Pause before you paste
community data into ChatGPT.
Do not put identifiable donor, beneficiary, case-management, health or safeguarding information into a public ChatGPT account. The OAIC recommends, as best practice, that organisations avoid entering personal information, especially sensitive information, into publicly available generative AI tools. Use public, synthetic or properly de-identified information, an organisation-approved service and account, a defined purpose, minimum necessary data, and human review.
ChatGPT
Microsoft Copilot
Claude
GeminiFor Australian organisations
For fundraising, communications, service delivery, grants, volunteer and operations teams using ChatGPT or another general-purpose AI assistant.
What the key terms mean
Donor data
Information linked to a donor or prospect, including identity, contact details, giving history, preferences and notes.
Beneficiary data
Information about a person receiving, seeking or connected to a service, grant or program, including case and outcome information.
De-identified
Altered so a person is no longer identifiable or reasonably identifiable. Removing a name alone is not always enough.
Law and recommended practice
Legal position: privacy obligations depend on whether the Privacy Act applies, the information, the purpose and the disclosure. Recommended practice: the stop-and-check workflow below is deliberately conservative for community data and does not replace legal, privacy or safeguarding advice.
A practical implementation workflow
Use this sequence to turn the resource into an owned, reviewable process rather than a document that sits on a shared drive.
Stop and classify
Before opening ChatGPT, identify every person, field and file involved. If it contains personal, sensitive, confidential or safeguarding information, stop the public-account workflow.
Question: could someone reasonably identify a person?
Minimise or replace
Use public facts, synthetic examples or a short abstract that removes unnecessary details. Do not paste the source record merely because the desired output is brief.
Question: what is the minimum information needed?
Check tool and account
Confirm that the exact product, account tier, feature and integration are approved for this purpose and data class. Personal subscriptions are not a substitute for approval.
Question: what terms, retention and access apply here?
Prompt with boundaries
State the task and constraints without asking the model to infer facts. Never ask it to fabricate a quote, consent, impact story, diagnosis or eligibility reason.
Question: can the task be completed without real community data?
Review and record
A responsible person checks facts, tone, bias, privacy, accessibility and any action that follows. Record higher-risk approved uses according to policy.
Question: who owns the final output and decision?
Report mistakes quickly
If information was pasted into the wrong service, preserve the facts, stop further use and contact the privacy or security lead. Do not conceal or independently decide notification duties.
Question: who needs the tool, account, time, prompt and file details?
Green, amber and red prompt guide
Classify the source material before drafting. These examples illustrate a conservative default; your approved-tools register and risk assessment control the final decision.
Green: public or synthetic material
Usually suitable for an approved AI tool when the task is low impact and a person reviews the result.
- Rewrite a published event description for plain English.
- Brainstorm subject lines using a fictional campaign brief.
- Summarise a public ACNC or government page with source checking.
Amber: internal or de-identified material
Pause for owner approval and verify that de-identification is robust, the account is approved and re-identification risk is low.
- Analyse aggregated campaign results with small groups removed.
- Draft a generic service FAQ from an approved, de-identified issue list.
- Improve a report paragraph that contains no person-level detail.
Red: identifiable or sensitive community data
Do not use a public or unapproved generative AI tool. Move to an assessed workflow or complete the task without the data.
- Paste donor names, emails, giving history or relationship notes.
- Upload case notes, support plans, health or disability information.
- Ask AI to rank people for services, grants, safeguarding or hardship support.
If you already pasted it
Act promptly and follow the incident process. Fast internal reporting lets the organisation assess and contain the issue.
- Stop the session and do not paste more data.
- Record the tool, account, time, prompt, file and output without spreading the information further.
- Contact the nominated privacy or security lead and follow their containment instructions.
Common NFP tasks and safer alternatives
A useful output does not justify unnecessary disclosure. Redesign the task around the minimum data.
| Team request | Risky approach | Safer starting point |
|---|---|---|
| Write a donor thank-you | Paste the donor record and notes | Use an approved mail-merge template and public campaign context |
| Summarise a case for reporting | Upload the case file | Create an approved aggregate or de-identified summary outside public AI |
| Draft an impact story | Ask AI to invent a beneficiary narrative | Use consented facts, human interviews and documented editorial review |
| Triage service enquiries | Let AI rank named people | Use a documented human triage process and separately assess any AI support |
Give teams help at the moment of use
Aona can help an NFP see workforce AI use, guide people towards approved tools and apply data controls to prompts and files. It complements, but does not replace, privacy analysis, safeguarding judgement, staff training or human ownership of communications and services.
Discover ChatGPT and other AI services used across the workforce.
Guide staff towards approved tools and acceptable uses.
Apply controls to sensitive prompt and file content.
Retain evidence to improve policy and training.
Primary Australian sources
Use these official sources alongside advice specific to your organisation and sector.
OAIC guidance on commercially available AI
Official guidance on personal information in AI inputs and outputs, due diligence and best practice.
Open official sourceAustralian Privacy Principles
Official text covering use and disclosure, overseas disclosure, data quality and security.
Open official sourceACNC governance for good
Official guidance for Responsible People on charity governance and accountability.
Open official sourceFrequently asked questions
See and govern AI use
across your workforce
Aona helps not-for-profits discover AI use, guide teams to approved tools, protect sensitive information and bring evidence into governance reviews.
Continue through the NFP governance cluster
AI governance for NFPs
Create the accountability and controls behind this frontline guidance.
NFP AI policy template
Set clear organisation-wide prompt and data rules.
Charity AI risk assessment
Assess a proposed ChatGPT workflow before approval.
AI data loss prevention
See Aona's approach to prompt and file data controls.