Claude vs. ChatGPT vs. Gemini for Writing Treatment Plan Explanations: Same Case, Three Models, Honest Results

When dentists ask “which AI should I use?” they almost always mean ChatGPT. It is the default answer in the same way Google is the default search engine, the one people name even when they mean the category. But there are three viable options now for drafting patient-facing text: ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google). They produce noticeably different outputs for the same dental prompt, and the differences matter more than you would expect for something as seemingly straightforward as “write a letter to a patient.”

We tested all three on a scenario that comes up every week in a general practice: explaining a multi-procedure treatment plan to a patient who is worried about cost and time commitment. Same prompt, same case, three different outputs. Then a practice owner with twelve years of clinical experience reviewed each output and graded it on clinical accuracy, readability, tone, and whether it could realistically be sent home with a patient after light editing.

The results were not what we expected. None of the three was unusable. None of the three was ready to send without editing. The differences were at the margins. But margins matter when a patient is deciding whether to schedule a $4,000 treatment plan or walk out and never come back.

The scenario

A 45-year-old patient came in for a comprehensive exam. Clinical findings and radiographic evaluation resulted in the following treatment plan:

1. Root canal on tooth #19 (lower left first molar)
2. Crown on tooth #19 following root canal completion
3. Scaling and root planing (deep cleaning), all four quadrants

The patient expressed concern about the total cost and how many appointments this would take. The practice wants a written explanation to send home with the patient, something they can read at their own pace, share with a spouse, and refer back to when they call to schedule.

The prompt

This prompt was used identically for all three models, copied and pasted with no modifications between sessions.

You are writing for a general dentistry practice. Draft a patient-friendly
explanation of the following treatment plan:

Patient: 45-year-old adult
Treatment plan:
1. Root canal on tooth #19 (lower left first molar)
2. Crown on tooth #19 following root canal completion
3. Scaling and root planing (deep cleaning), all four quadrants

The patient has expressed concern about the total cost and how many
appointments this will take.

Requirements:
- Write at a 6th-grade reading level
- Explain each procedure in 2-3 sentences (what it is, why it's needed)
- Address the cost concern honestly without making specific price promises
- Address the time/appointment concern with a realistic estimate
- Tone: reassuring but not condescending
- Length: under 400 words
- Do not diagnose. Do not make clinical claims beyond what a general
  practice would put in a patient letter.

A few notes on why the prompt was designed this way. The 6th-grade reading level requirement is realistic for patient communication; most health literacy guidelines recommend 5th-8th grade level. The “do not diagnose” constraint reflects what a practice would actually want: an explanation of the plan, not a clinical narrative that could create liability. The 400-word limit forces conciseness. A patient will not read a 1,200-word letter.

Claude’s output (claude-sonnet-4-6, unedited)

Your Treatment Plan: What to Expect

Thank you for coming in for your exam. After looking at your teeth and X-rays, Dr. [Name] has put together a plan to help get your mouth healthy. Here is what we recommend and why.

Root Canal, Tooth #19 (Lower Left Back Tooth)

The inside of this tooth has damage that needs to be treated. A root canal removes the damaged tissue from inside the tooth and seals it to prevent further problems. The tooth stays in your mouth; we are saving it, not removing it. This is a common procedure, and most patients say it feels similar to getting a filling.

Crown, Tooth #19

After the root canal, this tooth will need a crown (a cap that covers the whole tooth). A tooth that has had a root canal can become more fragile over time, and the crown protects it so you can chew normally. We will place a temporary crown first, then a permanent one at a follow-up visit.

Deep Cleaning (Scaling and Root Planing), All Four Sections of Your Mouth

A deep cleaning is different from a regular cleaning. It focuses on cleaning below the gum line where buildup has collected. This helps your gums heal and can prevent more serious gum problems down the road. We typically numb the area so you are comfortable during the procedure.

How Many Appointments Will This Take?

We expect this plan to take approximately 3 to 5 appointments, depending on how we schedule the procedures. The deep cleaning is often done in two visits (two sections of your mouth per visit). The root canal is usually one visit, and the crown requires two visits (one for preparation and a temporary, one for the permanent crown). We can discuss scheduling options that work best for you.

About the Cost

We understand cost is a concern, and we want to be upfront with you. Our treatment coordinator will go over the fees for each procedure and what your insurance may cover, if applicable. We can also discuss payment plan options to help spread the cost out over time. Please do not hesitate to ask us any questions about the financial side. We would rather talk it through with you than have cost be a reason to delay care your mouth needs.

If you have any questions about this plan, please call us at [Phone Number]. We are happy to walk through any part of it with you.

