You’ve got a hospital letter full of words you don’t recognise, or you woke up with something odd and Google has already convinced you it’s fatal. This is where an AI chatbot actually earns its keep, if you use it right. Not as a doctor. As a translator and a bit of prep help before you talk to an actual human who’s trained for this.

What you’ll be able to do

By the end of this, you’ll have a free AI assistant set up and ready to explain what a symptom or a letter actually means in plain English, help you get your thoughts straight before a GP appointment, and, importantly, know exactly when to stop typing and ring 111 instead. That last bit matters more than the rest of it combined.

Before you start

AI is genuinely good at some things here. It can take a phrase like “raised inflammatory markers” and tell you what that means without the Latin. It can help you turn “I’ve felt a bit off for weeks” into something more useful, like a list of when it started, what makes it worse, what you’ve already tried. Doctors appreciate specifics, and most of us are rubbish at giving them under pressure in a ten-minute slot.

What it absolutely cannot do: examine you. It has no idea what your skin looks like, whether that lump is soft or hard, whether your history includes something relevant from three years ago that you forgot to mention. It doesn’t know you. It’s working entirely off what you type, and it can sound completely confident while being wrong. That’s the bit to hold onto through all of this.

Get set up

You don’t need anything fancy. Pick one of the mainstream assistants, ChatGPT, Claude, Gemini or Copilot, and make a free account. They all work roughly the same way for this: you type a question, you get an answer, you can ask follow-ups.

Go to the website for whichever one you fancy (chatgpt.com, claude.ai, gemini.google.com, or Copilot inside Microsoft’s apps or on its own site), sign up with an email address, and you’re in. No downloads required, no card details for the free tier. If you’ve got a hospital letter as a photo or a PDF, check whether the app lets you upload files, most of the mainstream ones do, which saves you typing out medical jargon by hand and getting it wrong.

That’s genuinely it for setup. The skill here isn’t technical. It’s knowing what to ask and, more importantly, when to stop asking.

Try it yourself

Here are three prompts that do three different jobs. Copy them, tweak the details in the brackets, and see what comes back.

First, understanding a letter or a result:

I've received a letter/test result from the NHS and don't understand some of the language. Here's what it says: [paste the text]. Can you explain what this means in plain English, without diagnosing me or telling me what to do? I just want to understand the words.

Second, getting ready for an appointment:

I have a GP appointment coming up and want to describe my symptoms clearly. Here's what's been happening: [describe it: when it started, what it feels like, what makes it better or worse, anything you've tried]. Can you help me organise this into a short, clear summary and suggest three or four questions I should ask my doctor?

Third, the urgency check, which is the one that matters most:

I have [symptom], and it started [when]. Based on general NHS guidance, is this something that could wait for a routine GP appointment, or does it sound like I should contact NHS 111 or seek urgent care? Please tell me if you're not certain, and remind me to call 111 or 999 if in doubt.

Notice that last prompt asks the assistant to admit uncertainty. That’s deliberate. You want it hedging, not reassuring you like a mate down the pub.

Check the result

Once you’ve got an answer, don’t just accept it and move on. Run a few checks.

Does it actually say when it’s not sure? A good answer includes phrases like “I can’t be certain” or “this needs a professional to assess properly.” If it sounds completely confident about a diagnosis, that’s a red flag, not a good sign.

Does it point you towards 111 or a GP rather than trying to resolve things itself? For anything with even a whiff of urgency, a decent response nudges you towards NHS 111 online, or 999 if it’s chest pain, stroke signs, breathing trouble, heavy bleeding, or an overdose. If it doesn’t mention any of that when the symptoms sound serious, be suspicious.

Does it avoid pretending it knows more than it does? Watch for it inventing details about you that you never gave it, or assuming your medical history. It shouldn’t be saying things like “given your history of…” unless you actually told it that history in the conversation. If it starts making things up about you, that’s the model filling gaps with guesses, and guesses are exactly what you don’t want here.

Try asking it directly: “How confident are you in this, and what would make you wrong?” A good assistant will answer that honestly. One that just doubles down isn’t being useful, it’s being convincing, and those aren’t the same thing.

If it doesn’t work

There are a few signs it’s time to stop using the chatbot and do something else instead.

If it keeps reassuring you and that reassurance doesn’t sit right, stop. Chatbots can be weirdly soothing, telling you it’s probably nothing, when actually you should just ring 111 and let a real person triage it. Reassurance from an AI is not the same as reassurance from someone qualified. Never let a calm-sounding chatbot talk you out of getting real care. If something feels wrong in your body, that feeling outranks anything on a screen.

If you’re pasting medical records that belong to someone else, stop immediately. That’s not your data to share, and it’s not appropriate to hand over regardless of how the tool markets its privacy. Similarly, be careful with your own sensitive details. Full names, dates of birth, NHS numbers, addresses, anything that could identify you alongside health information, think twice before typing it into a general-purpose chatbot. You don’t need to include your name for the AI to explain what “elevated troponin” means.

If you find yourself going back and back to the chatbot instead of booking the GP appointment, that’s a sign the tool has become a delay tactic rather than a help. It’s meant to prepare you for the conversation with a professional, not replace it.

On the privacy point more broadly: for anything you’d rather not put through a cloud-based assistant, particularly detailed personal health information, a locally run model on your own device is worth looking into, since nothing leaves your machine. That’s a more technical setup and not something a complete beginner needs to worry about on day one, but it’s worth knowing the option exists if sensitive detail is a concern.

And the golden rule underneath all of it: when in doubt, 111. Not the chatbot. 111.

Keep exploring

If you want to get better at spotting when an AI is guessing rather than knowing, have a look at “How to get AI to be honest with me”. If you want to check whether something it told you actually holds up, “How to get AI to fact-check something” walks through that. And if you’ve got a PDF hospital letter or a discharge summary you want help reading properly, “How to get AI to read a PDF” covers the upload and reading process in more depth than we’ve gone into here.

Sources and review notes

Review date: 15 September 2026 — every source above was opened and checked against the text on that date.