Can you spot a deepfake?

Which? researchers put the general public to the test to see whether they could identify deepfake videos and found that, more often than not, people cannot reliably spot AI fakery.

When asked to spot the deepfakes from a selection of real and fake videos, 70% failed to correctly identify all of the videos shown to them. Watch the videos below and read our clues to see if you picked up on the tell-tale signs.

A version of this article was originally published in Which? Tech magazine, Jun/Jul 2026 issue.

Video: Can you spot deepfakes?

We took two deepfake videos and two genuine videos (not created using AI) that had been circulating on social media.

A representative sample* of the UK adult population watched the videos – we showed a random selection of three to each participant and asked them to tell the real from the fake.

Can you spot the signs that these videos are AI-generated? Read on to see what we noticed.

*Our research: The results are based on a survey of 514 members of the public and are representative of the UK population aged 18+. Each participant saw three videos out of a potential four (two real, two fake). Data collection was conducted online, March 2026.

About the videos

When we originally identified these videos on social media, both led to dodgy phishing websites, most likely set up to steal your personal and payment information. 

The Boots video is the most convincing. The main clue to spotting it was that it wasn't posted from the official Boots verified page.

Boots confirmed that the advert is fake and urged its customers to be vigilant. A spokesperson told us: 'We proactively monitor for this type of content on social media and report any we identify to the platform owners and request that they are urgently removed'.

We found deepfakes on social media sites

Most of the videos we found were adverts placed on social media sites, meaning the platforms they were posted on profit from them. Scam ads on social media continue to be a massive problem, as we have uncovered and reported on time and time again.

Digital technology analyst Juniper Research estimated that in 2025 alone, social media companies made nearly £3.8bn in revenue from scam ads in Europe and that these scam ads were viewed almost 1 trillion times (known as ‘impressions’, in social media jargon).

Meta is being sued in the US by the not-for-profit Consumer Federation of America for allegedly failing to protect users from scam ads. Meta responded by stating that the allegations misrepresent the reality of its work and it will fight them. Meta also said that it aggressively combats scams on its platforms.

What needs to be done

We believe that the most effective way to disrupt the fraudsters behind these deepfakes and scam ads would be to make tech giants take responsibility and prevent the flood of scams from appearing on their platforms to begin with.

Deepfake videos are just the latest tool in a long line of means and methods that have made scams more believable.

They get more convincing by the year. Whether it’s appearing at the top of a search engine’s results page, placing an advert on social media or getting verified with a blue tick to appear more genuine, fraudsters are constantly using new online developments that are intended for legitimate businesses.

What stands between them and us are the social media companies that monopolise internet traffic. But if they’re not properly vetting what gets onto their platforms, we’re all at risk of being exposed to scams, forcing us to rely on an evolving handbook of tactics to spot them.

If legislation and enforcement continue to lag in forcing big tech platforms to block scams, this handbook will have to keep growing.  

Keep reading to find out how our scams expert, Tali Ramsey, spots deepfakes.

Is Big Tech doing anything?

The only way to distinguish between real and deepfake videos on social media is content labelling, where a video on a website or app is marked as being AI-generated.

An Ofcom survey in 2024 found that around six in 10 respondents thought it was very important for online platforms to tell them when content had been made or edited using AI.

However, big tech platforms have different approaches to labelling AI-generated content, as we found when we looked at the policies on their websites.

• Meta (which owns Instagram and Facebook), TikTok and YouTube label AI-generated content using digital labels attached to videos by the AI platforms they were created on. They also enable users to disclose that the content they’re uploading has been created using AI.
• YouTube states that, in some cases, it uses other methods to detect AI-generated content – but it’s not clear what those methods are.
• Snapchat adds AI content labels to AI images or videos that are created on its platform, but its website says it can’t guarantee that AI content made away from its platform will be labelled.
• Reddit doesn’t label AI-generated content but encourages disclosure.
• On X, users can voluntarily label their own posts containing AI-generated content.

None of the videos we found on Meta platforms were labelled. We shared the videos with Meta and it said it had removed them.

What about the government?

Earlier this year, the government published its long-awaited fraud strategy, outlining how it was going to tackle one of the country’s biggest crimes.

In it, the government stated how it would design and implement a framework to detect deepfakes.

However, overall, it failed to adequately address actually stopping fraud from reaching the public in the first place – something that Which? has been campaigning on for years.

The Online Safety Act, which came into force in March 2025, was implemented to make big tech platforms accountable for fraudulent content on their platforms. It applies to all scams, including those that are AI-generated.

However, due to a legal challenge to the implementation of part of the Act, we most likely won’t see the rules around scam ads come into force until 2027 at the earliest.

Find out more about the Which? take on the fraud strategy

What next?

Rocio Concha, Which? Head of Policy and Advocacy, said:

'Our investigation into deepfakes on social media shows that the public – and even experts – are increasingly at the mercy of scammers exploiting inadequate security on social media platforms. 

'This is troubling, as we believe that the tech giants who profit from scams on their platforms will not take enough action against fraudsters unless they are legally compelled to do so. 

'Ofcom must stop kicking the can down the road and implement the Online Safety Act’s measures on fraudulent advertising as soon as possible, and the Government must ensure that it does so urgently and robustly.'

How I spot deepfakes

Checking the profile behind a post is important. If it’s been posted from an unfamiliar profile, supposedly on behalf of a well-known brand, avoid it. Anything that sounds too good to be true is often a red flag.

These are still the most reliable ways to spot all types of scams circulating on social media, including deepfake videos.

Categorically being able to say whether a video is a deepfake often requires more detailed analysis.

While lip syncing, robotic voices and blurry details still show up on some videos, their sophistication necessitates new detection methods.

Try looking beyond the obvious, such as shadows and reflections in the eyes that don’t match the scene, and keep an eye out for emotional mismatches between facial expressions and the theme of the video.

For example, is the person smiling while discussing a serious topic? Or is there something odd about the light, such as a missing shadow?

More tips to try

• Study videos frame by frame to spot inconsistencies. Is there a facial feature or something in the background that disappears, or does the person in the video age or experience a slight change in their skin, hair or eye colour over the course of the video?
• Use reverse image searches by uploading images – select the camera icon on Google search or visit TinEye. This could lead you to other versions of the deepfake video, which may have more obvious signs of being fake.
• This is slightly challenging, but if you know how, you can also open the video on your computer’s default video program and play around with the brightness to reveal hidden watermarks.

Get tech help from humans - save 25%

Solve your tech issues and get expert buying advice by chatting to our support team as often as you need for only £36.75 year.

Buy and save

Save 25% – was £49, now £36.75 for a year, offer ends 8 June 2026