Subscribe for Free
Join over 5,000 doctors, providers, and staff by subscribing to im+, the largest digital educational resource for medical.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Understanding Language AI Models (AI Part 2)

Published on
June 20, 2024
Last Updated
minute read
Christian Shepherd
Christian Shepherd
Content Strategist
/ Founder

This article is one of a five-part series exploring the fundamentals and practical application of artificial intelligence for professionals in the medical aesthetics industry. We are not responsible for any hostile AI takeovers, Armageddon, human extinctions, or cheesy dystopian plots. We will, however, take full credit for any incredible improvements to your marketing program. That we are used to.

Before ChatGPT became such a popular instrument for the masses, there were already a ton of tools on the market using older versions of OpenAI's natural language models and other AI programs. 

Though, to be completely honest, they kind of sucked. 

Or at least, that's what my first impression was. I am not sure if it was some sort of unconscious self-preservation or if it was just a severe oversight on my part; because as I read the drab, disconnected copy it spit out, I emphatically, with the confidence of someone pushing doTERRA oils in the 2010s, announced to my peers that language AI was basically a non-factor after some mild testing in the middle of 2022.

I was so wrong.

In fact, the next week, I caught wind of some buzz on social media about some different AI tools. I spent a couple of hours exploring them, enamored with the speed at which they produced not great, but passable content. 

By the end of my discovery, I realized my error: I was looking in the wrong places. 

I messaged my content strategy team, addressing my own automation anxiety with a thinly veiled quip:

"I would like to apologize to our AI overlords formally and humbly request that I not get outsourced by AI." 

Those programs last year were less sophisticated than they are now. Still, they were already making quick work of simple things like social media captions, blog snippets, topic creation and a handful of other tasks typically reserved for content specialists. 

Hell, some power users were even pumping out nearly fully developed content.

But in the coming months, the digital dust settled, and as the flaws and limitations of language models started to surface, two things became clear:

  1. Artificial intelligence could be used by savvy marketers and practice owners to either bolster or expedite their marketing programs
  2. You would need to have a comprehensive understanding of both the material you are writing about and the engine of the tool you are using to use AI effectively

If you take the time to learn what an NLM is and focus on understanding how to prompt it best to give you the results you are looking for, there are some really cool ways to use it in your practice and daily operations.

What Is a Natural Language Model?

For those who are entirely in the dark, an NLM is an artificial intelligence model that uses machine learning techniques to generate human-like text…

Or at least get as close to human-like as possible.

These models are typically trained on large amounts of text data in a variety of languages and can be used to perform tasks like:

  • Text generation
  • Sentiment analysis
  • Text classification/categorization
  • Named entity recognition 
  • Translation
  • Text summarization
  • Question answering
  • Chatbot conversation
  • Text-to-speech
  • Semantic similarity
  • Automatic proofreading
  • Keyword extraction

The ultimate goal of an NLM is to understand, interpret, and generate human language in a meaningful and contextually relevant way for your needs — in your case, creating content or offloading some clerical task for your medical practice.

How Do NLMs Work?

Tokens, prompting, and parameters are the three tenets of generative language content. Essentially, the AI model looks at what you are asking it to do, recognizes any limitations or requirements you've asked for and then, one phrase at a time, strings words into sentences, sentences into paragraphs, and paragraphs into fully expressed ideas.

Tokens. A token refers to a unit of text the model reads, generates, or considers at once. Tokens can be as small as a single character or as large as a word or phrase. AI uses the information it was trained on to determine the most likely string of tokens you are looking for. Many models also charge for the AI's content by the token. 

Prompting. This refers to the information that you feed the language model. For example, "Please give me ideas for a blog about Botox" would be the prompt. AI will search the data it was trained on to string together the most likely sought-after series of words until it has fulfilled this request. 

Parameters. These are the limitations or requirements you asked the AI to follow. Parameters can be a minimum word count, avoiding a particular phrase, avoiding specific topics, using a certain voice or a host of other minor changes to the model's task that will change the output you receive. 

