How to Fact-Check What ChatGPT Tells You About Your Teeth

In short AI can sound confident when it's wrong. A simple routine of checking against NHS pages, asking for sources and spotting vague answers catches most mistakes.

On this page 5 sections
  1. The mistakes AI actually makes about UK dentistry
  2. The routine: four checks, in order 1 min
  3. A worked example from start to finish 4 min
  4. What to keep out of the chat box 6 min
  5. When to stop asking the AI 6 min

You’ve just left the dentist with a printed treatment plan. It says something like “UL6 RCT, PFM crown, 2x MOD composite” and a total of £1,140. You type it into ChatGPT, and within seconds you get a tidy explanation, a price comparison and a recommendation. It reads well, and it sounds sure of itself.

That is exactly the problem. So, is ChatGPT dental advice accurate? Often it is. When it’s wrong, though, it sounds just as confident as when it’s right, and you have no clinical training to tell the two apart. You don’t need that training. A ten-minute checking routine catches most of the mistakes, and this post walks you through it using real examples.

The mistakes AI actually makes about UK dentistry

AI tools rarely invent nonsense about what a root canal is. The errors cluster in three predictable places.

Out-of-date prices. NHS dental charges in England usually change in April. Ask a chatbot what Band 3 costs and you may get £306.80, which was the 2023 figure. From April 2025 it was £326.70, with Band 1 at £27.40 and Band 2 at £75.30. If a tool’s training data stops before a price rise, it will quote the old number with total confidence and no warning.

Mixing up the four UK nations. England’s three-band system is not used in Scotland, where NHS treatment is charged item by item at a set percentage of the fee. Wales and Northern Ireland have their own arrangements too. ChatGPT, Gemini and Copilot all have a habit of answering “UK” questions with English rules. If you live in Glasgow and your chatbot is telling you about Band 2, it’s answering a question you didn’t ask.

Overstating or understating what the NHS covers. Typical examples:

  • Claiming dental implants are available on the NHS “in most cases”. They are only offered in limited circumstances, usually through hospital services, and almost never at a high street practice.
  • Saying that white fillings on back teeth are “never” available on the NHS. The real rule is that the NHS provides what is clinically necessary, and the dentist makes that call.
  • Missing the two-month rule. If you need more treatment from the same band or a lower one within two months of finishing a course, you shouldn’t pay again. Plenty of AI answers leave this out.

None of these are exotic. They’re the exact questions someone holding a quote will ask.

The routine: four checks, in order

1. Make it show its working

Don’t ask “is this a fair price?” Ask a question that forces the tool to tell you where its answer came from:

My dentist in Leeds has quoted NHS Band 3 for a root canal and crown
on an upper molar. What is the current Band 3 charge in England?
Give me the source you're using and the date that source was last updated.
If you aren't sure the figure is current, say so.

That last line matters. You’re giving the model permission to admit uncertainty, and most will take it. When a tool says “as of my last update” or “this may have changed”, treat that as useful information: this is a number you need to check yourself.

If you’re using ChatGPT with search switched on, or Perplexity (which shows its sources by default), you’ll get links. Without search, you’ll get a figure from memory, and memory is where stale prices live.

This is where most mistakes get caught. Run three quick checks on each link:

CheckWhat a bad answer looks like
Does the link open?404 page, or a URL on nhs.uk that doesn’t exist. Chatbots sometimes make up plausible-looking addresses.
Is it the right country?An NHS England page cited for a patient in Cardiff or Belfast.
Does the page actually contain the figure?The page exists but says £326.70 while the AI told you £306.80.

For NHS charges, the page you want is the “How much will I pay for NHS dental treatment?” page on nhs.uk. Scottish patients should use NHS inform, and Welsh patients should use the 111 Wales site. The NHS figure always beats the chatbot’s figure. No exceptions.

Private prices are harder because there’s no national list. A chatbot might say a private molar root canal costs “£300 to £1,500”, which is so wide it’s useless. A better check is to look at two or three local practice websites, since many publish fee lists. In much of England a private molar root canal often comes in somewhere between £500 and £1,200, and a specialist endodontist will charge more than a general dentist. If your quote is £2,400 for a single molar and three local fee lists top out at £1,100, you now have a real question to ask, and it came from real evidence.

