The short version
- In EY’s 2026 Consumer Beverage Survey (opens in a new tab) of 2,512 adults, 27% of US consumers had used AI-based beverage recommendations in the past year, and 45% of Brazilian consumers had.
- Shoppers choose drinks by what is in them: 58% of US consumers pay attention to ingredients, 66% choose lower-sugar, lower-calorie drinks and 52% will pay more for drinks that support their health goals (EY).
- The prize is large and shifting toward function: shoppers spent nearly $295 billion on US retail beverages in 2025, according to FMI’s Power of Beverage 2026 (opens in a new tab) report built on Circana data, and energy drinks grew 15%.
- A won shopper is a habit, not a single can: NIQ found functional-beverage spend per buyer rose 12.0% in a year, against 5.4% for all beverages, with millennials spending $153 a year each.
- For wine, beer and spirits, a 2025 survey commissioned by the alcohol platform DRINKS (opens in a new tab) found 31% of US adults over 21 had already used AI to help choose alcohol.
Who buys beverages now, and what is one new drinker worth?
One shopper buys on need, taste and price, and a new drinker’s value lies in a year of repeat purchases.
Beverages remain one of food retail’s strongest categories. FMI’s report, written with Circana sales data and a May 2026 survey of 2,003 US grocery shoppers, puts 2025 retail beverage spending at nearly $295 billion, up 3% on 2024. Food Business News’ summary (opens in a new tab) of the same report says non-alcoholic beverages grew 5% in dollar sales, led by energy drinks.
What shoppers say defines value is telling for anyone hoping to be recommended. Sixty-one percent say low price defines good value in a beverage, followed by taste (46%), trusted quality (40%) and a brand they trust (39%).
The purchase is cheap and frequent, so a single sale is small. The value sits in repetition. NIQ’s June 2026 analysis of functional beverages, drinks sold for energy, hydration, gut health, mood or focus, says “Habit is replacing trial.” It found a functional-beverage buyer spent $119 a year on average, millennials $153 and Gen X $136. Millennials drive 40% of the category’s dollars. For a brand, the commercial question is how many new people try the drink once, and how many of them make it part of a routine.
Big companies pay heavily for brands that have built that routine. In May 2025 PepsiCo closed its acquisition of poppi (opens in a new tab), the prebiotic soda, for $1.95 billion. PepsiCo credited poppi’s community- and culture-first approach, including viral TikTok campaigns and influencer partnerships. Our reading: discovery, not distribution alone, built that brand.
Where do AI assistants already sit in a drink shopper’s decisions?
Early, at the moment a shopper names a need or an occasion and asks what to buy.
The EY survey, fielded online in late 2025 and weighted to the US and Brazilian census, gives the clearest picture so far:
- 27% of US consumers and 45% of Brazilian consumers used AI-based beverage recommendations in the past year, and 70% of Brazilians said they are very likely to use them next year.
- US consumers also find functional drinks through online grocery recommendations (19%), fitness and health apps (17%) and loyalty apps (16%).
- 80% of Gen Z and 75% of millennials drink functional beverages at least every two weeks, against 65% overall.
Alcohol shoppers are moving in the same direction. In the DRINKS survey of 1,000 US consumers, run by Dynata in 2025, 71% said they would be interested in AI help if liquor stores and online retailers offered it, 44% trusted AI to recommend a bottle of wine or liquor, and 39% would let it pick drinks for a party. Only 30% called human expert advice “very important” to alcohol purchases. DRINKS sells AI tools to alcohol retailers, so it has an interest in that result.
The assistants have also tried to become places to buy. In December 2025 Instacart launched an app inside ChatGPT (opens in a new tab) that built a grocery cart from a conversation and let the user pay through OpenAI’s Instant Checkout without leaving ChatGPT. That launch was documented by both companies. We follow that path from meal plan to cart in our food ecommerce guide. OpenAI scaled the feature back in March 2026, according to FashionUnited (opens in a new tab), though its help page still describes it for some eligible merchants. Across retail generally, Adobe’s data reported by TechCrunch (opens in a new tab) showed AI traffic to US retail sites up 393% year over year in the first quarter of 2026. Adobe did not break out beverages.
