TL;DR

Royal Mail has confirmed to Deadline that the narrator of its 2025 Christmas advertising campaign was generated by AI, a fact it did not disclose when the ads ran on ITV. Lucy Wynne, daughter of the late broadcaster Paul Vaughan, believes the voice is her father’s, and a forensic speech scientist at the University of York found “considerable phonetic similarity”. Deadline published the exclusive on Friday 9 October.

Whose voice?

Vaughan, who died 11 years before the ads aired in November 2025, was heard on the BBC’s Horizon, Radio 4’s Kaleidoscope and Paddington audiobooks, and narrated the drama Threads. That body of recorded work, Deadline notes, is just the sort of material that could train a voice model.

Royal Mail says the voice, which delivers the closing line “Royal Mail, it’s so much more than just a card”, was “not designed to replicate any specific individual”, and that the producers’ checks raised no resemblance concerns. Its agency, Darwin & the Machines, passed the voiceover to Fameplay, a Czech AI specialist backed by Czech Media Invest. Its chair is Daniel Křetínský, and his EP Group owns Royal Mail.

Fameplay told Deadline its prompts made no reference to Vaughan and that it licenses technology from ElevenLabs. ElevenLabs would confirm only that it was no direct clone of Vaughan, and declined to say whether its model had been trained on his recordings. It has since put him on its No Go Voices list, which blocks direct cloning on its platform.

The evidence

Professor Vincent Hughes at York found overlap in how consonants and vowels were pronounced and in the overall quality of the voice, although the short clip limited his analysis. Relatives are very good at picking out a loved one’s voice, according to research by Queen Mary University of London voice psychologist Dr Nadine Lavan. Deadline also ran three open-source AI comparison tools, all of which suggested a good chance the two voices came from the same speaker.

Wynne says she has had no apology. Royal Mail told Deadline it understood the concerns and had taken them seriously.

Looking forward

In our view, the case shows a gap that consent rules built around deliberate cloning do not close: a generic prompt can still produce a voice a family recognises, and a producer’s checks may not catch it. It adds to the pressure covered in our reporting on the UK legal gap around voice cloning and on performers asking for a legal right to their voice. Brands commissioning synthetic voices would be wise to run a resemblance check against well-known voices before broadcast, and say plainly when a voice is not human.