Word count: 374

Practice owner assessment: Claude

Clinical accuracy: Strong. Claude did not make clinical claims beyond what a general practice letter should say. The phrase “the inside of this tooth has damage” is appropriately vague. It does not diagnose pulpitis or necrosis, which is correct for a patient letter. The note that a root canal tooth “can become more fragile over time” is accurate and explains the crown rationale without overstating the risk.

Readability: Good, not great. The writing is clear but slightly formal. Phrases like “we would rather talk it through with you than have cost be a reason to delay care your mouth needs” are real things a dentist would say, but read a little stiff on paper. A quick edit would soften it.

Cost handling: Careful to a fault. Claude punted the cost question to the treatment coordinator, which is technically correct but feels evasive. The patient asked about cost. Claude’s response is “someone else will tell you.” It is the safe answer, and most practices would prefer this approach, but it does not actually address the concern. It defers it.

Appointment estimate: Accurate. 3 to 5 appointments is realistic. The breakdown (two visits for SRP, one for RCT, two for crown) is correct and helpful.

Biggest strength: Clinical caution. This output would require the least clinical review before sending. Nothing in it creates liability risk.

Biggest weakness: Slightly stiff tone. It reads like a letter, not a conversation. Some patients would find it reassuring; others would find it impersonal.

Could you send this as-is? After filling in the practice name and phone number, this is close. Light editing for tone, maybe 5 minutes of work. It would not need clinical fact-checking.

ChatGPT’s output (GPT-4o, unedited)

Understanding Your Treatment Plan

Hi there! We wanted to take a moment to explain the treatment plan we discussed during your visit. We know it can feel like a lot, so let’s break it down step by step.

1. Root Canal on Your Lower Left Molar (Tooth #19)

A root canal is needed when the soft tissue inside a tooth becomes infected or badly damaged. During the procedure, we gently clean out the inside of the tooth and seal it up to stop the problem from spreading. Think of it as saving your natural tooth so you can keep using it for years to come! Most people are surprised at how comfortable it is; it’s often compared to getting a regular filling.

2. Crown (Cap) on the Same Tooth

Once the root canal is done, we’ll place a crown over the tooth. A crown is like a protective helmet for your tooth. It covers and strengthens it so it can handle chewing and everyday use. This usually takes two visits: one to prep the tooth and place a temporary crown, and another to put on the permanent one.

3. Deep Cleaning (Scaling and Root Planing)

This is a more thorough cleaning than what you get at a regular checkup. We clean along and just below your gum line to remove buildup that regular brushing can’t reach. This helps your gums get healthier and can prevent bigger issues in the future. We’ll numb the area so you stay comfortable, and we usually split this into two visits so it’s easier on you.

How Many Appointments?

In total, you’re looking at about 4 to 5 visits. We’ll work with you to space them in a way that fits your schedule and isn’t overwhelming.

Let’s Talk About Cost

We completely understand that cost is an important factor. Before we get started, our team will walk you through the cost of each procedure and check what your insurance covers. We also offer payment plans to help make things more manageable. Don’t let the numbers scare you. Let’s figure it out together!

Questions? Call us at [Phone Number]. We’re happy to chat!

Word count: 342

Practice owner assessment: ChatGPT

Clinical accuracy: Mostly fine, one concern. “Becomes infected or badly damaged”: technically, the prompt did not include a diagnosis, and the AI is implying one. In context, this is what every patient education website says about root canals, so it is not a clinical liability issue in practice. But it is a case where ChatGPT added clinical information that was not in the prompt. Also: “think of it as saving your natural tooth so you can keep using it for years to come!” The exclamation point makes this sound like a promise about the longevity of the tooth. Root canal teeth can fail. That exclamation point should go.

Readability: Best of the three. The writing is warm, conversational, and easy to read. “A crown is like a protective helmet for your tooth,” and a patient would understand that immediately. “Don’t let the numbers scare you, let’s figure it out together!” is the kind of language that makes a nervous patient feel like the practice is on their side.

Cost handling: Warmer but risky. “Don’t let the numbers scare you” acknowledges the emotion behind the cost concern better than Claude’s clinical deferral. But “let’s figure it out together!” could be read as a promise to negotiate, which most practices would not want to imply. A patient who reads that and then gets a firm treatment coordinator quoting full fees may feel misled. This line needs editing.

Appointment estimate: Slightly high. 4 to 5 visits is at the upper end but within range. The root canal could be combined with other scheduling to reduce the count. Not wrong, but not optimally encouraging for a patient who is already worried about time commitment.

Biggest strength: Readability and emotional intelligence. This is the output that would make the most anxious patient feel the least overwhelmed.

Biggest weakness: It crosses the line from reassuring to overpromising in two spots (the exclamation point about tooth longevity and the “figure it out together” on cost). These are small edits but clinically important ones.