What do drink shoppers ask AI assistants?
They ask about a need, an occasion or an ingredient, and rarely about a brand by name at first.
The prompts below are illustrative, written by us to show the shape of beverage questions. They are not captured from real users.
| Stage | Illustrative prompt |
|---|---|
| Need state | “Best electrolyte drink with no added sugar for long runs” |
| Alternative | “What’s a healthier soda that still tastes like soda?” |
| Comparison | “Olipop or poppi: which has less sugar?” |
| Ingredient check | “Energy drinks without sucralose or artificial colors” |
| Occasion | “Non-alcoholic drinks for a baby shower that feel festive” |
| Pairing and budget | “A red wine under $25 that goes with grilled salmon” |
| Where to buy | “Where can I buy this sparkling tea near me?” |
These questions match how the category is now sold. FMI says shoppers increasingly buy on functional benefits and specific occasions, and it urges retailers to organize the aisle around need states. NIQ’s advice to brands is to focus “less on claim variety and more on use occasions tied to consumer need states.” Our inference: an assistant answering a need-state question is doing exactly that sorting, out loud, for one shopper. Sportswear brands meet the same activity-based questions, as what shoppers ask AI about training gear shows.
Each question can also turn into several searches. Google says AI Overviews and AI Mode may use a “query fan-out” technique (opens in a new tab), and that AI Mode draws on shopping data for billions of products (opens in a new tab). Both are documented by Google. In our hidden-searches study, ChatGPT looked for reviews in 46.2% of its answers to buyer questions.
How does a mention in an AI answer become sales?
Through a trial purchase at a store, then repeat purchases if the drink earns a place in a routine.
The path for a beverage differs from a software purchase or a booked appointment:
- Named for a need. The assistant puts a few drinks on a short list for “energy without a crash” or “alcohol-free for dinner.”
- Checked on the facts. The shopper compares sugar, caffeine, ingredients and price, the attributes EY and FMI say drive the choice.
- Bought where it is convenient. Most drinks are bought at grocery, convenience, club or online retailers; from late 2025 that list also included a cart built inside ChatGPT through Instacart. Some brands also sell bundles on their own sites. Turning those into repeat orders is covered in winning subscribers through AI search.
- Repeated. If the drink works for the shopper, the purchase becomes a weekly habit, which is where NIQ’s $119 to $153 a year per buyer comes from.
The first purchase may not even happen on your website, so the result shows up as retail velocity, the rate at which a product sells at each store, rather than as a web visit. When shoppers do click through, Adobe found AI visitors to US retail sites converted 42% better than other visitors in March 2026.
For wine, the direct channel is under pressure, which raises the value of every recommendation. Sovos ShipCompliant’s 2026 report (opens in a new tab) found winery shipments to consumers fell 15% in volume and 6% in value in 2025, a loss of 967,000 cases. Napa wineries held up best, adding 1% to the value of their shipments. Our inference: when fewer people find wineries by visiting them, a recommendation for a specific bottle carries more weight.
Why does an assistant name one drink and not another?
Mostly because of product facts, reviews and third-party coverage it can find; the exact rules are not public.
What is documented: OpenAI says (opens in a new tab) ChatGPT’s product results are not ads and are selected by ChatGPT based on the query and context. It names structured metadata from first-party and third-party providers, such as price and product description, plus other third-party content and review summaries drawn from public websites. Merchants on Shopify are already connected through Shopify Catalog, and others can apply to send a direct product feed.
What has been observed in studies (none of them on beverages):
- Facts beat fame, until products look the same. In tests of three AI systems on skincare products, Chu and Hou (opens in a new tab) found rating, price and reviews explained 82.4% of how products were ranked and brand name only 1.2%. When every product had identical specs, the famous brand won all 670 valid trials. A challenger drink needs a clear, checkable difference.
- Each assistant reads different sources. For the same consumer shopping questions, Uberti-Bona Marin and colleagues (opens in a new tab) found ChatGPT and Gemini shared only 5.4% of the websites they displayed, and editorial and review sites made up 56.7% of ChatGPT’s sources.
