Free versions of ChatGPT, Claude, Gemini, and Amazon's Alexa were tested a week ahead of Black Friday to find discounts on a 55-inch television. Across the board, the systems generated frustrating errors, including dead links, out-of-stock items, and misstated costs like a £60 pricing error from Gemini. The inconsistent results arrive just as tech firms roll out specialized shopping features, even though consumer skepticism remains high in markets like the UK.
Key points
Leading artificial intelligence platforms currently deliver unreliable recommendations, o…: Leading artificial intelligence platforms currently deliver unreliable recommendations, outdated prices, and broken links when tasked with finding retail bargains.
Tech firms continue to introduce commercial features like OpenAI's Shopping Research tool…: Tech firms continue to introduce commercial features like OpenAI's Shopping Research tool despite the persistent performance limitations of existing systems.
Low regional adoption rates indicate that consumers remain deeply hesitant to trust artif…: Low regional adoption rates indicate that consumers remain deeply hesitant to trust artificial intelligence with their holiday shopping needs.
Sources checked by our analysis pipeline: 1
- Published
Whether it's scrolling through endless choice, spotting fake reviews, or working out if a "deal" is a deal at all, Black Friday sums up a lot of online woes.
Retailers promise bargains - but the sheer number of offers can leave us feeling no closer to finding the best deal.
It's an area where generative AI products, such as ChatGPT, Claude and Gemini, have promised to help.
But are they any good at finding Black Friday deals? We had a go ourselves a week ahead of the big day.
Figures suggest relatively few UK shoppers currently go to AI tools for help.
A recent YouGov survey of 19 countries found the UK ranked second-last in the proportion of people who said they had used AI to help them shop.
Crucially, 36% of respondents who said they would be shopping around Black Friday didn't plan to use an AI tool at all.
But the survey suggested a growing number of people were considering it - and tech firms have released features to capitalise on this.
Engineers at MIT have solved a persistent failure mode in predictive artificial intelligence where small simulation errors compound over time into physically impossible values. By embedding mathematical guardrails directly into nonlinear algorithms, the framework guarantees outputs remain strictly bounded even during extended simulations. This breakthrough provides critical operational stability for complex environments like weather forecasting, aeronautics, and ocean current modeling.
Organizations deploying distributed artificial intelligence models can now eliminate idle compute expenses caused by localized request queuing. The managed routing system optimizes expensive GPU infrastructure by automatically shifting high-volume inference traffic to healthy, underutilized clusters across geographical boundaries. This layer removes the overhead of building bespoke cross-region networking tools while maintaining native enterprise access controls.
Foundation model infrastructure teams can now use a declarative approach via an API or the AWS console to manage settings and software across active nodes. This helps eliminate software drift that traditionally occurs when custom lifecycle scripts fall out of sync with evolving tooling requirements. Administrators still retain deep system control, preserving root SSH access and the option to run direct node customizations.