Starbucks’ new beta app in ChatGPT looks simple on the surface: tell the AI what you’re craving, describe your mood, or upload a photo, and it recommends a drink. Want something bright to start your morning? Craving an afternoon boost that isn’t too sweet? ChatGPT can now translate that moment into a Starbucks order, then help the customer customise it, choose a store and finish checkout through Starbucks’ app or website.
But the bigger story is not that Starbucks found a clever new way to sell coffee. The bigger story is that product discovery is moving from search bars and menus into conversations.
For marketers, that shift matters. A lot.For years, brands optimised for clicks. They built websites, bought media, refined SEO strategies and tried to get in front of customers when they searched for a product. But generative AI is changing the playing field. Search is no longer just a list of links. In many cases, it is becoming a single synthesised answer, a recommendation, a conversation and, increasingly, a commerce experience.
If your brand is not part of that answer, you may be invisible at the exact moment customers are deciding what to buy.
This is your last chance to book tickets to the first-ever Campaign Breakfast Briefing: Retail & Commerce Media on Friday, 22 May. Get yours now for actionable strategy on enhancing ROI for retail and commerce media in the region.
From search intent to mood intent
Traditional product discovery was built around keywords. A customer searched for iced coffee near me, low sugar drink, or best afternoon coffee order. Brands optimised content, menus and ads around those phrases.
AI discovery does not work that way.
A customer may not know the product name. They may not even know the category. They may start with a feeling: I want something refreshing but not too sweet. They may start with a moment: ‘I need a drink for a long workday.’ They may start with a photo, an outfit, the weather or a vibe.
That is exactly why the Starbucks integration is important. Starbucks is not asking customers to navigate a menu first. It is letting the customer describe the need, then letting AI translate that signal into a product recommendation. That is a very different kind of intent.
We have been saying this for a while: the best brands do not just buy media; they build smarter audiences. We are not in the ad business. We are in the audience intelligence business. AI does not just help brands find customers. It helps brands find the ones that matter most. The Starbucks example shows what that looks like in real life.
Predictive personalisation is now mainstream
Personalisation is no longer about adding someone’s first name to an email or retargeting them because they visited a product page.
The best brands are predicting customer behaviour in real time: who is likely to buy, who is at risk of churning, who is browsing but has not clicked, who is responding to a creator, who is showing intent before they ever land on the brand’s website.
AI models are running that maths behind the scenes and making sure each customer gets the right message, at the right moment, on the right channel.
Starbucks is applying that same principle to discovery. Instead of waiting for a customer to know the exact drink, the brand is meeting the customer earlier in the journey, at the moment of curiosity.
That is where product discovery is headed. The next wave of commerce will not be driven only by people who searched for X. It will be driven by people who show patterns, signals, preferences and behaviours that indicate what they are likely to want next.
It is not just first-party data anymore
For years, marketers were told first-party data was the holy grail. CRM lists, loyalty data, purchase history and app activity still matter. But they are no longer enough on their own.
Brands are now layering in behavioural signals – app usage, store visits, content engagement, purchase patterns, creator interactions and device-level behaviour – to train lookalike and predictive models.
That is the real opportunity behind AI-powered discovery.
A Starbucks customer who asks ChatGPT for a not too sweet afternoon boost is not just placing an order. They are creating a signal. That signal says something about taste, timing, occasion, energy level, dietary preference and potential future behaviour.
For a marketer, that is powerful. The question becomes: how do you turn those moments into audience intelligence without crossing the creepy line?
That is where deterministic, consent-based data matters. Facial recognition and overly invasive personalisation may get attention, but they do not scale in a way that builds trust. Brands need signals that are useful, privacy-conscious and actionable.
The goal is not to make personalisation feel invasive. The goal is to make discovery feel effortless.
AI search changes the funnel
The old funnel assumed that customers started broad and moved slowly: awareness, consideration, conversion. AI compresses that journey.
A customer can ask a detailed question, get a recommendation, compare options, customise a product and move toward purchase in the same conversation. That means AI assistants are not just top-of-funnel discovery tools. They are increasingly bottom-of-funnel decision engines. That changes how brands need to think about visibility.
In traditional SEO, the question was: How do we rank No. 1 for this keyword?
In AI search, the question is: How do we become one of the brands the AI mentions?
That requires a different strategy. Brands need content that is structured, specific and easy for AI systems to understand. They need to anticipate follow-up questions. They need to show up across the sources AI trusts, not just on their own website. And they need to understand which audience segments are most likely to use AI-powered discovery in the first place.
If customers are asking ChatGPT what to buy, where to go, what to try and which product fits their needs, brands need to know how to be part of that answer.
Portable audiences will matter more than platform audiences
One of the biggest mistakes marketers can make right now is assuming AI discovery will live inside one platform. It will not.
Customers are already moving across ChatGPT, Google AI Overviews, Perplexity, TikTok, Instagram, Reddit, CTV, retail media networks and brand-owned apps. Starbucks’ ChatGPT integration is one example of a larger shift: discovery is becoming more conversational, more fragmented and more signal-rich.
That is why portable lookalike modelling matters.
Facebook look-alikes are not enough. They are stuck inside Meta. Smart brands are building portable lookalike audiences that can be activated across TikTok, CTV, programmatic, direct mail and owned channels. You own your data, you control your segments and you scale on your terms.
As AI discovery expands, marketers will need to understand not only who their best customers are, but where those customers are most likely to discover, compare and buy. That is the difference between buying reach and building intelligence.
The new personalisation playbook
The Starbucks integration also points to a broader change in creative strategy.
Micro-personalisation at scale is becoming the standard. Brands are no longer limited to broad demographic buckets. They can build and activate segments like Plant-Based Snackers, App-First Shoppers, High-Intent Travellers or Creator-Influenced Buyers, then tailor creative and channel mix accordingly.
For a CPG brand, that could mean serving one version of an ad to health-conscious moms on Hulu and another to Gen Z shoppers on TikTok – even if both ads promote the same product.
For Starbucks, it could mean understanding which customers respond to seasonal flavours, which are driven by social trends, which want lower-sugar options, and which are most likely to try a new drink because ChatGPT recommended it.
That is where AI personalisation gets interesting. Not because it replaces the human experience, but because it helps brands understand the human context behind a purchase.
Product discovery is becoming audience discovery
The most important takeaway from Starbucks’ ChatGPT launch is not that consumers want AI to pick their coffee.
It is that consumers are becoming more comfortable letting AI help them narrow choices, interpret preferences and make decisions. That means marketers need to rethink product discovery as audience discovery.
Who is asking AI for recommendations? What are they asking? What language are they using? What signals suggest they are ready to buy? Which audiences are being influenced by creators, social content, app experiences or AI-generated answers? Which customers are likely to respond to a personalised recommendation versus a traditional ad?
Those are the questions that will define the next phase of marketing. The best brands will not just chase impressions. They will build signal-rich, intent-based audiences. They will use AI to understand not just what customers bought, but what they were trying to solve, feel or experience when they bought it.
Starbucks is showing where discovery is going.
The brands that win will be the ones that stop thinking about AI as a novelty and start treating it as a new layer of the customer journey.
Because in the next generation of commerce, the customer may not start with a search.
They may start with a sentence. And the brands that understand that sentence – the mood, the signal, the intent behind it – will be the ones that get recommended.
By Mike Ford, CEO of Skydeo.
