The trail, written out #
Practices talk about letters as though they were a single task. Followed end to end, one letter passes through seven states, and it can sit in any of them indefinitely without anybody being told:
| State | What it is waiting for | Who notices if it stalls |
|---|---|---|
| Promised | The clinician to write it | Usually the patient, on the phone, later |
| Drafted | A check and a signature | Nobody, unless somebody keeps a list |
| Signed | Sending | Nobody |
| Sent | Arrival at the right recipient | Nobody, until it turns out it did not arrive |
| Copied to the patient | A second send, often forgotten | The patient |
| Filed | Attachment to the record | Whoever needs it next, months later |
| Acted on | The recipient to do something | Nobody in your practice at all |
Look at the third column. Six of the seven states have no owner. This is the reason a letter can be four weeks old and nobody in the building knows, and it is also why "we need to write letters faster" is usually the wrong diagnosis. The writing was rarely the bottleneck.
What the published guidance actually asks for #
Two UK sources set the expectation here and both point the same way. The Academy of Medical Royal Colleges publishes Please write to me, first issued in 2018 and updated in April 2026, which asks clinicians to write outpatient letters directly to the patient.
NHS England said the same thing in its responsible officer information sheet on clinical letters in December 2018, noting that most patients and GPs prefer letters written that way. The Academy guidance also sets out what the writing should look like.
- Everyday language instead of jargon
- Any medical term explained
- A target of a Flesch reading ease score of 70 or above, with a UK reading age of 9 to 11
- Ask each patient whether they have accessibility needs affecting how they read and process information, and record that in their health record so other clinicians know
The last of those is a practical instruction that is pure admin and routinely skipped, and it is worth pausing on. It is a field on a record, checked before a letter goes out. Knowing whether it has ever been asked is a query. Asking it is a conversation.
Who owns the words #
The clinician who signs it. The General Medical Council's Good medical practice, in effect since 30 January 2024, carries duties on recording work clearly, accurately and legibly, and on delegating safely and appropriately. Neither of those changes because a draft was produced by something rather than someone.
This is the same line we hold everywhere on this site: the candidates are the admin around the care, not the care. A letter is not admin. Knowing which letters have not been signed is.
Which parts are genuinely rule based, and which are not #
Runs on rules
- A consultation with a letter promised and no draft after a set number of days
- A draft that has been waiting for a signature longer than your own standard allows
- A signed letter with no record of being sent
- A letter sent to a referrer with no copy to the patient, where one was intended
- A letter that bounced, or an address that has failed before
- A patient record with no accessibility preference ever recorded
- A referral made with no acknowledgement back from the receiving service
Every one of those is a state and a date, which is the definition of bookkeeping. None of them requires reading the letter.
Needs a clinician
- What the letter says
- Whether the referral is the right one
- Whether a finding needs a phone call rather than a letter
- How much detail a particular patient should be given in writing, and in what words
- Whether a letter should be copied to a family member
None of that is administrative, and treating it as administrative is the failure mode worth naming out loud.
That distinction, between watching the envelope and reading the contents, is the one that keeps this safe, and the general version of the test is on AI automation, explained.
The data protection point, briefly #
A referral letter is about as concentrated a piece of special category data as a clinic produces: identity, condition, treatment and often family circumstance in one document.
Your practice stays the controller for it whichever supplier touches it in transit, so a lawful basis and an Article 9 condition are needed, and misdirection is the risk that matters most. We point at the ICO rather than interpreting it, and the patient data page collects the relevant guidance in one place.
What to look at first #
Count the letters currently sitting in each of the seven states above. Most practices have never counted the middle three, and the number is usually the argument.
If the pile is in "drafted, unsigned", you have a clinician time problem. If it is in "signed, not sent", you have an admin problem, and it is a tractable one. The results and follow ups page covers the same shape of problem on the incoming side, and the free tools will turn your own counts into hours.