Executive summary
AI has rewritten the rules of online shopping. Nearly half of consumers can no longer tell the difference between human and AI-generated product recommendations—a watershed moment that signals algorithmic guidance has gone mainstream. Companies are betting big on this shift, pouring 26.8% of their total marketing budgets into personalisation as they chase an estimated $2 trillion in economic value.
- 45% of consumers no longer differentiate between human and AI product recommendations, signalling mainstream acceptance of algorithmic guidance.
- 64% of shoppers now use generative AI tools daily, up from just 29% in 2023, creating familiarity that extends naturally into shopping contexts.
- Younger consumers lead adoption: 45% of Gen Z and 41% of millennials use GenAI for shopping, compared to 24% of consumers overall.
- Major retailers deploy sophisticated AI assistants – Walmart’s Sparky, Amazon’s personalised descriptions – that anticipate needs and streamline decisions.
- Cambridge researchers predict the emergence of an “intention economy” where AI systems anticipate and steer motivations before conscious decisions form.
- Only 14% of marketing leaders fully utilise their personalisation platforms despite massive investment, revealing a significant capability gap.
The question isn’t whether AI will shape consumer behaviour – it already does. The real challenge lies in ensuring these systems enhance choice rather than exploit vulnerability. As algorithmic recommendations become indistinguishable from human advice, understanding the mechanics, motivations, and potential manipulations becomes essential for anyone who shops online.
The way we make purchasing decisions has changed dramatically over the last few years. Whereas we once relied on shop assistants, magazine reviews, or recommendations from friends, we now increasingly turn to algorithmic systems that analyse our behaviour, predict our preferences, and surface products we’re likely to want. These systems operate quietly in the background of nearly every digital shopping experience, shaping what we see, how we evaluate options, and ultimately what we buy. And when it works, it can feel like a sort of dark magic.
For all its promise, however, this shift raises some big questions about consumer autonomy and choice. When an algorithm knows your preferences better than you do – when it can anticipate your needs before you fully articulate them – does that constitute helpful guidance or subtle manipulation? The answer matters deeply; not just for individual shoppers navigating their next purchase, but for the broader relationship between technology companies, consumers, and the nature of free choice in digital marketplaces.
Consumers embrace AI
From scepticism to trust – how shoppers have learned to rely on algorithmic recommendations.
Consumer wariness about AI recommendations – once a major sticking point – is quickly evaporating. The 2025 State of E-Commerce report from Constructor and Shopify surveyed more than 1,500 consumers across the US, UK, and Germany. What they found is striking: 45% of consumers no longer distinguish between human and AI-generated product recommendations. Nearly half of online shoppers have reached a point where the source of advice matters less than whether it actually works. When AI recommendations prove helpful, people simply accept them.
This acceptance springs from something simple: familiarity. Today, 64% of consumers use generative AI tools like ChatGPT daily, compared to just 29% in 2023. That familiarity bleeds naturally into shopping behaviour: if you’ve already spent months asking ChatGPT to explain complex topics or help with work tasks, turning to an AI shopping assistant feels like a logical next step. Research from the Capgemini Research Institute shows generative AI is increasingly influencing consumer behaviour. 68% of consumers – more than two-thirds – are already prepared to act on its recommendations, while 58% actively prefer product recommendations from generative AI tools over traditional search engines. Around one-quarter of consumers used generative AI in shopping experiences in 2024, compared with 20% the previous year.
Leave the change to Gen Z
Nearly half of consumers say they’re enthusiastic about how generative AI affects their online shopping, with younger people leading the charge as usual. Gen Z and millennials want automated, personalised, real-time generative AI customer support at rates of 51% and 45% respectively, compared to 36% overall. The appetite for ChatGPT-style chatbots also runs high across these age groups, with 70% indicating they want them. Overall, 57% of consumers want hyperpersonalised content and product recommendations, with 64% of Gen Z and 62% of millennials comfortable letting generative AI ask personal questions to deliver them, and 55% overall accepting this as part of shopping.
But consumers aren’t naive. A 2024 BCG survey of 23,000 people globally reveals that while roughly four-fifths feel comfortable with personalised experiences, and most expect companies to deliver them, there’s a catch. The three benefits respondents cited most frequently are value, enjoyment, and convenience. However, two-thirds of customers say they recently had at least one personalised experience that felt inaccurate or invasive. In many instances, these missteps caused customers to unsubscribe, disengage, or simply not come back.
