By Lisa A Young M.A. CCC-SLP

How to use ChatGPT for speech therapy in 2025

AI tools are becoming more useful and more popular, even within the field of speech pathology. The 2024 ASHA Convention had several AI-related presentations, and the ASHA Leader has published multiple articles on AI and specific tools like ChatGPT. We even have a systematic literature review published already! In this article, I’m sharing tips for how to use ChatGPT for speech therapy.

Free DIRECT downloads: ChatGPT prompting for SLPs (cheat sheet). (Email subscribers get free access to all the resources in the Free Subscription Library.)

Outline:

5 tips for prompting ChatGPT

Giving instructions to a large language model like ChatGPT is called prompting, and crafting good prompts is called prompt engineering. I’ve been learning about AI and prompt engineering for several years now, from people such as Joanna Penn, Elizabeth Ann West, Joseph Michael, and Jason Hamilton.

Just fyi, I tried but failed to write fiction with AI, so I’m writing the old-fashioned way. But I’ve found AI to be very helpful in many other ways!

These prompting techniques aren’t specific to ChatGPT, but it’s the large language model I use most often.

return to top

Traditional search engines simply try to find the best matches for our queries. We’re very comfortable typing keywords into that search bar, knowing we’re likely to find something useful. For example:

  • “Free articulation worksheets for /r/ sound.”
  • “Downloadable PDF for dysphagia exercises.”
  • “Best apps for AAC.”

AI tools based on large language models are much more sophisticated. It’s complicated and beyond the scope of this blog post, but ChatGPT does not search a database for answers. Here is part of ChatGPT’s answer to how it works:

ChatGPT… doesn’t just predict the next word–it also considers the context, structure, and intent of the entire input to produce coherent, contextually relevant, and often creative responses, … capable of crafting responses that simulate understanding and expertise.

ChatGPT 4o, generated 1/18/2025

We can use ChatGPT for:

  • Answering questions.
  • Brainstorming ideas.
  • Problem-solving.
  • Learning new skills.
  • Organizing and planning.
  • Creating images or visuals.

The next four tips will suggest useful ways to interact with ChatGPT.

return to top

2. Assign ChatGPT a persona

In the quote above, ChatGPT said that it “simulates expertise.” But how does it know what type of expertise to provide? For example, just imagine the different answers this question might get:

“What does a speech-language pathologist do?”

  • A patient’s wife asking an SLP on a hospital ICU floor.
  • A patient asking a general practitioner who is making the referral.
  • A roommate asking an SLP graduate student.
  • An eighth-grader asking a guidance counselor.
  • A Hollywood script writer asking a friend.
  • A young child asking their mother.

Each of those answers is going to be different! Each person asking the question has a different reason for asking and a different ability to understand the answer. Each person answering the question has different knowledge.

ChatGPT will give different answers based on what it knows about the person asking. So we can get better answers if we ask ChatGPT to take on a particular persona. (Michael, n.d.)

For example, here are 7 personas that an SLP may want ChatGPT to adopt:

  • “Act like you are an SLP preparing materials for a session and create….”
  • “Pretend you are a linguist and make a list of words such that…”
  • “Assume the role of a neurologist and explain…”
  • “You are an AAC expert. Please give me step-by-step instructions for…”
  • “Let’s say you are an expert in creating plain language materials. Please list ten ways I might explain [insert a medical concept] to patients and families.”
  • “Act like a utilization management specialist working for a large health insurance company. Please help me word this…”
  • “Imagine that you are an IT expert and explain to me how I can…”

return to top

3. Be specific

As SLPs, we know we get better results when we use specific instructions with our patients. If we’re vague, our patients may misinterpret our intent. Same thing goes for ChatGPT. (Coursera, 2025.; FreySoft, n.d.)

  • Use clear, specific vocabulary. Tell it who you are, what you’re looking for, and why.
  • Avoid ambiguity. If you want 3-syllable words, ask for “exactly 3 syllables”, otherwise you may get words that have “at least 3 syllables.”
  • Be clear what format you’re looking for. For example, a numbered list? A table? A line-drawing? A photo-realistic image?

return to top

4. Give examples

We can also improve our results by including examples in our prompt. (Coursera, 2025.; FreySoft, n.d.)