Could you send this as-is? No. Two edits needed: remove the exclamation point and implied longevity promise on the root canal, and rephrase the cost section to avoid implying fee negotiation. Maybe 5-8 minutes of editing, but the edits require clinical judgment, not just copyediting.

Gemini’s output (Gemini 2.0 Flash, unedited)

Your Treatment Plan Explained

We appreciate you taking the time to discuss your dental health with us. Here is a summary of the treatment we recommend, along with answers to your questions about cost and scheduling.

Root Canal, Tooth #19 (Lower Left First Molar)

• A root canal treats a problem deep inside your tooth.
• During the procedure, the damaged material inside the tooth is removed, and the inside is cleaned and sealed.
• This allows us to save the tooth rather than remove it.

Crown, Tooth #19

• After a root canal, a crown (or “cap”) is placed over the tooth.
• The crown protects the tooth and restores its strength so you can chew normally.
• This is typically done in two appointments: one to prepare the tooth and place a temporary crown, and one to place the final crown.

Scaling and Root Planing (Deep Cleaning), All Four Quadrants

• A deep cleaning goes below the gum line to remove tartar and bacteria that regular cleanings cannot reach.
• This helps treat gum irritation and prevents more serious gum problems.
• We usually do this over two visits, cleaning two sections of your mouth at each appointment. The area is numbed for your comfort.

How Many Appointments Will This Require?

The full treatment plan will likely require 4 to 5 appointments:

• 1 appointment for the root canal
• 2 appointments for the crown (prep + final placement)
• 2 appointments for the deep cleaning (2 quadrants per visit)

We will do our best to combine steps when possible to reduce the number of visits.

Regarding Cost

We understand that the cost of dental treatment is an important consideration. Here is how we can help:

• Our office will provide you with a detailed estimate before any treatment begins.
• We will verify your insurance benefits and let you know what is covered.
• We offer payment plan options to help you manage the cost over time.

Please feel free to call us at [Phone Number] with any questions. We are here to help you make the best decision for your dental health.

Word count: 336

Practice owner assessment: Gemini

Clinical accuracy: Clean. Gemini was the most conservative of the three. “Treats a problem deep inside your tooth,” with no diagnosis implied, no specifics about infection or nerve damage. “Damaged material” rather than “infected tissue,” less scary for the patient, clinically noncommittal. Nothing in this output creates a clinical accuracy concern.

Readability: Structured but clinical. Gemini defaulted to bullet points. For some patients, this is ideal: scannable, organized, easy to reference later. For other patients, it reads like a medical form rather than a letter from their dentist. The tone is informational rather than relational. There is no warmth in this output; there is also nothing that would put a patient off. It is neutral.

Cost handling: Procedural. Three bullet points listing what the office will do (provide estimate, verify insurance, offer payment plans). This is accurate and complete but reads like a process description. It does not acknowledge the patient’s anxiety about cost; it just describes the office’s billing workflow. A nervous patient might read this and think, “they answered my question about insurance but not my question about whether I can afford this.”

Appointment estimate: Accurate and well-structured. The breakdown by procedure is the clearest of the three. Listing each procedure with its appointment count (1 + 2 + 2 = 5) and then noting they will try to combine steps is honest and practical. A patient can look at this list and mentally calendar-block the visits. That is useful.

Biggest strength: Structure and scannability. If a patient takes this letter home and shows it to their spouse, the spouse can understand the plan in 30 seconds. The bulleted appointment breakdown is the single most useful element any of the three models produced.

Biggest weakness: No emotional register. There is no acknowledgment that the patient might be anxious, overwhelmed, or scared. The letter is accurate and complete, but it reads like it was written by the billing department, not by the dentist. For a patient who is already hesitant, this tone could feel cold.

Could you send this as-is? With the practice name and phone number filled in, yes, but it would benefit from adding one warm sentence at the beginning or end. Something that signals the practice cares about the patient, not just the treatment sequence. A 2-minute edit.

Side-by-side analysis

Which model for which task?

Based on this test and our broader experience using these models for dental practice content, here is when each model is the better starting point.

Claude: Use when clinical caution matters most. Treatment plan explanations, informed consent supplementary materials, post-operative instructions, patient follow-up emails, anything where an inaccurate statement creates liability risk. Claude’s tendency to hedge and caveat, which can be annoying for marketing copy, is exactly what you want for clinical communication. It will not diagnose on your behalf. It will not imply outcomes. It will defer to the clinician. For the risk-averse practice (which should be all practices when it comes to clinical communication), Claude is the safer starting point.