“In the year ahead, successful leaders will be the ones who embrace generative AI to improve customer experiences and accelerate business growth.”
Sharyn Leaver, Chief Research Officer at Forrester
The corporate rush toward personalisation
Companies invest billions into algorithmic recommendations, yet many struggle to make the technology work.
The business world has seen the stakes already. BCG estimates that companies excelling at tailored recommendations could capture an additional US$2tr in economic value from personalisation alone. That number alone explains why marketing budgets have shifted so dramatically. Gartner reports that organisations now allocate 26.8% of total marketing budget to personalisation, marking a 30% year-over-year increase. Despite this massive financial commitment, there’s a glaring capability gap: only 14% of marketing tech leaders say they’re actually fully utilising their personalisation platforms. Companies are throwing money at personalisation tools, while most have yet to actually unlock their potential. The disconnect between investment and execution reveals an industry still figuring out how to use technologies it already owns.
Still, executive confidence remains sky-high in spite of these struggles. Twilio’s 2024 survey of marketing leaders found that 89% believe personalisation will be crucial to business success in the next three years. “Generative AI’s influence is dominating all aspects of business and consumers’ lives,” says Sharyn Leaver, Chief Research Officer at Forrester. “In the year ahead, successful leaders will be the ones who embrace generative AI to improve customer experiences and accelerate business growth.” Over 70% of leaders agree that AI adoption will fundamentally change personalisation strategies, while 61% worry that inaccurate data will hamper their efforts. The business elite expect AI-driven recommendations to keep evolving rapidly, but they’re also aware of the risks that could derail results.
How major retailers deploy AI shopping assistants
Walmart, Amazon, and Instacart reveal what algorithmic personalisation looks like in practice.
To understand what algorithmic personalisation actually looks like in practice, we need to look at what major retailers are deploying right now. Walmart’s approach centres on Sparky, a generative AI-powered shopping assistant accessible via the ‘Ask Sparky’ button in the Walmart app. Sparky helps customers search for items, synthesise reviews, and get insights for any occasion, from looking up current sporting events and finding the right jersey to planning celebrations and picking out toys. But Sparky does more than simple product search. You can also ask what sports teams are playing tonight or check the weather at the beach you’re heading to and get outfit recommendations accordingly.
Sparky’s future roadmap gets even more ambitious. Soon, the bot will let customers customise their experience – automatically reordering household essentials, booking services that simplify complex shopping tasks. It will become multi-modal, understanding text, images, audio, and video, weaving seamlessly into customers’ lives to unlock instant access to products and services whenever and however they shop. The goal: solving everyday problems and freeing up time.
Personalised product recommendations
Amazon takes a different approach, building on decades of machine learning infrastructure. For years, Amazon has leveraged AI to provide personalised product recommendations across the homepage and shopping journey, personalised deals during events, and personalised emails suggesting products. Now, with generative AI, Amazon is pushing personalisation further, creating recommendation types and product descriptions tailored to individual shopping activity. Instead of generic recommendations like ‘More like this’, you see specific suggestions based on your shopping history. For example, if you regularly search for gluten-free products and search for ‘gluten-free cereal’, AI intelligently positions ‘gluten-free’ prominently within product descriptions, even if Amazon or the selling partner buried it at the very end. This makes it faster to locate products based on attributes that matter most to you.
Amazon analyses product attributes alongside customer shopping information – preferences, search, browsing, purchase history – using a large language model to edit product titles, highlighting features most relevant to each customer and their current shopping activity. Then another LLM, functioning as an evaluator, challenges and improves these results. “If the primary LLM generates a product description that is too generic or fails to highlight key features unique to a specific customer, the evaluator LLM will flag the issue,” explains Mihir Bhanot, Director of Personalisation at Amazon. “This feedback loop allows the system to continuously refine suggestions, ensuring that customers see the most accurate and informative product descriptions possible.”
Welcome to the smart shop
Instacart’s Smart Shop technology shows how AI personalisation can also extend into health and nutrition. The platform uses generative AI and machine learning to analyse customer habits and dietary preferences, surfacing the most relevant products faster. Alongside Smart Shop, Instacart unveiled AI-powered Health Tags – detailed, transparent nutritional information across their catalogue – and Inspiration Pages featuring expert-backed health recommendations and shoppable recipes. The first Inspiration Page, developed in partnership with the American Diabetes Association, contains evidence-based nutrition guidance and diabetes-friendly grocery and recipe recommendations.
