Every NHS service manager has, at some point, watched a colleague hold up a phone to a patient and hope for the best. It is quick, it is free, and it is currently keeping NHS England’s patient safety team up at night.
NHS England’s Improvement Framework for community language translation and interpreting services, published in May 2025, put this plainly: while translation apps provide a convenient, familiar and timely means of translation, they can also carry risks, particularly regarding accuracy and the potential impact on patient safety. That is not a ban. It is an invitation to think harder about where the line actually sits – and to write that line down, rather than leaving it to whoever is nearest the phone charger.
Why “just use Google Translate” is no longer a defensible policy
The National Register of Public Service Interpreters has been blunter still, advising that automated online translating systems or services such as Google Translate should be avoided in healthcare settings as there is no assurance of the quality of the translations. Meanwhile, NHS England’s own framework acknowledges the scale of the problem: there are concerns about the appropriate use of AI translation apps that are currently widely used across the NHS to communicate with patients with limited English proficiency.
For solicitors advising trusts, ICBs or GP practices on governance, this matters because the gap between “widely used” and “properly governed” is exactly where clinical negligence and consent claims tend to live. A tool that is informally everywhere but formally nowhere in the policy file is a liability sitting in plain sight.
The three-tier framework
Rather than a blanket rule, NHS services need a tiered sign-off approach that matches the tool to the stakes. Here is a structure that service managers can adapt and put in front of their safety and governance leads.
Tier 1 – Green: low-risk administrative use
Think appointment reminders, directions to a clinic, opening-hours notices, or translating a poster about the flu jab clinic. NHS England’s own case study showed Frimley ICB using exactly this approach, translating reminder texts into the area’s most common languages – crucially, with bilingual NHS staff checking AI-generated translations to ensure accuracy and check nuances.
- Machine translation permitted for drafting.
- Human review by a bilingual staff member or linguist before publication, even for low-stakes content.
- No patient-specific clinical data involved.
- Sign-off: service manager or communications lead.
Tier 2 – Amber: routine clinical administration and triage
This covers booking confirmations referencing a specific condition, routine history-taking support, or discharge logistics (follow-up dates, where to collect a prescription). The risk is higher because patient-specific information is involved, but the decisions being supported are not irreversible.
- Machine translation may support, but should not replace, a professional interpreter for the actual consultation.
- Any machine-translated patient-facing document must be checked by a qualified linguist before it leaves the building.
- Clear audit trail: date, tool used, who reviewed it.
- Sign-off: clinical lead plus information governance.
Tier 3 – Red: consent, medication and safeguarding
This is the zone where machine translation alone should not be used, full stop. Consent discussions, medication dosing instructions, mental health assessments, safeguarding disclosures and end-of-life conversations all sit here. The legal literature on this is consistent: working with professional interpreters will minimise legal risk of misinterpretation of important clinical information, for example informed consent to undergo clinical treatments and procedures.
Commissioning guidance for primary care is equally direct about who should not be doing this job informally. The use of anyone under the age of 16 for interpretation is not acceptable in any circumstance other than when immediate and necessary treatment is required. If a child should not be interpreting a consent conversation, an app with no concept of clinical nuance certainly should not be either.
- Professional, qualified interpreter required – telephone, video or face-to-face depending on acuity.
- No unreviewed machine output presented to a patient or recorded as the basis for consent.
- Where a family member is the patient’s preference for support (not interpreting), the patient should give informed consent in their own language, sought from them independently of the family member or friend, and the consent must be noted in the patient’s record.
- Sign-off: responsible clinician, with safeguarding or consent documentation cross-referenced.
Building this into a sign-off framework, not just a poster
A tiered checklist only works if someone actually owns it. NHS England has called for exactly this kind of structure nationally, recommending that organisations ensure clinical safety and accuracy of AI outputs, particularly for sensitive tasks like medication instructions, provide clinical assurances and governance frameworks including indemnity and responsibility for AI use, and outline when AI tools are suitable and when alternative methods should be prioritised.
Practically, that means:
- Name an owner. Someone – information governance, patient safety, or both – signs off which tools are approved, for which tiers.
- Record language need at the point of registration, not at the point of crisis, so Tier 3 situations are never a surprise.
- Train reception and admin staff on the difference between Tier 1 and Tier 2 use, since this is where informal app use creeps in fastest.
- Audit, don’t assume. Spot-check what staff are actually using against what the policy says is approved.
Where a human interpreter earns its keep
For anything in the amber-to-red range, a professional interpreter remains the safer, more defensible choice – not because the technology is hopeless, but because nobody can cross-examine an app about why it chose a particular word. Certified translation of consent forms and discharge letters, backed by a documented review trail, gives trusts something they can actually stand behind if a case is ever scrutinised later.
Connect has been supporting NHS teams, alongside courts, solicitors and the Legal Aid Agency, with interpreting and certified translation since 2005. As the owner puts it, using a service provider who is well established like Connect can help – particularly when a governance committee wants evidence of process, not just good intentions. You can find out more about the agency here.
A policy worth writing down
None of this requires banning AI tools outright – NHS England itself is pushing for sensible, documented use rather than prohibition. What it does require is a line, drawn in advance, between “this can wait for a human check” and “this cannot go near a patient without a qualified professional in the room.” Get that line agreed, signed off and reviewed, and the next safety audit becomes considerably less eventful.
If your service needs help setting up interpreter cover for Tier 2 and Tier 3 situations, get in touch with Connect or use our online booking request to arrange a professional interpreter.
Frequently asked questions
Can NHS staff use Google Translate or similar apps at all?
NHS England's own framework warns that these tools carry accuracy and patient safety risks, and guidance from bodies like NRPSI recommends avoiding them in healthcare settings. Low-risk administrative uses, reviewed by a bilingual staff member before anything reaches a patient, are more defensible than using them unchecked for clinical content.
Is it ever acceptable to use a family member instead of a professional interpreter?
NHS primary care guidance says that if a patient expressly wants a family member or friend involved, the patient's informed consent to this must be obtained independently, in their own language, and recorded in their notes. Children under 16 should not be used to interpret except in immediate, necessary treatment situations.
Who should sign off on machine translation use in an NHS service?
NHS England recommends clear governance, including named responsibility and indemnity arrangements, for any AI translation use. In practice this means a named clinical or information governance lead approving which tools are permitted for which tier of use, with regular audits of actual staff practice.