chatgpt-personalisation-and-privacy
ITIAN ChatGPT Academy
Technology Simplified — Solutions That Work
Personalisation and Privacy
Make ChatGPT more useful without sharing more than necessary. Learn how personalisation sources, Data Controls, connected tools, shared links and Temporary Chat affect your privacy decisions.
Learning Outcomes
By the end of this lesson, you should be able to:
Minimise Data
Provide only the information genuinely needed for a useful response.
Separate the Controls
Distinguish personalisation, chat history, model improvement, connected tools and sharing.
Assess Destinations
Understand when information could reach memories, files, apps, actions, shared-link viewers or other people.
Choose Safely
Select a normal chat, Project, Temporary Chat, anonymised prompt or an approved non-AI system based on risk and authorisation.
Privacy begins before you type
The safest information is often the information you never enter. Before sharing context, ask whether ChatGPT actually needs it, whether you are authorised to disclose it, whether it can be anonymised and which tools or people may receive it.
Personalisation should reduce repetition—not increase unnecessary exposure.
The ITIAN Privacy-by-Design Principles
Purpose
Know exactly why the information is needed.
Minimum
Use the least detail that can achieve the outcome.
Authority
Share only information you are allowed to disclose.
Destination
Know which service, person or third party may receive it.
Control
Review settings, retention, removal and sharing regularly.
Five Controls People Commonly Confuse
| Control | What it governs | What it does not automatically do |
|---|---|---|
| Memory and personalisation | Whether relevant personal context can shape future responses | Does not decide whether eligible content improves models. |
| Chat history | Whether ordinary conversations remain available in your history | Deleting a chat does not necessarily delete a separately saved memory. |
| Improve the model for everyone | Whether eligible consumer conversations help improve models | Turning it off does not remove ordinary chats from history. |
| Temporary Chat | A conversation that is absent from history and does not create or use personalisation memories | Does not disable enabled Custom Instructions or all safety-related processing. |
| Shared link | Creates a viewable conversation link for others | Deleting the link cannot remove copies someone already imported. |
Information Risk Levels
Lower Risk
General questions, fictional examples, public facts and broad preferences that you are comfortable using for personalisation.
Still check accuracy and avoid unnecessary identifying detail.
Needs Care
Private plans, unpublished work, personal circumstances, internal procedures or information that could identify another person.
Anonymise, minimise and confirm authorisation.
Do Not Enter
Passwords, authentication codes, private keys, payment credentials, confidential records, highly sensitive identifiers or information you are prohibited from sharing.
High-Stakes Decisions
Health, legal, financial, employment and safety information needs qualified human oversight, approved systems and careful data handling.
Interactive Privacy Decision Guide
Use a general label only—do not enter the actual information. This tool runs locally and provides educational guidance, not legal or regulatory advice.
Memory and Personalisation Privacy
Review What Is Used
Inspect the Memory Summary, saved memories, Custom Instructions and available personalisation source indicators.
Correct and Remove
Update inaccurate information and delete it from every source when full removal is required, including chats, files and connected apps.
Choose the Smallest Scope
A one-off detail belongs in the current prompt. Project information belongs in that Project. Only stable, suitable details belong in persistent personalisation.
Expect Variation
Memory features and sources vary by plan, region, device and rollout. Use the controls present on your account and check official guidance.
Model Improvement and Data Controls
Open Data Controls
On web or desktop, open your profile → Settings → Data Controls. On mobile, open the sidebar, profile and Data Controls.
Review the Training Setting
“Improve the model for everyone” controls whether eligible consumer conversations help improve models.
Choose Your Preference
Turn the setting off if you do not want eligible conversations used for model improvement. The choice syncs across devices.
Understand History
Ordinary conversations can remain in history when model improvement is off. These are separate controls.
Know Workspace Differences
OpenAI states that ChatGPT Business, Enterprise and Edu content is not used for training by default. Organisational policies may add controls.
Review Regularly
Re-check after account, plan, workplace or product changes.
Temporary Chat: Useful, Not Magical
Privacy Benefits
Temporary Chats do not appear in normal history, do not create or use personalisation memories and are not used to improve models.
