Five Stars, Zero Human: How to Spot a Bot-Written Tech Review Before You Buy

You are two clicks from buying a pair of headphones. The listing has 4.7 stars, a few hundred reviews and a top comment that calls the sound 'rich but never muddy' and the app 'a breeze'. It could be a genuine owner. It could also be a language model with a thesaurus and a deadline.

That is the small, annoying problem at the centre of UK tech shopping in 2026. Reviews are still one of the few places where buyers expect to hear from other humans, yet the cost of producing convincing fake ones has collapsed. The result is not a world where every review is fake. It is a world where the signal is noisier, and the star rating is doing more work than it should.

The star rating is easier to move than it looks

Synthetic reviews rarely need to fool you for long. They need to do a job: move a product from 4.1 to 4.5, push a listing onto the first page, or bury an early batch of complaints. AI makes that cheaper because it can produce dozens of reviews in different tones, with UK spelling and plausible complaints, without a room full of people and a spreadsheet.

The tell is often not one review. It is the pattern. A sudden burst of five-star comments over a weekend. A reviewer who has also reviewed a phone case, a garden hose and a robot vacuum in the same week. Uniform paragraphs that all mention the same three features in the same order. A product with lots of praise for 'premium build quality' and almost nothing about the annoying firmware update that changed the app layout.

That last point matters for phones, headphones and smart home gadgets. These are products with quirks. Real owners notice quirks. Bots tend to notice brochure copy.

What UK law and the platforms actually do

In the UK, misleading commercial practices are on the regulator's radar. The Digital Markets, Competition and Consumers Act 2024 gave the Competition and Markets Authority stronger tools, including powers to tackle fake and misleading reviews. Those provisions have been brought in through a staged commencement process rather than one dramatic switch, so the practical picture is a moving one. The direction is clear, though: a seller that plants reviews or suppresses bad ones is not just being annoying, it is in the territory of consumer law.

Platforms have their own defences. Amazon marks 'verified purchase' reviews, which helps but does not prove the reviewer is human or unbiased. Google says it uses machine learning to remove policy-breaking reviews, and Trustpilot has fraud-detection teams that look for patterns rather than individual comments. None of these systems is perfect, because the game is cat and mouse. Detection improves, synthetic text improves, and the shopper is left doing a bit of homework.

One nuance is worth keeping: a genuine owner who uses AI to tidy their grammar is not the same as a seller flooding a listing with invented opinions. The problem is deception, not the keyboard. If a review describes a real purchase, real frustrations and real use over time, the fact that it was polished by software is not the scandal.

The tells that survive better grammar

AI writing has become smoother. It has not become better at having a life. That is where to aim your scepticism.

Specificity beats polish. A real headphone review might mention clamping force after two hours, the difference between the ANC on a train and on a busy street, or the exact firmware version that fixed a connection drop. A bot prefers 'excellent sound quality' and 'comfortable for long use'. Both can be true. Only one reads like a person who has actually worn them.

Timelines are hard to fake well. Look for phrases like 'after a month', 'three washes in', 'the first update broke it'. Synthetic reviews often live in a permanent present tense. They describe the product as a finished experience, not something that changed.

Beware the perfect balance. A review with three generic pros, three generic cons and no real trade-off can be a bot's idea of credibility. Real owners tend to be lopsided. They care about one thing more than the rest, and they go on about it.

Check the reviewer's back catalogue. If someone has posted twenty five-star reviews in a fortnight across unrelated categories, treat their verdict as entertainment rather than evidence. This is especially useful on marketplaces where profiles are public.

Look for the boring friction. Delivery delays, app sign-in problems, a smart plug that would not connect to a specific router band, the fact that a phone case blocked a particular cable. Bots can invent friction, but they often invent the same friction. Real complaints are oddly specific and slightly petty, which is exactly what makes them useful.

Photos are not proof. A photo can be reused, generated or borrowed. A messy kitchen counter and a scuffed gadget are mild evidence. A studio-perfect shot is not.

Read the threes and twos first. Five-star reviews tell you what the seller wants you to feel. Three-star reviews tell you what the product is actually like when the novelty wears off.

Three quick checks before you tap buy

First, search the listing for a phrase. If the same sentence appears in several reviews with a word swapped, you have found a batch. Second, cross-check the product against a specialist forum, a YouTube teardown or a UK consumer group. You are not looking for a single definitive verdict; you are looking for a consistent complaint. Third, check the returns policy. A generous return window is a better safety net than any star rating, especially for smart home kit that may not play nicely with your existing setup.

If you do spot a suspicious review, report it. It takes a minute and it feeds the pattern-detection systems that platforms rely on. Then leave your own review if you bought the thing. Specific, boring, honest reviews are the cheapest defence the rest of us have.

AI reviews will get better. They will get more British, more nuanced and better at pretending to have owned a phone for six months. The counter is not cynicism. It is a habit: look for lived detail, check the pattern, and treat a five-star average as a claim to be tested rather than a conclusion.