AI Guides › Workbench

One Health File an AI Can Read: Building Your Own Health Ledger

By Nigel Guy · 8 min read

The usual way to use AI for your health goes like this: you photograph one blood test result, paste it into a chat, get a reassuring paragraph back, and close the tab. The next time you go, the AI has no idea what your last result was, which medicines you take or what the GP said in March. Each answer sounds sensible, but none of them takes your history into account, and history is the part that matters.

The rule: keep one plain, dated file of your readings, results, medicines, notes and appointments, and only ask the AI questions about that file. Without the file you're getting guesswork.

The kit at a glance

Tool What it does Cost at time of writing Best for Catch
NHS App Shows your GP record, test results, documents, prescriptions and appointments Free Getting the raw data Your surgery decides what you can see. You can't download test results from the app.
The Health Ledger (a document you keep) Holds everything in one dated, structured file Free (any word processor or notes app) Being the single source of truth It only stays useful if you keep it up to date
Claude Project Keeps the ledger and standing instructions so every chat starts with your history Free plan allows up to 5 Projects. Paid plans are listed in US dollars, so check the £ price at checkout. Asking questions across all your records It's a general assistant, not a medical device
ChatGPT Health / Claude health connectors Pull medical records and wearable data in automatically Varies by plan Automatic syncing US only at time of writing. Skip them if you're in the UK.

Item 1: The NHS App, your source

In England, the NHS App is where most of your raw material lives. According to the NHS help pages, it has these sections:

Wales, Scotland and Northern Ireland have their own services, which this guide hasn't checked. The rest of the kit works the same.

Item 2: The Health Ledger, your file

This is the "dashboard". It's a single document with five fixed sections, and every entry is dated. Skip the photos and screenshots: AI tools misread numbers in images more often than in typed text, and you can't check a photo by searching it.

Section What each entry needs Example format
Readings Date, measure, value, unit, context 2026-09-14 · Blood pressure · 138/86 mmHg · morning, seated, before coffee
Lab results Date, test, value, unit, the lab's reference range, who ordered it 2026-08-02 · HbA1c · 44 mmol/mol · range shown on report · GP
Medicines Name, dose, how often, start date, reason, stop date if stopped Ramipril · 2.5 mg · once daily · started 2026-05-10 · BP
Doctor's notes Date, who, what they said or decided, in your own words, and what happens next 2026-09-20 · GP · wants repeat bloods in 3 months
Appointments Date, place, purpose, what to bring 2026-12-01 · surgery · repeat bloods · fasting

Always copy the unit and reference range from the report, because labs differ and an AI will otherwise guess. Don't delete stopped medicines: give them a stop date, because "what changed and when" is the question you'll ask most.

Item 3: A Claude Project, your workspace

Anthropic's help centre describes Projects as self-contained workspaces with their own chat history and knowledge base, plus instructions that apply to every chat inside them. Free accounts get up to five. Paid plans add retrieval, which helps with large documents. Set one up like this:

  1. In Claude, create a new Project called something dull like "Health Ledger".
  2. Upload the ledger as a file in the Project's knowledge.
  3. Paste the instructions below into the Project instructions.
  4. Whenever you add new entries, replace the old ledger file with the new version so there's only one copy.

Fill in the square brackets before pasting it in.

You are a careful health-records assistant helping me, [YOUR_FIRST_NAME], understand my own records. You are not my doctor and you do not diagnose or recommend treatment changes.

Context: The attached Health Ledger is my only source of truth. It has five sections: Readings, Lab results, Medicines, Doctor's notes, Appointments. Every entry is dated. I live in [COUNTRY], and my relevant conditions are [CONDITIONS_OR_"NONE_RECORDED"].

When I ask a question:
1. Find the relevant entries in the ledger and quote them with their dates.
2. Answer using only those entries. If the ledger doesn't hold the information, say "not in the ledger" and tell me what I would need to add.
3. Use the reference range printed in each lab entry. Never substitute a general range without saying so.
4. Point out gaps, contradictions or entries more than [MONTHS] months old.
5. List anything worth raising with a clinician as a question, not a conclusion.

Format: a short answer first, then a table of the entries you used (date, item, value), then "Questions for my GP or pharmacist".

Guardrails: if anything suggests something urgent, tell me to contact NHS 111 or 999 and stop there. Ask me for missing details rather than assume them.

Before replying, check: did every number you quoted come from the ledger, with its date?

Item 4: The three questions that make it worth having

Use these inside the Project.

1. What changed since my last appointment? Fill in the appointment date.

My last appointment was on [DATE]. Using only the ledger, list everything that has changed since then: new or stopped medicines, dose changes, new results compared with the previous result for the same test, and readings that trend up or down. Show each change as old value → new value with both dates. End with up to five questions I could ask at my next appointment, ordered by importance. If you can't compare a result because there's no earlier value, say so.

2. How has one measure moved over time? Fill in the test or reading.

Show every ledger entry for [TEST_OR_READING] in date order as a table: date, value, unit, reference range, context. Then describe the trend in two plain sentences without interpreting what it means for my health. Flag any entries where the unit or reference range differs from the others, because those values may not be comparable. If there are fewer than three entries, say the trend can't be judged yet.

3. Is my medicines list straight? Fill in what you're actually taking today.

Here is what I am actually taking today, from the packets: [LIST_NAME_DOSE_FREQUENCY]. Compare it with the Medicines section of the ledger. Produce a table with three columns: in both and matching, in the ledger but not in my list, in my list but not in the ledger (including over-the-counter items and supplements). Do not judge whether anything is safe to combine; instead, write the exact question I should ask my pharmacist about each mismatch.

How to choose

What to skip

Guardrails

Sources

All 751 AI guides · JulieMango plans from £17/mo