For example:

  • “Give me a list of household tasks an older individual could do while sitting down. Another adult will be able to set them up at the table. For instance, sorting mail or dusting knickknacks.” 
  • “What are some things an older individual could do while sitting down for leisure? Common examples might be a crossword puzzle or solitaire. Could you make a list of less common activities?”

return to top

5. Treat ChatGPT like an assistant

People who work with AI tools have learned that the results generally improve by having a dialogue with the tool. If the first response isn’t what you were looking for, don’t simply accept it or click away in frustration.

Since ChatGPT is software running on human language, we can get better results by following the conventions we’d use in any conversation. Since humans are more helpful when we’re nice to them, so are AI tools!

And since we are the human and ChatGPT is the software, we’re the boss. Elizabeth Ann West likes to say that AI is our junior [writing] partner, meaning that we have to take the lead and direct the tool to get the results we want.

So treat ChatGPT like an assistant. Remind it of background material that is important. This can be as easy as giving it a persona. Provide a context by telling it who you are and why you’re asking. Be specific with your instructions and desired output format.

return to top

Ethical concerns with AI tools

As you can imagine, ethical use of AI is of primary concern for SLPs. This section draws on the handouts from three ASHA Convention presentations (Mamlekar, 2024; Ramachandra, 2024; Rodrigues, 2024). There are three primary ethical issues that clinical SLPs should be concerned about:

  • Privacy and security.
  • Accuracy and completeness.
  • Bias and discrimination.

It’s important to read the terms of use for any piece of software we use. What information, if any, is collected? How is it used? How do you opt out of having your contents help train the model? Per my last read of OpenAI’s Terms of Use, you can opt out of having your content being used to help train the model.

Next comes concerns about accuracy and completeness. While AI tools develop at a very rapid pace, they still make mistakes. In addition, they may misunderstand your goal and only provide superficial or partial answers.

Finally, it’s important to remember that tools like ChatGPT learn from human language. If you look at books, movies, or newspapers from 50 years ago, you’ll probably find examples of bias or stereotypes that will make you cringe. So we need carefully review any content generated by ChatGPT.

But even beyond these obvious examples, as SLPs, we need to be aware that most media don’t include people with conditions we often see in our work. This means that ChatGPT might assume everyone communicates and behaves in “typical” ways. It might produce responses that aren’t inclusive or respectful.

return to top

Tips for using AI ethically

Ethical use of AI is a big topic, but here are four important considerations for SLPs. I’m drawing these points from the handouts from three presentations at the 2024 ASHA Convention. (Mamlekar, 2024; Ramachandra, 2024; Rodrigues, 2024)

  • Understand the benefits, risks, and limitations of any tool you use.
  • If you suggest a tool to patients, ensure they understand the benefits, risks, and limitations.
  • Check the accuracy of any responses that could affect the patient (such as using AI to analyze a language sample or asking for therapy approaches).
  • Use critical thinking and professional judgment.

return to top

Case example: Personalized articulation materials

When you interact with AI, you are providing a prompt that it responds to. The better your prompt, the better the outcome.

I gave several tips above, but let’s look at an example. Let’s imagine that we have an adult patient with dysarthria who has difficulty articulating /s/, and we don’t want to copy a list of irrelevant words from a workbook. Let’s say our patient is interested in houseplants and cooking. We may want to create a personalized list of words or sentences for practice.

*We can use ChatGPT to create images with DALL-E, and there are AI programs that create videos and songs!

return to top

My basic prompt

I opened ChatGPT using a private browser (so I can pretend to be a new user), and I’m using the free version (ChatGPT 4o mini). Here is the text I typed in as my first prompt:

Please generate a list of 10 words beginning with “s”.

That’s a simple, straightforward prompt, but I bet you can already see a problem with it!

Here is ChatGPT’s response:

Here’s a list of 10 words beginning with “s”:

  1. Sunshine
  2. Serene
  3. Sculpt
  4. Swift
  5. Symphony
  6. Sincere
  7. Summit
  8. Sparkle
  9. Silhouette
  10. Strategy

So , here we have a list of words. Boring? Maybe, maybe not. But certainly not personalized.

return to top

Personalizing the prompt

So let’s try a different prompt right in the same chat window:

I’m a speech-language pathologist working with an adult patient who has dysarthria. They have difficulty making the “s” sound and they would like to focus on improving. They are interested in houseplants and cooking. Please create a list of 10 interesting words beginning with “s”.