What Are the Most Popular NLMs on the Market?

If you stay up-to-date on the latest and greatest in artificial intelligence, you'd know about a few different models making waves in the industry that three leading tech companies are piloting.

And, to the surprise of absolutely no one, the biggest two companies in the AI race are Google and Facebook. But, to the surprise of everyone, they aren't winning the consumer race right now — that award goes to a drastically smaller company: OpenAI. Let's break down all three companies' progress in the AI world.


BERT stands for Bidirectional Encoder Representations from Transformers. Super sexy, I know. But being sexy has never really been Google's thing — it's always been more focused on practicality in usage and productivity. 

The goal of BERT is not something the average person gets too excited about: it wants to teach AI to understand a sentence from both the left and right sides simultaneously rather than moving from left to right like a human would. 

I told you, it's about productivity, not raw sex appeal

Google did have a bit of teenage-romance-star-removes-her-glasses-and-wins-prom-queen moment recently. Without spoiling too much, another large AI company recently stole the spotlight of the AI conversation by dropping a new user interface that made using its AI as easy as sending a text message. 

Spoiler: It was called ChatGPT, and it opened the power of AI to a dialogue box where users could have a "conversation" with the AI model to get information and assistance with tasks.

But — in true Google fashion — they went full pick me girl and told the world about Bard: an NLM that uses LaMDA (Language Models for Dialogue Applications) to do the same thing.

But as the artificial intelligence ecosystem became even more competitive, Bard wasn't holding its own, especially as multimodality — the ability to process text, photo, video, and audio simultaneously — became the new arms race.

So, it launched its new product: Gemini.

Gemini's potential as a primary LLM is huge, especially as it continues to evole into general artificial intelligence.

Google technically has another model, PaLM (Pathways Language Model), but it primarily focuses on improving large-scale language models' learning process. So, it is of little use to the average person. Moving on.


Meta's artificial intelligence team has gotten in on the action as well. Its model, LLaMA or Large Language Model Meta AI, is being trained on a smaller database of information to focus on fine-tuning their program for specific product applications, i.e., actually using it in something like Facebook or the Metaverse.

They seem to have found some success with their approach, and have implemented it in things like Facebook and Instagram search, but only more time will tell how this model will compare to larger models like those by Google and OpenAI, who really seem to be shaping the industry.


If you have ever heard of ChatGPT, Jasper, Rytr, Copy.AI, or any of the countless artificial intelligence platforms bringing language AI to consumers, the odds are that the GPT model created by OpenAI is powering it.

There are numerous iterations of the GPT model. The newest and most powerful is known as GPT-4o. But OpenAI earned enormous mass appeal when it launched ChatGPT, the conversation box interface we mentioned earlier. Almost every professional field — not to mention hordes of students — sought to explore how they could use ChatGPT. 

Its 3.5 model is still the basis for many applications today, but with the addition of GPT-4o, it has become more complex in its ability to comprehend instructions and information and provide meaningful feedback. Couple that with some recent beta features like web surfing and plugin support, and you have a potent tool that has earned its place at the forefront of natural language processing prestige. 

In fact, OpenAI was technically the first to announce their version of multimodality, a day before Google's announcement. Just to really flex on 'em.

Why Is Natural Language AI So Useful?

There was a genuine concern among content creators that NLMs would flat-out replace them. On-screen personalities and videographers were probably feeling alright, but content writers, copywriters, technical writers, screenwriters, journalists, bloggers or anyone whose marketable skill was writing eloquently, all had a gut-wrenching moment of terror.

To some degree, it was warranted. If you aim to produce 1,000 words of informative content without concern for voice, personality, web optimization, storytelling, structure or any of the other hundreds of writing elements that make something feel special, then tools like ChatGPT are perfect. 

See, language AI doesn't excel at being creative. It excels at producing focused and straightforward content, coming up with ideas for a topic you feed it or sifting through large amounts of information and either synthesizing or parsing it out.

