Ask ten people what an "AI shopping assistant" is and you will get ten answers. A chat window on a retailer's website. A general-purpose chatbot that recommends a jacket it has never seen. A browser tool that quietly learns what you like. The phrase has become an umbrella for anything that puts software between you and a purchase, and the vagueness is starting to matter, because the differences between these tools are larger than the similarities.
They differ in what they can do, in who they ultimately work for, and in what happens to the information you hand them. This guide separates the three things people actually mean, and gives you the questions that cut through the marketing.
The moment we're in
The money arrived before the vocabulary did. Amazon launched Rufus, a generative-AI assistant built into its shopping app, in early 2024; by May 2026 it had folded Rufus into a broader Alexa shopping agent, a sign of how quickly the category is being rebuilt even inside the company that dominates it. OpenAI switched on Instant Checkout in ChatGPT in September 2025, letting American users buy from Etsy sellers without leaving the chat; Etsy's shares jumped sixteen per cent on the news, and OpenAI confirmed it takes a fee on completed purchases. Perplexity had already launched a shopping assistant of its own in late 2024, with one-click checkout through merchant partners.
The startups are moving just as fast. Daydream, founded by former Stitch Fix executive Julie Bornstein, raised a fifty-million-dollar seed round to build an AI search engine for fashion. Phia, founded by Phoebe Gates and Sophia Kianni, raised thirty-five million dollars to compare prices across retail and second-hand sites.
Shoppers are following. Adobe measured a 1,200 per cent jump in traffic arriving at US retail sites from generative-AI sources between July 2024 and February 2025, and by the 2025 holiday season that traffic was up a further 693 per cent year on year, converting to purchases at notably higher rates than other channels. People are no longer experimenting with AI shopping. They are shopping with it.
The three things people actually mean
Retailer chatbots are the assistants shops build into their own sites and apps; Rufus was the most prominent example. They can be genuinely useful for "does this come in a 42?" questions, and they know their own catalogue deeply. But they know only one shop, and their job, in the end, is to help that shop sell to you. That is not sinister. It is simply worth knowing which side of the counter the tool sits on.
General AI chatbots — ChatGPT, Perplexity and their peers — sit further from any single retailer, and the arrival of built-in checkout makes them feel like genuine assistants. They are fluent, broad and impressively quick at research. But they know the internet in general rather than you in particular, and as checkout fees and merchant partnerships accumulate, the incentive question returns in a new form: a fee on every completed purchase is still a stake in what you buy.
Personal shopping intelligences are the newest and least familiar category: tools that sit on your side of the counter, learn your taste over time, and work across shops rather than for one of them. Daydream and Phia are building in this direction; FRIDAY belongs to this category too. The promise is different in kind. Not "let me sell you something", but "let me remember what you actually like".
Search resets every session. A personal intelligence compounds every session.
The three questions that cut through
Whoever builds the tool, three questions do most of the work.
Who does it work for? Follow the money. If the assistant earns a commission or a checkout fee when you buy, it has an interest in you buying, and sometimes in where you buy. That does not make it useless; it makes it a salesperson, and you should weigh its enthusiasm the way you would weigh a salesperson's.
What does it remember? A tool that forgets you every session can never really know your taste. A tool that remembers you raises the next question: can you see that memory, correct it, and delete it? If the answer is buried or absent, treat that as an answer in itself.
Where does your data live? On your device, or on a server being monetised in ways you cannot inspect? UK law gives you real rights over the personal data an organisation holds about you, including access and erasure. But architecture is stronger than policy: data that never leaves your device does not need a promise attached.
Free tools deserve the sharpest version of these questions, because the money has to come from somewhere. Sometimes the answer is honest and harmless. Sometimes the product is you. We have written separately about how free shopping extensions make their money; it is worth five minutes before you install anything.
Where FRIDAY sits
FRIDAY is personal intelligence for shopping. It learns your taste on your device, and helps you shop across every store on the open web — no ads, no noise. Every purchase, every skip, remembered. Your taste lives in an encrypted vault sealed with a key only you hold, and you can pause it any time.
Put against the three questions, that is the shape of the honest answer: it works for you, its memory is yours, and no ads, no data sale — the architecture makes the promise for us. Ask like you'd ask a friend, and what comes back is the shortlist, not the scroll.
The category will keep shifting; this year's assistants are already being folded into next year's agents, as Amazon has shown. The three questions will not shift. Ask them of every tool that offers to shop with you — ours included.
