AI-generated voices can perform as strongly as human voiceovers in audio advertising, while locally matched synthetic accents may produce greater brand uplift, new UK research suggests.
The study was commissioned by omnichannel advertising platform Azerion and conducted by market research consultancy Differentology.
Researchers tested audio advertisements with 3,000 UK respondents during March and April 2026. Participants heard creative treatments for two household brands using a neutral human voice, a neutral AI voice, regional AI voices, or copy personalised through dynamic creative optimisation.
Overall brand uplift was 3% for both human and AI-voiced advertisements. The AI treatments were nevertheless considered more distinctive, more likely to attract attention, and more likely to give listeners a reason to choose the advertised brand.
The largest difference appeared when AI-generated regional voices were matched to a listener’s location. Among that group, 33% said they would recommend the brand, compared with 10% of those who heard the neutral human version.
Average uplift across the measured brand indicators reached 9% for the regionally matched treatments, against 3% for the neutral human voice. The accents tested were Geordie, Scottish, Yorkshire, and Welsh.
Listeners also struggled to determine whether voices were synthetic. Some 37% believed an AI-generated voice was human, while 29% correctly identified the AI treatment. Another 26% believed the human voice had been generated by AI.
Those results contrasted with expectations before exposure, when 39% of respondents believed that advertisements read by people would be more effective.
Ruth Reynolds, insight and strategy director at Azerion UK, said: “This research has closed the gap between synthetic and human voice creatives in advertising – and with that opened opportunities for brands to make the most of audio’s advantages. Those already using it can hone their campaigns, whether standalone or part of an omnichannel drive, with the new insight offered. Smaller brands meanwhile, previously deterred by the barriers of production and talent costs and long timescales, can introduce audio into their marketing repertoire.”
Dynamic creative optimisation also improved performance. Average brand uplift increased by 6% against the neutral AI benchmark across all respondents and rose to 24% among listeners who believed that personalisation represented a fair exchange for the use of their data.
That result links effective personalisation with perceived consent and value. A campaign may use accurate location data and still underperform when the listener does not understand why it was used or considers the benefit too small.
Audio production has traditionally involved fixed costs covering talent, studios, direction, editing, alternate scripts, approvals, and repeat sessions. Synthetic voice tools can reduce the cost of producing multiple versions, allowing campaigns to be adapted according to location, audience, weather, time, or other contextual factors.
Lower production costs may allow smaller advertisers to use audio more frequently, while larger brands can test a greater range of treatments instead of applying one national voice to every market.
Channel 4’s expansion of postcode-level streaming advertising has already demonstrated the growing role of location in media planning. Synthetic audio adds another layer by changing not only where an advertisement appears, but how the creative sounds.
Regionalisation requires careful control. An accent can convey familiarity, although it can also feel artificial, stereotyped, or patronising when the voice, script, or context does not reflect the location accurately. Local review will remain necessary rather than assuming that geographic matching automatically creates authenticity.
Voice rights create a separate contractual issue. Advertisers and agencies need to establish how a synthetic model was trained, whether the performer consented to the intended uses, and whether agreements permit future adaptation. The capacity to generate unlimited variations may create disputes where talent contracts were designed for conventional recording.
Brand governance becomes more difficult as the number of creative versions increases. A campaign containing hundreds of audio treatments needs controls over pronunciation, product claims, tone, regulatory wording, and suitability. A small mistake can be repeated across several markets before a human team identifies it.
Disclosure also remains unsettled. The study suggests that many listeners cannot identify a synthetic voice, but commercial effectiveness does not resolve whether consumers should be told when AI has been used.
Expectations may vary according to the product, the apparent identity of the speaker, and whether the voice resembles a recognisable individual. Financial services, healthcare, political communications, and recruitment may also carry different trust requirements from ordinary consumer advertising.
The research itself was conducted using two household brands in controlled test and survey conditions. Results may not transfer uniformly to every sector, media environment, audience, or type of purchasing decision.
The production advantage lies in the ability to create and test controlled variations at lower cost. Replacing one human read with one synthetic read offers a narrower benefit than adapting creative according to audience and context.
Successful use will require media targeting, consent, voice rights, brand controls, and local judgement to operate together. Synthetic production can increase speed and lower expense, but it can reproduce weak creative just as efficiently as strong creative when governance is absent.
You must be logged in to post a comment.