From the Speech Pathology Resources archive. View the original article →

Is it safe to use ChatGPT in clinical practice?

Suitability depends on the tool, account, information and purpose. Configuration matters, but cannot eliminate risks such as inaccurate output, data exposure or bias. Start with generic tasks and assess a proposed clinical workflow before using it.

Risks can come from the product itself as well as the way it is configured and used. Assess both.

Clinicians must actively configure and evaluate their AI tools.

 

What are common AI mistakes in healthcare?

The most common mistakes I am hearing about include:

  • Copying and pasting full reports into LLMs for wording improvements

  • Removing the client’s name but leaving contextual identifiers

  • Using personal AI accounts for professional tasks

  • Not checking whether model training is turned off

  • Having no AI consent or policy framework

Removing a name is not de-identification.

If someone could reasonably identify a client from age, diagnosis, location, schooling context, or rare circumstances, the risk remains.

What is re-identification risk?

Re-identification occurs when multiple small data points combine to make an individual identifiable.

Even if each detail seems harmless in isolation, together they may narrow identity significantly.

Clinicians must reduce information to what is necessary for the task.

 

What should I check in my LLM settings?

If using ChatGPT:

  1. Confirm model training is turned off

  2. Enable two-factor authentication

  3. Avoid using personal accounts for clinical workflows

  4. Understand how data is stored and processed

Tool evaluation should include privacy policies, storage location, compliance considerations, and appropriate use cases.

 

Do I need to be tech savvy to use AI safely?

No.

You need to be structured and informed. 

Ethical AI use requires:

  • Human-in-the-loop review

  • Clear de-identification workflows

  • Transparent staff practices

  • A documented AI consent and policy process

  • Understanding core AI terminology

 

How can I build a safe AI framework in my clinic?

Start with foundations:

  • Audit your current AI use

  • Document a de-identification workflow

  • Create an AI consent process

  • Train staff on appropriate use

  • Evaluate each tool before adoption

If you want structured guidance on professional setup, de-identification, privacy guardrails and tool evaluation, explore my AI Toolkit and upcoming live workshops.

Want More Practical AI Guidance?

If you are a speech pathologist wanting to use AI confidently and ethically, join my email list for weekly AI insights and practical frameworks designed specifically for allied health professionals.

https://www.speechpathologyresources.com.au/weekly-ai-tips 

 

 

 

 

Privacy settings are one part of the assessment

Turning off model training is not the same as deleting data or guaranteeing confidentiality. Review the provider’s retention, processing, access and overseas-disclosure arrangements, alongside consent and your service’s obligations. Avoid identifiable client information in public AI tools.

OAIC guidance on commercially available AI · De-identification guidance

← Back to articles