3. Spot the vague answer

Vagueness is how AI hides the gaps in what it knows. Learn to see it on the page:

PhraseWhat it often meansWhat to ask next
“Costs can vary widely”It doesn’t have a figure“Give me a range for a private practice in [your town] and cite where it’s from.”
“Your dentist will advise”Correct, but it’s dodging“What questions should I ask my dentist about this specific item?”
“In some cases this may be covered”It doesn’t know the rule“Under what specific NHS rule is it covered, and where is that written?”
“Generally considered safe and effective”Filler“What are the main risks, and how often do they happen?”

A good answer has numbers, a named source and a clear “it depends on X”. A bad answer has adjectives.

4. Ask a second tool the same question

Paste the identical prompt into a different assistant. Claude, Gemini, Microsoft Copilot and Perplexity are all free to try. If two tools give the same figure and both cite the same nhs.uk page, you can be fairly confident. If they disagree, one of them is wrong, and you check the official source to find out which.

It’s also worth asking the same tool twice in a fresh chat. If you get £306.80 one time and £326.70 the next, the model is guessing.

A worked example from start to finish

Priya, in Nottingham, gets this plan from her NHS practice:

Course of treatment: NHS Band 3        £326.70
  - Root canal treatment, UR6
  - Crown, UR6 (metal-ceramic)
Private (optional):
  - Tooth-coloured filling, LL7         £145.00
Total if all accepted:                  £471.70

She asks ChatGPT, without search, whether this is fair. It tells her Band 3 is £306.80, so she’s being overcharged by about £20, and that “white fillings are not available on the NHS for back teeth, so the private charge is normal”.

She runs the routine.

Check 1: she asks for the source and date. The reply admits the figure is “based on information that may not reflect recent changes”. That’s a flag.

Check 2: she opens the nhs.uk charges page herself. Band 3 is £326.70, so the dentist’s figure is correct and the AI’s was out of date. She wasn’t being overcharged.

Check 3: the filling answer said “not available”. That’s an absolute, so she asks what the actual rule is. The second answer is more careful: the NHS will provide a clinically suitable filling, which on a back tooth is often amalgam, and a dentist can offer tooth-coloured as a private extra if NHS treatment would do the job. That’s a different claim from “not available”.

Check 4: she asks Perplexity the same filling question. It cites the NHS page and agrees with the corrected version.

Priya’s question for her dentist is no longer “am I being ripped off?” It’s this: “Would an NHS filling be suitable for LL7, and if so, what’s the actual benefit of the £145 option for me?” That’s a question a dentist can answer properly, and it might save her £145.

Notice that the plan lists NHS and private items separately. That’s correct practice. Your dentist should give you a written plan showing which treatment is NHS and which is private, and the NHS part can’t be made conditional on you accepting the private part. If an AI tells you otherwise, check it.

What to keep out of the chat box

Don’t paste your full name, date of birth, NHS number or practice address into a general-purpose chatbot. You don’t need them to get a useful answer. “Woman, 40s, Nottingham, NHS patient, upper right molar” is plenty. Also check the settings: ChatGPT and Gemini both let you turn off use of your chats for training, and it’s worth doing before you start asking about your health.

Uploading a photo of your treatment plan is convenient, but it shares everything printed on it. Typing out the line items takes two minutes and shares much less.

When to stop asking the AI

Some questions can’t be fact-checked against a web page, because the answer depends on your mouth. Whether a tooth can be saved, whether a crown is really needed or whether a filling would do, whether a wisdom tooth is likely to cause trouble: these need someone who has looked at your X-rays. AI can help you write the questions. It can’t answer them.

If you want a second opinion, a different dentist can look at your existing X-rays (you can ask your practice for copies). You can also check that any dentist is registered by searching the General Dental Council register at gdc-uk.org, which takes about 30 seconds.

For more on writing good prompts, what different tools are good at and how to read a dental treatment plan line by line, see our full guide to asking AI about your teeth.

Next time a chatbot gives you a price, ask it for the date that price is from. You’ll be surprised how often the answer is “I’m not sure”.