Data allowed: job title and generic reason only. Never paste the candidate's CV or personal details.

Prompt:
You are an HR assistant. Write a short, kind and clear email informing a candidate that their application for [job title] was not selected.
Tone: respectful, no empty promises. Length: under 120 words. Do not give a detailed reason. Leave the greeting as [First name].

Check before use: tone, no false promises, consistent with our policy, no personal detail.
Known limits: sounds generic; add a personal touch by hand.

Naming and versions

Treat prompts like small documents. Give them a clear name ("email-reply-supplier-delay"), a version number, a date and an owner. When someone improves a prompt, increase the version and note what changed and why. If a result gets worse after a change, you can go back to the previous version. Keep a "retired" folder rather than deleting.

Folders, templates and context

• Group prompts by task type (writing, summarizing, analysis, customer relations), not by tool.
• Keep short, reusable "context blocks" (your company in three sentences, your tone of voice, your standard disclaimers), without confidential content.
• Save good outputs as examples to include in future prompts.
• Decide who can edit the shared library and how often it is reviewed.

A ready-to-use prompt: build your context block

Help me write a short "context block" that I can paste at the beginning of my prompts, so that the assistant understands my work without any confidential details.
Information, in general terms:
- My role: [job title]
- My organization in 2 or 3 sentences: [sector, size, type of customers]
- My writing style: [e.g. warm but concise, formal, plain language]
- Things to avoid: [e.g. jargon, promises on deadlines]

Produce a block of at most 100 words in the second person ("You are helping...") and tell me what I must NOT add to it (names, personal data, secrets).

Bad use versus good use

Bad use	Good use
Using a free personal account for client files "just this once".	Asking which tool is approved, or anonymizing the content and using only the approved tool.
Keeping your best prompts in a chat history you can't search.	Saving them in a library with title, version and checking notes.
Assuming the paid plan automatically protects your data.	Reading the terms on retention, training and hosting for that specific offer.

Your turn: set up your safe workspace

Exercise. (1) Find out which AI tools your organization allows, or, if you work alone, pick one tool and note where its terms say how data is kept and whether it is used for training. (2) Create your data traffic light with three real examples of your own data for each color. (3) Create the first entry of your prompt library using the template above, for one of your three chosen use cases.

See a solution

Julien checks with the partner in charge of IT: only one tool is approved, with a business contract. He notes: "Green: general explanation of a public tax rule. Orange: a client case with names removed. Red: any client file, tax number, bank detail, anything under professional secrecy, which never goes into the tool unless the partners explicitly allow it for that purpose."

His first library entry is "Reminder for missing documents": prompt with placeholders for the client's first name and the list of missing documents, an indication that no amounts or account numbers are ever added, a check (read it, compare the list with the file) and a known limit (the tone can be too formal, to be adapted by hand).