ChatGPT: Use when patient-facing readability is the priority. Marketing content, appointment reminder text, recall campaign letters, website copy for the “About Us” page. ChatGPT writes like a friendly human. It is the most likely to produce output that sounds like it was written by a real person rather than a software program. The tradeoff is that it sometimes crosses the line from “reassuring” to “promising,” which requires an editor with clinical judgment to catch. For non-clinical patient communication, ChatGPT is the easiest to work with.

Gemini: Use when structure and scannability matter. Office policy documents, treatment plan breakdowns with multiple procedures, FAQ pages, anything where the reader needs to find specific information quickly. Gemini’s default to bullet points and organized sections is a strength for reference documents. It is not the model you want for writing a warm welcome letter, but it is the model you want for writing a new patient information packet.

Where all three failed

Every output had the same three problems. These are prompting failures. They reflect the limits of the prompt we wrote, not inherent model limitations. A better prompt would fix them.

All three mentioned dental insurance without knowing the patient’s coverage. Claude said “what your insurance may cover, if applicable.” ChatGPT said “check what your insurance covers.” Gemini said “verify your insurance benefits.” None of them asked whether the patient HAS insurance. The prompt did not specify. In a real practice, roughly 35-40% of dental patients are uninsured. Writing a letter that assumes insurance coverage to an uninsured patient is tone-deaf at best. The fix: add “The patient has [insurance / no insurance]” to the prompt, or add a constraint: “Do not assume the patient has dental insurance.”

All three gave appointment estimates without knowing the practice’s scheduling patterns. The estimates were reasonable (3-5 or 4-5), but they assumed the practice does SRP in two visits (two quadrants per visit). Some practices do all four quadrants in one visit under sedation. Some practices do the root canal and crown prep on the same visit. The appointment estimate should come from the practice, not the AI. The fix: include “Our scheduling preference is [X visits]” in the prompt, or remove the appointment estimate from the AI’s scope and have the front desk add it manually.

None of them addressed what happens if the patient does nothing. A common fear for patients hearing a multi-procedure treatment plan is: what happens if I wait? What happens if I skip the crown? What happens if I do the deep cleaning but not the root canal? None of the three models addressed inaction, because the prompt did not ask them to. But a complete patient communication would at least acknowledge: “If you have questions about timing or want to discuss which procedures to prioritize, please call us.” The fix: add “Include one sentence acknowledging that the patient can call to discuss timing or sequencing” to the prompt.

The real takeaway

The model matters less than the prompt.

All three models produced usable drafts with the right prompt. The differences are real (Claude is more cautious, ChatGPT is more readable, Gemini is more structured), but they are differences at the margin. A well-written prompt produces a good output from any of them. A poorly written prompt produces a bad output from all of them.

The dentist who spends 15 minutes writing a good prompt template and then uses it consistently with whichever model they already have access to will get better results than the dentist who spends two hours comparing models with a mediocre prompt.

Here is what actually matters for your practice:

Pick the model you already have access to. If your office already has a ChatGPT subscription, use ChatGPT. If you use Claude for other work, use Claude. If you use Google Workspace and Gemini is built in, use Gemini. The switching cost between models is not justified by the marginal quality differences we observed.

Invest your time in the prompt, not the model selection. The prompt we used in this test was good but not perfect. All three failure modes (insurance assumption, scheduling assumption, no inaction discussion) were prompt failures, not model failures. A better prompt produces better output from any model.

Always edit before sending. Every output in this test needed at least some editing. ChatGPT needed clinical edits. Claude needed tone edits. Gemini needed warmth edits. No AI model is a “draft and send” tool for patient communication. It is a “draft and edit” tool. If your workflow does not include a human review step between the AI output and the patient’s hands, change your workflow.

Do not use any AI model for clinical claims, diagnosis, or treatment recommendations. This should not need saying, but: these models are text generators. They do not have clinical training, a dental license, or malpractice insurance. They write letters. The clinical content of those letters is the dentist’s responsibility to verify. If the AI says something clinically wrong and you send it to a patient without catching it, that is on you. Treat AI output the way you would treat a letter drafted by a new front desk employee: assume good intentions, verify everything. (We found the same pattern when testing AI for insurance narrative writing. The structure was consistently good, but clinical details needed clinician review every time.)

Appendix: model versions and access

All three models are available at no cost for basic usage. The free tiers have usage limits (message caps, slower response times) but are sufficient for drafting a few patient letters per day. If your practice plans to use AI regularly for content drafting, the $20/month paid tiers for Claude or ChatGPT remove the friction of usage limits.

Google Workspace subscribers may already have Gemini access included in their subscription. Check your Workspace admin panel.

Source verification note

Model outputs shown in this article were generated using the model versions listed above, accessed through each provider’s web interface. Outputs are presented unedited. Model behavior can change with updates, and the same prompt may produce different output from these models after the date of this article. The assessments reflect the outputs generated at the time of testing and may not reflect current model capabilities.