- Lists change between asks. In our consistency study, only 25.2% of the brands ChatGPT named appeared in all five repeats of the same question.
- Independent mentions matter. In our brand entity study, each tenfold increase in the number of independent websites naming a brand went with 4.7 times the odds of being recommended.
Our inference for beverages: the trust factors are accurate nutrition and ingredient facts, retailer ratings and review volume, taste tests and “best of” lists from known publications, creator reviews, and wide retail availability. FMI found 45% of Gen Z say a drink becomes more appealing when they learn about it from a creator, so creator coverage may matter twice, to people and to the assistants that read about them. We explain the research on product facts in what drives AI product recommendations.
What does a drink brand lose when it is missing from AI answers?
It loses trial at the moment a shopper picks a new drink, where the category’s growth now comes from.
Direct evidence of lost sales does not exist yet, so we label the reasoning:
- The growth is in need states. NIQ found energy claims added $2.7 billion in dollar growth, and FMI reported double-digit sales growth for drinks offering mood support, immune health and digestive health. These are exactly the questions shoppers bring to assistants.
- Habits lock in. NIQ’s point that habit is replacing trial cuts both ways. We infer that a shopper who settles on a rival’s drink after an AI answer may not try another for months.
- Some product pages cannot be read. Adobe found about 34% of retail product pages could not be properly accessed by AI. A drink whose details sit only in images or behind age gates may simply be skipped.
Treat any precise estimate of lost revenue from AI absence as a guess. No beverage company has published one.
How does GEO work for a beverage brand?
Generative engine optimization (GEO) makes your drinks easy for AI assistants to find, describe accurately and support with outside evidence.
For a beverage company, that work usually covers:
- Clean product facts everywhere. One consistent name per product and flavor, with sugar, calories, caffeine, key ingredients and pack sizes stated in text on your site, your retailer listings and your product feeds. Adobe’s finding on unreadable product pages makes this the first fix. If assistants already get your facts wrong, see how to fix wrong brand information in AI answers.
- Pages for need states and occasions. Plain pages that answer “drink for afternoon focus,” “alcohol-free options for dinner parties” or “what to pair with this wine,” with honest comparisons to the alternatives people consider.
- Third-party coverage. Taste tests, roundups in food and drink publications, creator reviews on YouTube and TikTok, and genuine community discussion. For small brands, this is the main lever; see how a small brand gets recommended by AI. Gift guides play a similar part for toy brands answering gift questions.
- Retail reputation. Ratings and reviews on Amazon, Walmart, Target and grocery sites, plus clear where-to-buy information, because the purchase usually happens there.
- Product feeds and catalogs. Keep Shopify Catalog, merchant feeds and any OpenAI product feed complete and current, since OpenAI documents that it reads that metadata.
- Responsible claims. State functional benefits only as your evidence and labeling allow. In tests, manipulative product copy was flagged and demoted, and an exaggerated health claim is a legal and reputational risk, not a shortcut.
- Measurement. Ask a fixed set of need-state and occasion questions in ChatGPT, Gemini and Google’s AI features many times, track how often your drinks appear and which sources are cited, and watch retail sales in the accounts those answers point to.
None of this guarantees a recommendation. It raises the odds that when an assistant looks for the best answer, it finds yours, stated accurately and backed by others.
What can’t the evidence tell beverage brands yet?
It shows shoppers asking AI about drinks, not how much any brand sells because of it.
- Self-reported surveys. EY, FMI and DRINKS asked people what they did and would do. Behavior can differ, and the DRINKS survey was commissioned by a company that sells AI tools to alcohol retailers.
- No beverage-specific studies of AI picks. The ranking research we cite tested skincare and general consumer products such as electronics. We infer it applies to drinks because the purchase is similarly attribute-driven, but no one has tested beverages directly.
- Store sales are hard to trace. A recommendation that leads to a purchase in a supermarket aisle leaves no click. We found no beverage company that publishes sales attributed to AI answers. The broader question is weighed in does AI visibility drive business results.
