What changed
On 15 October 2026, Which? published a test of the selfie skin scanners on retailer and beauty brand sites. Senior researcher James Aitchison and colleagues found 17 online tools, including Boots, Superdrug, La Roche-Posay, L'Oréal, and Neutrogena. Which? said at least seven, Boots included, run on a white-label engine it called Revive. L'Oréal, it said, uses its own ModiFace framework.
The measured tool was the Boots Skincare Diagnostic Tool. Thirty people completed more than 120 scans in early September 2026. Each person filled in the pre-scan questionnaire, ran two scans back to back in good light, then a third with deliberately wrong answers, then a fourth in dim or patchy light. Which? told Boots what it found. Boots did not answer how products are chosen. It did say the tool was no longer available online.
A separate, older source uses a close spelling. On 18 October 2022 Revieve announced that Boots had launched a "Boots Skin Analysis" service on boots.com. That release does not describe the 2026 test. It shows that a Boots selfie advisor and the Revieve name have been public for years. Which? spelled the engine Revive.
Figure: A phone held to the face for a retail skin scan. Image: Which?
Why it matters
When light and distance stayed the same, the overall score barely moved. The average shift between the first and second scan was less than one point. Only four of the 30 testers saw a change of five points or more, and the top and bottom categories, plus the suggested products, stayed almost the same.
The misses were the local issues people had already named. Of 14 testers with visible acne who listed it as their main concern, the scanner failed to spot it or contradicted them 10 times. Of nine testers with dark spots, the app rated pigmentation as perfectly healthy six times. Of seven testers with redness, not one had redness flagged, and three received a top score for clear skin. Which? said about 63 percent of scores matched the mirror, and 37 percent did not.
Light moved the number much more than a second scan in the same room. For about half the group, a change of light swung the overall score by as much as 39 points up or 25 points down. The app will not take a photo when the room is too dark, so the warning is on the screen. The swing still happened inside lighting the app accepted.
Skin tone changed the report. Which? said many of these tools struggle on darker skin, Fitzpatrick types IV to VI, because training sets often underrepresent them. In this tryout the app struggled to detect hyperpigmentation on participants with types IV and V, and it gave them higher overall health scores on average than testers with lighter skin. Lighter skin saw the larger swings when the light changed.
The questionnaire did not move the overall score much. It did change the category scores and the shopping list. One tester with clear skin, scored as having little or no acne, then selected spots as a concern and received a low acne score. Twenty-five of the 30 testers got a completely different morning product list after they changed their answers. The default list was four morning products and five evening products. L'Oréal was suggested 214 times and Cosrx 154 times. Olay and La Roche-Posay barely appeared.
Figure: Which? published this example of scanner output. The numbers on the screen are the example, not the averages from the 30 person test. Image: Which?
Who is affected
Shoppers on Boots, Superdrug, and brand sites are the people who see the score and the product row. Retailers that embed a white-label engine, and the brands that dominate the list, decide what the selfie is allowed to sell.
Which? asked consultant dermatologist Dr Emma Wedgeworth for a view. She said a tool can help someone narrow a crowded shelf, and that these screens must not be confused with a medical diagnosis. She said labels such as radiance, dullness, normal, combination, and oily are not recognised in clinical dermatology, and that anyone with a skin condition should seek professional help. This desk is not adding a treatment claim on top of that.
Which? also recorded one product list that shows the limit. A tester with rosacea, who had said the skin was sensitive and dry, was recommended products containing retinol and salicylic acid. The tester told Which? those ingredients would irritate the skin. That is the tester's report of what the tool suggested. It is not advice from this desk to use or avoid any ingredient.
What to do next
Treat a retail skin score as a shopping hint taken under one set of lights. If a scanner is still online, use the lighting the app asks for, and read the ingredient list before you buy. Do not treat the number as a diagnosis.
Teams shipping these scanners should publish how the ranking is chosen. A score that misses acne the person can see, and that jumps when the lamp moves, is a retail feature with a known error bar. Calling the screen a diagnostic sets an expectation this test did not support.
What to watch
Which? said several high profile tools, including the Boots one, have been pulled, and that others remain. The next useful signal is a public note from Boots or from Revieve on the September test, or a fresh scan count from Superdrug or from a brand site Which? listed. Until that note exists, the 15 October write-up is the record of what the Boots tool did in matched light, in changed light, and on the concerns testers named.
Sources
- Primary — Which?, Can you trust AI skin analysers? (15 October 2026). Establishes the 17 tool survey, the 30 person Boots test, the score and miss counts, the lighting swings, the product counts, the removal of the Boots tool, and the Wedgeworth comments.
- Secondary — Revieve, Boots and Revieve launch an AI skincare advisor (18 October 2022). Establishes the 2022 Boots launch. It does not establish the 2026 test results.