Documented Retention
OpenAI may retain a copy for up to 30 days for safety. Temporary does not mean “never processed” or “safe for secrets”.
Custom Instructions Remain
Enabled Custom Instructions still apply. Check whether they include context you do not want involved in the task.
Third Parties Remain Separate
If a GPT or tool sends information through an action, the receiving party’s privacy policy applies and retention may differ.
Connected Apps, Plugins and Actions
| Before using a connection | Question to ask | Safer practice |
|---|---|---|
| Trust | Who operates the service and what privacy terms apply? | Use only reviewed and trusted tools. |
| Permission | What data can it read or change? | Choose the minimum required permissions and confirm write actions. |
| Transfer | Which part of my prompt, files or instructions may be sent? | Remove unrelated personal and confidential context. |
| Retention | How long can the third party retain the information? | Check its privacy policy before authorising access. |
| Revocation | How do I disconnect it and remove stored data? | Record the approved connection and review it periodically. |
Custom Instructions can matter here
OpenAI’s guidance states that relevant information from Custom Instructions may be provided to third-party plugin developers when those tools are used. Keep persistent instructions minimal and avoid content you would not want shared with a connected provider.
Shared Links
Anyone with the Link Can View
Shared links do not provide granular access permissions. A recipient can forward the link to someone else.
Review the Whole Snapshot
A shared link can include the conversation up to the point of sharing. Inspect earlier prompts and responses, not only the final answer.
Deleting Has Limits
Deleting the link or original chat stops access through that link, but cannot remove a copy another user already imported into their history.
Best Practice
Create a clean, separate conversation containing only material approved for sharing. Never share a chat containing sensitive information.
A Safer Prompting Transformation
| Risky version | Privacy-preserving transformation |
|---|---|
| Real customer name, address and account number | “A customer” plus only the non-identifying facts needed to draft a general response. |
| Full private medical record | A general symptom or workflow question without identifiers; consult a qualified professional for decisions. |
| Unpublished employee performance details | A fictional scenario focused on communication principles, following workplace policy. |
| Real API key inside a debugging prompt | A placeholder such as YOUR_API_KEY; rotate any key that was exposed. |
| Complete confidential contract | Authorised, anonymised excerpts or a general clause type; obtain professional legal review. |
Common Privacy Mistakes
Assuming “Private Chat” Means Secret
No online AI conversation should be treated as a password vault or approved confidential-record system.
Confusing Settings
Turning model improvement off does not delete history; turning Memory off does not delete existing memories.
Forgetting Earlier Messages
A shared conversation can expose prior context. Review the full snapshot before sharing.
Sharing Someone Else’s Data
Possessing information does not mean you are authorised to disclose it. Check consent, policy and legal requirements.
Recommended: 16:9 • Captioned • Settings and safe examples
Future Guided Demonstration
The video will use fictional information to demonstrate:
- Reviewing Memory and Data Controls.
- Anonymising a risky prompt.
- Inspecting a shared-link snapshot.
- Reviewing app permissions and Temporary Chat.
Practical Activity: Personalisation Privacy Audit
Complete this review without copying private details into the page.
0 of 12 completed — begin by reviewing personalisation sources.
Knowledge Check
Answer all five questions, then check your result.
Official Sources and Further Reading
Reviewed against current first-party OpenAI guidance on 13 July 2026.
OpenAI guidance used in this lesson
- Data Controls FAQ — model improvement, account controls and Temporary Chat.
- Memory FAQ — personalisation sources, connected context and removal.
- ChatGPT Custom Instructions — data use and third-party plugin considerations.
- ChatGPT Shared Links FAQ — access, snapshots, deletion and imported copies.
- Temporary Chat FAQ — memory, history, retention and third-party actions.
Lesson Summary
Seven ideas to remember
- Share the minimum information needed for a clear purpose.
- Confirm authority before entering another person’s or organisation’s information.
- Memory, history, model improvement and shared links are separate controls.
- Temporary Chat reduces personalisation and history use but is not a secret vault.
- Connected apps and actions can introduce third-party privacy terms.
- Anyone with a shared link can view and forward the linked conversation.
- Review personalisation sources, Data Controls, apps and shared links regularly.