- Alcohol is a special case. Age rules and state-by-state shipping laws shape where alcohol can be bought, and we found no public data on how assistants handle alcohol shopping across markets.
Where should a beverage brand start?
Start by asking the need-state and occasion questions your future drinkers ask, and see whose drinks are named.
That first check usually shows three things: which of your products appear and for which needs, which rival drinks and sources appear instead, and whether your sugar, caffeine, ingredient and price facts are quoted correctly. From there, the work is to fix the facts, earn the coverage and make your retail listings ready for the shoppers an assistant sends.
When growth depends on trial and repeat purchase, send us a note and we will show where your drinks appear in AI answers, why rivals are named instead, and which changes are most likely to put your products on more shoppers’ short lists. What that work covers, from clean nutrition facts and need-state pages to retail reviews and product feeds, is set out on our generative engine optimization service page.
Frequently asked questions
Do people really ask ChatGPT which drink to buy?
Many do. In EY’s 2026 survey, 27% of US consumers had used AI-based beverage recommendations in the past year, and in a 2025 DRINKS survey, 31% of US adults over 21 had used AI to help choose alcohol.
Does a famous beverage brand automatically win AI recommendations?
Not automatically. In one controlled study, the famous brand won every test when products were identical, but rating, price and reviews explained 82.4% of rankings once products differed. A clear, checkable difference gives challengers a chance.
Can a drink be bought inside ChatGPT?
Groceries could. From December 2025, Instacart’s app in ChatGPT built a cart from a conversation and offered checkout without leaving ChatGPT, so drinks sold through Instacart’s retail partners could be bought there. OpenAI scaled in-chat checkout back in March 2026, according to FashionUnited, though its help page still describes it for some eligible merchants.
Should functional drink brands make health claims to get recommended?
Only claims your evidence and labels support. Studies of AI shopping systems found manipulative copy was often flagged and demoted, and inflated claims create legal and reputational risk. This article does not give health advice.
Sources
- EY (2026-03-09), EY Consumer Beverage Survey: health-led choices, generational changes and digital discovery are redefining beverage expectations (opens in a new tab)
- Supermarket News (2026-07-31), Function is becoming primary beverage purchase driver: report (opens in a new tab)
- Food Business News, Caleb Wilson (2026-08-13), Functional beverages are having a moment (opens in a new tab)
- NIQ (2026-06), NIQ Perspective: Functional Beverages
- DRINKS via The AI Journal (2025-04-22), Consumer appetite for AI-driven drink recommendations is growing (opens in a new tab)
- PepsiCo (2025-05-19), PepsiCo completes acquisition of poppi, accelerating strategic portfolio transformation (opens in a new tab)
- Supermarket News, Mark Hamstra (2025-12-08), Instacart launches end-to-end shopping app on ChatGPT (opens in a new tab)
- TechCrunch (2026-04-16), AI traffic to US retailers rose 393% in Q1, and it’s boosting their revenue too (opens in a new tab)
- Sovos ShipCompliant (2026), 2026 Direct-to-Consumer Wine Shipping Report reveals record declines as market downturn deepens (opens in a new tab)
- FashionUnited (September 29, 2026), US consumers would let AI agents buy clothes, but not without a say (opens in a new tab)
- OpenAI (2026), Shopping with ChatGPT Search (opens in a new tab)
- Google Search Central (2025), AI features and your website (opens in a new tab)
- Google (2025-03-05), Expanding AI Overviews and introducing AI Mode (opens in a new tab)
- Chu and Hou (2026), Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems (opens in a new tab), arXiv:2606.17443.
- Uberti-Bona Marin and colleagues (2026), "If I Had to Buy Just ONE: Galaxy S26 Ultra": Auditing AI-Generated Product Recommendations (opens in a new tab), arXiv:2609.18729.
- Underneath (2026), The hidden searches AI assistants run before they answer
- Underneath (2026), Ask an AI the same question 5 times: do the brands change?
- Underneath (2026), Do Wikipedia and schema make AI assistants recommend a brand?