In a November 10, 2026 announcement, OpenAI described free ChatGPT workshops for family caregivers in five US cities, covering schedules, educational material and finding support. Those everyday jobs offer a useful starting point for families considering AI: choose something specific that can be checked before anyone relies on it.
The Oakland workshop listing explicitly excludes diagnosis, treatment and medical recommendations. This is training in administrative uses, not a new clinical service. The announcement does not establish that using AI saves caregivers time or improves care.
Caregiver Action Network, a workshop partner, launched its Humanity Baseline on October 15. It describes the framework as guidance, not product certification or an endorsement of a particular tool. A practical way to apply it is to choose the task and its final check together.
Draft a schedule when the arrangements are already agreed
Use AI to draft a family schedule from agreed arrangements, then check it before anyone acts. This fits the workshop’s schedule-and-checklist examples: the job is organizing information, not deciding what care someone needs or who must provide it.
For example, a hypothetical request could be: “Turn these agreed visits into a weekly list. Use Person A and Person B. Leave gaps marked unassigned. Do not invent availability or change any times.”
Compare every day, time and assignment with the original messages. Resolve an unfilled visit with the people involved; a neatly completed schedule is not evidence that somebody agreed to it. Keep medication timing and other clinical decisions outside this exercise.
Writing a message and sending it are separate decisions. CAN’s framework calls for authorization before an external action. Keep the result as a draft until someone with authority has checked its contents and recipients; getting writing help does not require connecting a messaging account.
Simplify a handout without replacing the original
Turning trusted educational resources into plainer language is another stated workshop use. Start with a current public handout from a health agency or condition-specific organization, or material supplied by the relevant professional—not an unexplained answer generated from scratch.
Ask for a shorter explanation that retains the source and date. Compare it with the original, especially exceptions, warnings and instructions about seeking help. If you cannot tell whether the meaning survived, ask the source or a qualified professional to explain it.
Keep the original accessible. This is a readability task, not a way to interpret an individual’s test results or choose treatment. For the broader distinction, RohitAI’s guide to saved AI memory and original documents explains why reusable notes should not replace the records behind them.
Find a service, then confirm it with the provider
A resource search should end with an organization you can contact, not a chatbot’s promise of help. Give a broad location and the type of support needed, rather than a full personal history. Ask for original provider links, then confirm the service area, eligibility, cost and current availability directly.
For older adults and their families in the United States, the Eldercare Locator offers a direct alternative: location-based service searches and trained staff available by phone, text or online chat. It is a public service of the Administration for Community Living. You can start there without asking AI to assemble a list.
Elsewhere, start with the relevant local public agency or established support organization. An unverifiable contact is not a useful lead, and a shortlist is neither a booking nor confirmation that someone qualifies. Urgent needs should go directly to the appropriate human service.
Permission comes before personal information
CAN advises sharing another person’s information only with appropriate permission or authority, while respecting their choices. Being a family caregiver does not settle that question. If authority is unclear or relatives disagree, resolve it through the appropriate human channels—not a chatbot.
For a schedule, use placeholders where possible. For an explanation, use a public handout instead of uploading a personal record. Changing a privacy setting does not supply permission to share someone else’s details.
For ordinary personal ChatGPT use, distinguish three controls:
Model improvement: turning off “Improve the model for everyone” lets you keep chat history while opting out of model-improvement sharing. It does not delete saved chats.
Temporary Chat: while temporary, a conversation stays out of history, is not used for model improvement and does not create or update memories. OpenAI may retain a copy for up to 30 days for safety.
Personalization: a personalized temporary chat can use existing memories. Choose Unpersonalized before starting if you do not want memory-based personalization. Saving a temporary chat converts it into a regular chat governed by your account settings.
These controls are not a blanket confidentiality guarantee. Third-party actions follow the recipient’s privacy policy; managed accounts have their own rules. This guide concerns ordinary text tasks, not specialized services or health-record integrations.
Count the work after the answer appears
As of November 10, 2026, ChatGPT’s Free plan includes text chat, with limits on uploads and other features. A simple text draft does not inherently require a subscription. OpenAI lists Plus at $20 a month; check local checkout terms and account limits before paying.
CAN’s workload guidance includes learning, setup, data entry, corrections and ongoing management—not just the subscription fee. For a practical comparison, choose the same period and the same completed task: compare your existing process with preparing inputs, generating the draft, checking it, correcting it and keeping it current. This is a decision aid, not a measured saving.
If your calendar already works, or checking an AI draft takes more effort than doing the job yourself, keep the existing process. If one recurring administrative job is awkward, start with that job alone. Expand only if the checked result is useful, the information can be shared appropriately and the people involved want to continue.