How to Master the "Prompting" Skill You Already Have

From SLPs Talk Tech Live, September 10, 2025 -

Here’s a topic that's both new and familiar: AI prompting. We all know what "prompting" means in our clinical practice as speech-language pathologists, eliciting a response or guiding a client with support. While the term 'prompting' has multiple meanings across different clinical contexts (i.e., behavioral techniques, EMR systems, and therapy approaches), in the world of AI, it's the art of crafting effective commands to get the best, most relevant outputs from a model. As speech-language pathologists (SLPs), we are uniquely equipped to excel at this. Our expertise in structured questioning combined with our deep domain knowledge gives us a powerful advantage over the average user.

SLPs are likely to approach an AI prompt differently than professionals from other disciplines, because of our expertise in prompting other humans to elicit a specific skill. Therefore, we wanted to revisit prompting with three specific techniques: RTF, zero-shot prompting, and meta-prompts. While prompting AI isn’t drawing the same spotlight it did in 2024, it remains a valuable competency for SLPs who want to build proficiency in the use of generative AI tools.

Your Clinical Skills Are Your Superpower

As SLPs, you are natural prompters. The same skills you use to elicit language and guide clients through therapy are directly transferable to working with AI. You understand how to use specific, structured questions and provide context to get a desired outcome. This "expert's advantage" allows you to quickly spot inaccuracies or biases in AI-generated content, a critical skill given that these tools can and do make mistakes.

SLPs are well-positioned to become proficient at AI prompting as they already have expertise in prompting to elicit language.

This meeting, we didn't just discuss what to ask, but how to ask, providing a strategic framework for crafting effective commands to get the best possible outputs.

1. The RTF Framework: A Prompting Recipe for Success

For clinicians new to AI, the Role, Task, Format (RTF) framework is a great starting point. It provides a simple structure for your prompts, helping you get what you need. By defining these three elements, you can guide the AI to perform a specific function and deliver the output in a usable way.

• Role: Specify the persona you want the AI to adopt (e.g., "Act as an SLP," or "Pretend you are an AAC expert"). This provides context for the AI's response.

• Task: Clearly state what you want the AI to do (e.g., "Generate a list of 24 core vocabulary words that I can target every week starting on Wednesday September 16, 2025," "Help me draft a session summary template for a client working on disfluency strategies," or "Write 10 simple sentences using the word “shell” ").

• Format: Tell the AI how you want the information presented (e.g., "Create a table for data collection" or "Provide a numbered list with checkboxes").

Actionable Advice: The next time you use an AI tool, try to phrase your prompt using the RTF framework. It's a great way to build confidence and get over the initial "imposter syndrome hump" that many people feel when they start using these tools.

2. Zero-Shot Prompting: The "No-Guidance" Approach

Zero-shot prompting is the simplest form of interaction. You ask a direct question or give a command without providing any examples or specific demonstrations of the desired output. The AI relies entirely on its pre-trained knowledge to generate a response. For example, a simple zero-shot prompt would be, "What is the capital of France?"Our meeting highlighted the significant risks of relying solely on this method. Even with a little bit of context, the output across platforms can vary significantly.

Actionable Advice: Use zero-shot prompts for basic, factual, and low-stakes tasks, but never for clinical recommendations or client-facing content. For anything that requires accuracy, context, or professional nuance, you'll need to provide more guidance.

3. Meta-Prompting: The "Prompt about the Prompt"

Meta-prompting is an advanced technique where you use an AI to help you create a better prompt. Instead of just asking for an answer, you are asking the AI to help you design the best possible question. You can even ask AI what information it needs to create a better RTF prompt. This approach is powerful because it leverages the AI's internal logic and understanding of “language” to create a prompt that is more likely to yield a high-quality, relevant response.

