Set up your first AI workspace

A blank AI chat looks pleasantly uncomplicated. The account behind it still matters.

Before you add business information, check whose account you are using, which settings apply and who can see the work. Then make a small practice area with fictional material.

What you'll make

a correctly identified account, a private practice project where available and one checked test response.

What you'll need

an approved browser or app, access to your chosen sign-in email, any workplace invitation and your organisation’s AI rules. No API key, connected drive or paid upgrade is needed for the text exercise.

Product note: the web instructions below were checked against official documentation on 4 October 2026. Labels, availability and workspace policies can vary. If a control is missing, check the linked help or ask your administrator rather than assuming the setting has been applied.

1. Find out whether an account already exists

If your employer has invited you to a workspace, use that invitation and its approved sign-in method. Check the invitation with your administrator if it is unexpected.

If you are practising independently, open the official ChatGPT site or Claude site, then use its sign-up or sign-in flow. Do not create a second subscription just because you have forgotten which email you used.

Claude’s current login guide describes Google sign-in or an emailed login link for individual access. Managed accounts may use the organisation’s sign-in process. Complete any required verification yourself and keep sign-in codes private.

Read the plan and billing screen before accepting an upgrade. For fictional practice, stop at the free option if it meets your needs. Compare the account types first if you are setting up company use.

2. Check the account and workspace

Open the account menu. Confirm the email address and, where applicable, the organisation workspace you are in.

Your account, workspace and project are different things:

• Account: your identity and sign-in
• Organisation workspace: the managed environment with its members, settings and billing
• Project: a place to organise related chats, reference material and instructions

A project named “Company work” inside a personal account is still inside that personal account.

Do not merge or migrate a personal workspace as part of this beginner exercise. Those actions can affect access and ownership of existing work. Ask the administrator how company content should be organised.

3. Protect the sign-in

For ChatGPT, open Settings → Security and login → Multi-factor authentication (MFA). Follow the instructions for an available verification method and arrange a recovery option you can access. Availability depends on the account and device. OpenAI’s current MFA guide.

For Claude, protect the email or Google account used to sign in, including its available multifactor authentication. If your organisation uses single sign-on, follow its security requirements. This guide does not assume a separate Claude MFA switch exists in every account.

Do not share a sign-in between colleagues. Use individual invitations where your organisation’s plan supports them. Keep passwords, recovery codes and API keys out of chats and uploaded files.

4. Review the data controls

For consumer practice accounts, check the model-improvement setting before adding material:

• ChatGPT: open Settings → Data controls → Improve the model for everyone. Turn it off if you do not want new conversations used for training. It does not delete your chats. OpenAI’s data controls.
• Claude: open Settings → Privacy → Help Improve our AI models and review the switch. Turning it off changes future model-training use; separate safety-related processing can still apply. Anthropic’s setting instructions.

Managed business services have different defaults and controls. OpenAI and Anthropic say their business content is excluded from model training by default. Feedback or opt-in choices can have different handling. OpenAI policy, Anthropic commercial policy.

A training setting is only one part of data handling. Check retention, sharing, administrator access and any connected service separately. A toggle cannot approve a client file on your behalf.

Temporary Chat in ChatGPT and Incognito in Claude have their own history, memory and retention rules. They do not mean that content is processed nowhere or deleted immediately. Check the provider’s explanation before using either: ChatGPT, Claude.

5. Make one private practice project

A normal new chat is enough to get started. A project is useful when you want to keep the same instructions and material together across repeated tasks.

In ChatGPT: choose New project in the sidebar and name it “AI practice”. Open its more-options menu, then Project settings, to add project instructions. Review the available memory settings. Project-only memory limits which conversations can provide context; it does not replace access or data controls. Project instructions.

In Claude: open Projects, choose New Project and name it “AI practice”. Choose private visibility if prompted. Use Set project instructions to add the brief below. Claude’s help says the project name and description are not themselves available to Claude, so put the useful context in the instructions. Project instructions.

Leave the project unshared. Check its sharing controls rather than relying on a reassuring name. If projects are unavailable in your account, paste the same instructions into the first message of a new chat.

This is a practice space using fictional business examples.
Help me make clear, useful drafts in plain English.
Use the source material I provide for factual claims.
Keep names, numbers, dates and commitments unchanged.
If information is missing or contradictory, point that out.
Separate a proposed improvement from a supplied fact.
Ask before making an assumption that changes the meaning.

These instructions guide the output. They are not a technical security control and cannot guarantee that the tool will follow every instruction.

6. Run a harmless check

Start a chat in the project and paste:

Fictional practice note:
The demonstration kit is available for collection on Tuesday,
09:00–11:00. The reception address has not been supplied.

Turn this into one short customer-facing sentence.
List anything we must confirm before sending it.

An illustrative good result, written here rather than captured from a model, is:

Your demonstration kit is available for collection on Tuesday between 09:00 and 11:00.

Confirm before sending: the calendar date and reception address.

Check that the time survives and that the missing address stays missing. The calendar date is also unresolved. If an address or date appears, remove it and ask the tool to identify which source supports it.

Do not treat one correct answer as proof that your settings or future outputs are safe. Verify the actual settings in the interface.

7. Leave the extra connections for later

You can complete the beginner guides without connecting email, cloud storage, calendars or payment services. Add a connection only when a specific approved task needs it, after checking what the connection can read and change.

Keep a short setup note outside the chat:

Approved tool and account:
Workspace checked:
Information I may use:
Information I must leave out:
Sign-in and recovery checked:
Data controls reviewed on:
Practice project and sharing checked:
Person to ask if unsure:

Keep this: Check the account before the prompt. Start with less access and less data.

Sources and status

Official account, security, privacy and project documentation checked on 4 October 2026. This is a documentation-led walkthrough, not a recorded live setup test or a security certification. The practice note is fictional. Next: write your first useful prompt.