In short, current AI models excel in three major categories: ideation, production and automation. 


If you have ever sat around thinking about the best way to market a new medical device or wanted to come up with a new idea to push your practice's branding, then you have participated in ideation. But it goes beyond that — you can ideate small things as well: 

What are some topics that we can cover with blogs? What are some specific titles that might be eye-catching for those blogs? What kind of social media coverage should we do? How should we approach our webpage on a new surgery technique? What kind of creative assets should we use for an ad campaign? What do we want our website to convey in terms of emotion? How do we achieve that emotion? What are some symbols that represent our brand identities or values? What should our practice name even be? 

All of this is ideation. And as a tool capable of producing thousands of answer permutations if you ask for it, ChatGPT is a great tool to get the process started. Here's an example:

In a matter of seconds, there are 15 blog topic ideas to ponder. Are any of these ready to go straight off the press? No, not really. I'd probably workshop these a little to make them more focused and catchier. 

But I will give a special shoutout to number 15 — that's a damn exciting topic.

Remember, ideation is not making decisions. It is about getting creative before deciding on a final approach, hoping that what you land on is as interesting and valuable as possible.


This is a complicated section because the truth is that you would probably be shocked at how quickly and effectively you could produce content with something like ChatGPT. Just look at this essay it wrote on The Intersection of Feminism and Mommy Makeovers topic from earlier. It did all that in about 60 seconds.

Is this essay going to win a Pulitzer? No, absolutely not. But it put relevant words down on paper, which alone could serve as a compelling outlining process for written content creation. 

Consider other marketing applications, though. Let's say you wanted to put together a list of minimally invasive or noninvasive treatments that can address different skin concerns on the face. You could spend a couple of hours compiling that between seeing your patients, or you could just…

…and get this in return:

And look: Gemini can do it too:

I'll let you decide which table is more useful or accurate. Regardless if you stan Gemini or GPT, the point is that you have a powerful tool to improve your content production. We will cover some non-marketing-related ways to use language AI for your practice as well later in this article. 

But before we do… you might be wondering “can I just automate all of my content writing now?”


For now, I will let you know this is a bad idea and promise that if you stick with this article, I will explain why aiming for full automation is generally a terrible idea.


Here is one page of the instruction manual for a Fraxel device:

Nothing about this makes me or anyone want to read it. And that may be intentional. These are scary things that are legally required to be here. Fraxel devices are not alone in their seemingly purposeful don’t-spend-too-much-time-looking-at-me-approach; instruction manuals for any medical device are usually the worst, even when written well. 

You can't exactly paste these warnings on your machine for easy reference as a reminder before you do a procedure. But language AI can definitely break it down for you. Here is the response I got after copying and pasting everything into the language model and asking it for a simple checklist:

These instructions could be followed by any operator using the device with ease. This is just one example of automating some tasks around your practice. But let’s dive into more:

How to Use Language Models In Your Practice

Now we find ourselves at the juicy part: how can I use AI outside of my marketing for my practice? Glad you asked.

Automate Low Stakes Communications

You could go to ChatGPT or Gemini and ask them to create an email confirming their upcoming appointment that includes other important information about their visit. Here's a template I made in a few seconds:

Protip: If your email marketing platform is set up correctly, personalization elements like the patient's first name can be automated. 

Create Patient Education

Providing patients with pre- or post-operative information is critical to their successful recovery and results. If you find yourself exhausted from creating these materials from scratch or updating them based on the patient you are treating, an NLM might be helpful.

Here is a prompt I gave ChatGPT:

Please create a list of post-operative instructions for a patient who underwent a tummy tuck. Provide information about preparing for the procedure, a timeline of the recovery, what to look out for during the healing process, and tips on making sure they heal successfully. 

This patient, in particular, is generally very active and may push themselves to exercise too early. Be sure to mention to avoid doing any activities without clearance from the surgeon. 