You can see that I’m much more specific, without giving any information away about my patient. Here is the response:

Certainly! Here’s a list of 10 interesting “s” words that might engage your patient, considering their interests in houseplants and cooking:

  1. Sprout (related to plants)
  2. Spice (in cooking)
  3. Succulent (a type of houseplant)
  4. Sauté (a cooking technique)
  5. Sear (a cooking method)
  6. Sizzle (a sound associated with cooking)
  7. Stem (part of a plant)
  8. Stir-fry (a cooking method)
  9. Soil (important for houseplants)
  10. Savor (related to taste and cooking)

These words are not only relevant to their hobbies but also provide opportunities for articulation practice in a fun and meaningful way.

Okay, that’s better, right? But we want to copy and paste this list into a document to print out. Who has time to delete all of the explanations in parentheses?

return to top

Asking ChatGPT to change the format

Let’s just ask it to delete those explanations.

Looking at that list of 10 words, I see you included an explanation for why you picked it. That’s great, but I’d like to print out this list, and the patient doesn’t need the explanations. Could you rewrite the word list without the explanations?

Did you notice my social nicety of “that’s great”? Remember, AI tools give better results when we are nice to them! Not because they have emotions or feelings, but because people are more helpful when others are nice.

Anyway, here is ChatGPT’s response:

Certainly! Here’s the list of 10 words without the explanations:

  1. Sprout
  2. Spice
  3. Succulent
  4. Sauté
  5. Sear
  6. Sizzle
  7. Stem
  8. Stir-fry
  9. Soil
  10. Savor

This version is ready to be printed for your patient. Let me know if you’d like further adjustments!

That’s a better word list for this patient, right? We could make it even better if we give ChatGPT some more information. What sorts of food does this patient like to make? What types of houseplants do they like?

But let’s take things a step further, in terms of complexity.

return to top

Increasing the complexity to the sentence level

Let’s say we want some sentences with words that contain /s/ and /z/. Here is my prompt:

That’s great, thanks! Now please generate a list of 10 sentences that contain several instances of the “s” and “z” sounds for this same person.

And here is the response:

You’re very welcome! Here are 10 sentences containing multiple instances of the “s” and “z” sounds for your patient to practice:

  1. The sizzling sauce smells so sweet as it slowly simmers.
  2. Sarah’s succulent strawberries surprised us with their zesty flavor.
  3. The soft soil settled around the small saplings in the garden.
  4. Zoe stirred the spicy sauce, savoring the sizzling sounds.
  5. The chef swiftly sliced the zucchinis for the sizzling stir-fry.
  6. A soft breeze rustled the leaves of the succulent plants on the windowsill.
  7. The sizzling steaks sizzled so slowly on the stove.
  8. Suzie’s sister loves to zest fresh lemons for savory salads.
  9. The sun set slowly, casting a soft, serene glow over the garden.
  10. The spicy seasoning added zest to the soup, making it so satisfying.

These sentences should offer a variety of “s” and “z” sound opportunities for practice, while keeping things engaging!

Those are pretty good sentences, right? But not perfect. In reading those sentences, I notice “sizzling” was used at least four times. We don’t generally grow saplings in the garden. And I don’t think succulent plants have leaves!

So, we have three options here. We could use this in therapy as-is, we could ask ChatGPT to only use “sizzling” once in the list and fix sentences #3 and #6, or we could do a quick edit ourselves. We could also personalize these sentences more by substituting family names for the generic names before printing.

But what do we do if ChatGPT gives a pretty useless response?

return to top

Tips if ChatGPT falls short

If you’ve been using ChatGPT or other AI tools for awhile, you may have noticed that one day, things can be fine, and then the next day, it doesn’t work. In other words, the AI tool randomly won’t give the sort of responses that it did before.

AI tools can behave in strange ways when there is an update on the back end. If it’s an unintended effect, the company usually fixes it pretty quickly. So if ChatGPT can’t perform a task that it used to do, just give it a few hours (or days) and then try again.