“At Instacart, we want to turn the ordinary task of grocery shopping into a delightful, personalised shopping experience that takes the mental load out of finding the exact items that meet your preferences,” says Daniel Danker, Instacart’s Chief Product Officer. “By combining our new Smart Shop technology, Health Tags, and Inspiration Pages, we’re not just improving online grocery shopping – we’re reimagining it, making it seamless to go from intention to action. By customising your shopping journey to match your personal health goals or fit your dietary restrictions, we can unlock possibilities that weren’t even on the table before.”
“The new AI tools are also making companies better at understanding customers’ vulnerabilities, which may enable them to manipulate user behaviour in ways beneficial for the company but not for customers.”
Dr Ali Makhdoumi, professor of marketing at Duke
When helpful recommendations turn into manipulation
Critics warn that algorithms designed to assist shoppers could exploit psychological vulnerabilities instead.
Unfortunately, the enthusiasm around AI recommendations tells only part of the story. Critics warn that opaque algorithms can subtly manipulate preferences, create economic ‘rents’ for platform owners, or trap users in narrow ‘filter bubbles’ of content. These aren’t abstract concerns, they represent fundamental questions about agency, autonomy, and the power dynamics embedded in algorithmic recommendation systems. Researchers at Cambridge’s Leverhulme Centre for the Future of Intelligence have coined a term for what they believe is emerging: the intention economy. Think of it as the successor to the attention economy, where social networks kept users hooked on platforms and served them adverts. Instead, the intention economy involves AI-savvy tech companies selling what they know about your motivations – plans for a hotel stay, opinions on a political candidate – to the highest bidder.
“For decades, attention has been the currency of the internet. Sharing your attention with social media platforms such as Facebook and Instagram drove the online economy,” explains Dr Jonnie Penn, a historian of technology at the centre. “Unless regulated, the intention economy will treat your motivations as the new currency. It will be a gold rush for those who target, steer and sell human intentions. We should start to consider the likely impact such a marketplace would have on human aspirations, including free and fair elections, a free press and fair market competition, before we become victims of its unintended consequences.”
A loss of agency
The aforementioned Cambridge study claims that large language models will be used to “anticipate and steer” users based on “intentional, behavioural, and psychological data”. An LLM could, at low cost, leverage your cadence, politics, vocabulary, age, gender, preferences for sycophancy, combining these with brokered bids to maximise the likelihood of achieving a given aim – selling a film ticket, for example. In such a world, an AI model steers conversations in the service of advertisers, businesses and other third parties. Advertisers will use generative AI tools to create bespoke online ads. AI models will tweak their outputs in response to “streams of incoming user-generated data”. Models can already infer personal information through everyday exchanges and even steer conversations to extract more personal information.
Dr Ali Makhdoumi, professor of marketing at Duke, delivers a blunt assessment: “The new AI tools are also making companies better at understanding customers’ vulnerabilities, which may enable them to manipulate user behaviour in ways beneficial for the company but not for customers.” He also warns that advanced AI personalisation could exploit human biases, noted that “if the platform can fool consumers for a long time, it will engage in behavioural manipulation”. Thus, he advocates safeguards to “limit the way platforms can offer tailored recommendations […] so that the platform can’t take the process to the extreme”.
Learnings
Consumers have crossed the threshold from scepticism to embrace AI-powered shopping experiences that actually deliver value. Nearly half can no longer distinguish between human and algorithmic recommendations, and the majority actively prefer AI tools for product discovery. Major retailers have deployed sophisticated assistants that anticipate needs, synthesise information, and streamline decisions across complex shopping journeys. Companies are racing to capture trillions in economic value, investing heavily even as they struggle to fully operationalise technologies they already own. But acceptance doesn’t equal understanding, and convenience has little bearing on ethics. The same technologies that help us navigate overwhelming product catalogues and discover items matching our preferences also build the infrastructure for manipulation.
When algorithms learn to anticipate intentions before conscious thought forms; when motivations become commodities auctioned in real time; when AI systems optimise for platform profit rather than consumer welfare, the line between enhancement and exploitation blurs dangerously. The question, then, isn’t whether AI will shape consumer behaviour – it already does, deeply and pervasively. Rather, it’s whether we’ll build guardrails, demand transparency, and insist that personalisation serves genuine needs rather than engineered wants. Platforms can engage in behavioural manipulation when they manage to fool consumers for extended periods. The challenge ahead is ensuring that the systems guiding our choices remain tools we control rather than forces controlling us.