"You are a speech-language pathologist working in a preschool with children who are minimally speaking and non-speaking. None of these children are able to meet their needs with spoken language. Many of them have significantly reduced comprehension of spoken language and use physical communication to get what they need. Generate a prompt for therapy activities based on the children’s unique interests. Check for what you missed, ensuring the prompt outline meets cultural and linguistic diversity standards and ethical best practice recommendations by the American Speech-Language Hearing Association (ASHA)." 
- Meta-prompt example

Actionable Advice: When faced with a complex task, consider a meta-prompt. Also, use a meta-prompt for something you are an expert in to observe bias and errors in the AI output. Use a two-step process: First, ask the AI to generate the prompt, and then, in a new chat, use that new prompt to generate the actual output. This helps you refine your requests and get more precise, actionable results. Cross check this output which a real expert in your network and/or a published text.

Not All AI Models Are Created Equal - Even with the Same Prompt

To illustrate the pervasive differences in output across models (e.g., Chat GPT, Claude, Gemini), access tiers (e.g., pro vs. free), and user history, we discussed two cross-platform experiments that compared results from the same prompt.

• ASHA 2024 poster by Drs. Houston Sanders, Howard Goldstein, and Lindsey Peters-Sanders titled The Potential Role of AI in Planning Vocabulary Instruction for Young Children

• AI for SLP’s Facebook group demo from September 4, 2025

These experiments, completed nearly a year apart, revealed that using a simple prompt to generate vocabulary words for instruction yielded wildly inconsistent results across different AI models. This variability raises serious reliability concerns for clinical use, as you cannot be sure of the information’s accuracy or quality.

"I am working with a kindergarten student who has language-based communication deficits. I am trying to enhance their word learning and vocabulary knowledge. Help me make a word list of 20 words that begin with the letter S that are related to the book “the very hungry caterpillar" - The prompt example from the AI for SLP’s Facebook group 

Actionable Advice: Don't rely on one platform. Use different AI tools to cross-reference and compare outputs for the same prompt. This simple step will help you develop a gauge for AI output consistency and reliability, and build your confidence in using these tools for clinical practice. Never copy and paste outputs without a thorough review. Verify all information, correct any biases or inaccuracies, and remember that you are the final authority.

"Human expertise is crucial for critically evaluating outputs, advocating for human-centered design, and creating safeguards against over-reliance."

Key Takeaways:

• SLPs are well-equipped to prompt AI.

• Three prompting approaches were discussed: RTF (Role, Task, Format), zero-shot prompting, and meta-prompting.

• AI outputs show significant variability.

• Human expertise is non-negotiable.

Insights

While AI offers powerful tools, its outputs are not reliable enough for direct clinical use without human oversight, we underscore that AI is best used for augmentation, not replacement. As SLPs, our existing expertise in prompting gives us a unique advantage in using AI effectively. However, AI outputs can be unreliable and inconsistent across different platforms, making human oversight non-negotiable. It’s important to develop a gauge for AI output consistency and reliability, and to build your confidence in using these tools for clinical practice while being aware of their limitations and biases.

Practical Applications

We also invite you to take on the challenge: Try comparing the same prompts across different AI platforms to observe the variations in responses. Share what you find with our community. The more we learn together, the better we can harness this technology to benefit our practice and, most importantly, our clients.

Take Your Next Step:

📅 Want to see these tools in action? Request a tools demo, a guided walk-through of the best AI tools for CSD professionals — no guesswork, just impact.

👩🏽‍💻 Join our next live meeting September 24, 2025 - Sign up here on ZOOM.

• September 24, 2025 - Firouza Eshonova from EZspeech

  • An interactive therapy platform, EZSpeech has created AI-powered assessments, personalized guidance, and real time feedback for speech therapy exercises.

• October 8, 2025 - Leila Denna, MS, CCC-SLP, Tools Demo: QuickPic AAC

  • Developed by Dr. Howard Shane, Christina Yu MS, CCC-SLP, and Mauricio Fontana de Vargas, this AI-driven mobile app creates personalized communication boards in seconds.

• October 22, 2025 - Julia Franklin, MS, CCC-SLP from Cephable

  • Cephable has created alternative access solutions that work across devices, with accessibility at the core of their work.

📝 Share this article with a colleague. Let’s keep growing our SLPs Talk Tech network!

Share SLPs Talk Tech