And here is what ChatGPT provided me:

Now, you'll need to review it for accuracy — after all, ChatGPT passed the medical licensing exam with flying colors, but it did not get a perfect score. Nor does it have your years of specialized training.

Automate Charting and Documentation

Here is some data from Statista about how many patients physicians generally see each day:

As you can see, most physicians will see between 11 and 30 people per day. This number might be smaller for the medical aesthetic industry, particularly for people who focus primarily on plastic surgery, but it still highlights a growing concern often voiced by patients: 

You simply don't have the time to talk with them. 

Really, what that means is, despite your current best efforts, you are in desperate need of more time to build a meaningful relationship. During the consultation, you might be plugging away at the patient's chart — attention focused on the screen or paperwork rather than the human in front of you.

This doesn't send a great message of confidence to the patient who came to see you.

But there is a better way. Voice-to-text generation is more powerful than ever. You could record (and securely store) a voice recording of your consultation with the patient's permission. This means the questions you are asking are being documented without having to fill them out yourself. 

Then, later on, you could have an NLM take the text from that recording that was automatically generated and distill it down into useful chart notes, which can be easily copied and pasted. It's really that simple.

Did… did we just solve charting?

Expedite Billing and Coding

With some third-party help, you could generate sheets of information to help with billing and coding and export them to text for ChatGPT, Gemini or any NLM you choose. 

If you had a list of patients and the treatments they received and needed to generate ICD codes for those treatments, ChatGPT could fill that in for you quickly. 

You might need to spend some time tinkering to get the correct coding and information you need, but if it's a tool you can commit to figuring out, then your billing department might be thankful for some automation magic to help them get through their backlog. 

Why You Should Avoid Using Raw AI-Generated Marketing Material

If you were reading this hoping that at the end, I could tell you that you'd be getting A+ content for free… sorry, that's not really in the cards. AI has limitations. Most notably, it does not understand the nuance of writing persuasive content. 

I am not talking about the "QUICK ACT NOW" call to action you see everywhere on the internet, but the subtle, genuine approach to educating someone to show them that you are a trustworthy source of information and treatment. The reputation building, the authentic approach to communication — it just can't understand the mechanics of your funnel that way.

That isn't to say it can't help you get there — don't throw the baby out with the bathwater — but you need someone who understands persuasive content marketing and the subject matter overseeing the use of AI so that the content will hit your goals. 

Its raw output just isn't branded for your needs, and more problematically, all these language models still have the capacity to be wrong. The only thing worse than unpersuasive content is content that is erroneous. 

(Erroneous? At least AI wouldn't be smug enough to write something pompous like eRrOneOuS.) 

For a real-life example of how trusting AI-generated content implicitly can go horribly wrong, check out this article about two lawyers whose court filing included fake cases that were made up by ChatGPT

AI Is a Tool, Not a Substitution

I wanted to achieve two things in this article about generative language AI: 1) get you excited about the technology and how you can utilize it at your practice, and 2) temper your expectations so that you understand any NLM is a tool to be used, not a replacement for people who generally perform those tasks. 

You won't be outsourcing your workload to ChatGPT anytime soon, but you will be bolstering your content and earning back some time if you dive into learning it. 

TL;DR Generative language AI models, like ChatGPT, have revolutionized content creation, offering tools that can produce close to human-like text. These Natural Language Models are trained on vast amounts of text data and can perform tasks ranging from text generation to sentiment analysis. The leading players in this space include Google with its BERT and Gemini models, Meta's LLaMA, and OpenAI's GPT series. NLMs are particularly useful for ideation, production, and automation. They can assist in automating low-stakes communications, creating patient education materials, streamlining charting and documentation, and expediting billing and coding. However, while AI can produce content rapidly, it's essential to approach it as a tool rather than a replacement. Raw AI-generated content might lack the nuance and branding required for effective marketing. It's crucial to have a human touch to ensure accuracy, authenticity, and relevance in the content.