Most people assume that if a cataract operation goes well, no one ends up in a courtroom. The data says otherwise. 

Cataract surgery is the most commonly performed surgical procedure in the NHS, with around 476,000 operations carried out in England and Wales each year. It is also one of the safest surgeries we offer. Yet it generates the largest share of clinical negligence claims within ophthalmology and contributes substantially to the specialty's total litigation costs. 

What is striking is that a good visual outcome does not protect against complaint or claim. When the analyses are done, a large proportion of cases can be traced back to patients feeling poorly informed, misunderstood, or unheard and not to surgical error. The procedure may go well, but the conversation around it does not. 

Why consent breaks down

Informed consent in the NHS is usually delivered through a brief face-to-face discussion supported by a written leaflet. In a high-volume setting like cataract surgery, that conversation is often short, time-pressured, and shaped by the individual clinician. Evidence shows patients only retain a small proportion of what they are told verbally, and the quality of consent discussions varies widely between clinicians and clinics. 

In London, 8-9% of the population report limited proficiency in English, with rates higher in older age groups, exactly the demographic most likely to need cataract surgery. The standard NHS leaflet is in English. Patients who cannot read it comfortably are, in effect, consenting on the strength of a rushed verbal exchange alone. 

These are the conditions in which misunderstandings and complaints quietly accumulate. 

The intervention

We designed a single-masked, randomised controlled trial at Moorfields Eye Hospital to test whether an AI-generated patient information video could improve the consent process. The video was built using Synthesia, a platform that uses generative AI to produce photorealistic avatars and supports more than 120 languages. The script was based directly on the existing Moorfields cataract leaflet, written in English, then translated into Bengali and reviewed by a native Bengali-speaking consultant ophthalmologist for clinical accuracy. 

East London serves a large Bengali-speaking population, and Bengali is the most commonly spoken language locally after English. If a multilingual approach can work anywhere, it should work here. 

We recruited 203 consecutive patients listed for first-time cataract surgery, 162 English-speaking and 41 Bengali-speaking, and randomised them to either standard care (face-to-face consent plus the leaflet) or the intervention arm (the same, plus the AI video). Both the consenting clinician and the operating surgeon were blinded to allocation. We measured information retention via a quiz one week later, anxiety at baseline and on the day of surgery, satisfaction with the consent process, and surgical cancellation rates. 

What we found

The video worked for understanding. Patients who received it scored significantly higher on the information retention quiz than those given the leaflet alone (4.16 vs 3.48 out of 5; p < 0.001). They also engaged with the materials for longer (25.6 vs 18.0 minutes) and greater engagement time was itself strongly associated with better quiz scores. This is one of the more useful findings of the trial: digital platforms allow us to see engagement in a way paper leaflets never could, meaning engagement itself can be a modifiable target. 

Importantly, there was no significant difference in retention scores between English- and Bengali-speaking participants, suggesting the multilingual format succeeded in closing the comprehension gap. 

But understanding is not the whole story. Anxiety rose from baseline to the day of surgery across every group, regardless of language or intervention arm. Better information did not calm patients down. While retention equalised across language groups, satisfaction did not: Bengali-speaking patients reported significantly lower satisfaction with the consent process than English-speaking patients, with non-significant trends towards higher anxiety and more cancellations. 

In other words, the video can help with what patients know. It cannot, on its own, help with how they feel about the care they receive. 

What this means

Our embedded qualitative work, group interviews with a subset of participants, pointed toward what the numbers could not capture. Patients spoke about trust, about cultural context and about wanting to feel that the staff in front of them knew who they were. They spoke warmly about specific people: a volunteer who held their hand during the anaesthetic, a nurse who explained things slowly. Several were clear that while a translated video would help, it would not substitute for a clinician who took the time to look up from the screen. 

For a litigation-prone, high-volume procedure like cataract surgery, this matters. The technology is now genuinely available: AI-generated multilingual video is cheap, fast to update, and scalable in a way traditional filmed content has never been. It can demonstrably improve patient understanding and address one real component of health inequity. But it works best as a complement to good clinical communication, not a replacement for it. 

The next steps for our group are to look at how this kind of intervention performs at scale, across more languages, and whether sustained improvements in understanding and engagement translate, over time, into reductions in complaints and claims. We will also continue to listen carefully to what patients themselves are telling us, because the loudest message from this trial is that the best consent process needs both scalable technology, and human attentiveness.