At the other end of the spectrum, you may be trying to ask ChatGPT to do a complicated task that it’s never been able to do. You may have to look for another AI tool or simply wait until the technology advances.

But in general, here are some things you can try if you’re not happy with ChatGPT’s responses:

  • Ask ChatGPT to explain its answer. (This is how I learned that I have to say “exactly three syllables.”)
  • Tell it what you liked and didn’t like about its answer.
  • Point out any mistakes and ask it to fix its answer.
  • Break down complex tasks into smaller steps.
  • Explain what you’re trying to do and ask ChatGPT to act as an expert prompt engineer and write a better prompt. (Coursera, 2025, Michael, n.d.)

return to top

Example ChatGPT prompts for SLPs

We can ask ChatGPT to “Act like…”.

  • a neurologist and explain which cranial nerves are important for speech articulation. Please explain how I, as a speech-language pathologist, could assess those cranial nerves.
  • an AAC expert and give me step-by-step instructions for how to create a new message in [insert product name] software.
  • a utilization management specialist. I’m going to describe my last speech therapy session, and I’d like you to suggest how I can word my visit note to show the skill I provided.
  • an IT expert and explain to me how I can set up an out-of-office message on my new work email.
  • a life coach and help me improve my work-life balance.
  • an SLP preparing for a session with an adult who has expressive aphasia. Their goal is to be able to call their salon and make an appointment. We’ve practiced a simple script, but they are concerned that the receptionist might say something unexpected. Could you make a list of possible things a receptionist might say in response to my client?
  • a prompt engineering expert and tell me how I could improve the following prompt, so that the results are as accurate as possible. Just for context, I’m an SLP working with an adult who has dysarthria. Here is my current prompt: Please create a list of 10 sentences such that each sentence contains several words containing the /s/ and /z/ sounds. Please remember the difference between phonology/phonetics and orthography.” This prompt has gotten me some good sentences, but the /z/ sound is mostly represented by words ending in “-ize”. This is repetitive and not reflective of actual conversation. How can I re-word this prompt to get words such as “cheesecake”, “pansy”, “browse”, or “fries”?

return to top

Other AI tools you may like to try

  • SLPeaceBot – a HIPAA-complaint AI tool that can turn a spoken summary of a session into a SOAP note.
  • Sora – OpenAI’s video generation tool. You can create videos from text prompts, images, or existing videos.
  • Runway Gen-2 or Gen -3 – Another tool for creating video, offering text to video, image to video, and text to image tools.
  • Google Notebook LM – Per my last read of the terms of use, this AI tool does not submit any information you share with Google unless you provide feedback (like a “thumbs up”). You can input one or more files into a virtual notebook, and it automatically generates study guides, FAQ’s, and more. You can converse with it, just like you would with ChatGPT. But a really cool feature is that you can create a podcast talk show – the AI characters discuss the material you shared in the notebook! It’s fun to listen to the AI hosts talk about the latest dysphagia article that you’re reading.

return to top

Learn more

return to top

Share your tips!

We’d love to hear about how you’re using AI tools to make your work life or personal life better! Please leave a comment below, or use the contact form.

return to top

References

return to top

return to top

Free DIRECT downloads: ChatGPT prompting for SLPs (cheat sheet). (Email subscribers get free access to all the resources in the Free Subscription Library.)

Featured image by Matheus Bertelli from Pexels, accessed on Canva.com.

Photo of Lisa Young
Website |  + posts

Lisa earned her M.A. in Speech-Language Pathology from the University of Maryland, College Park and her M.A. in Linguistics from the University of California, San Diego.

She participated in research studies with the National Institute on Deafness and other Communication Disorders (NIDCD) and the University of Maryland in the areas of aphasia, Parkinson’s Disease, epilepsy, and fluency disorders.

Lisa has been working as a medical speech-language pathologist since 2008. She has a strong passion for evidence-based assessment and therapy, having earned five ASHA Awards for Professional Participation in Continuing Education.

She launched EatSpeakThink.com in June 2018 to help other clinicians be more successful working in home health, as well as to provide strategies and resources to people living with problems eating, speaking, or thinking.

Being an SLP

Be First to Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.