Premium drink businesses have long understood the value of a good story. Vineyards have legendary soils. Winemakers have origin stories. Distilleries have canon events. Cellars have secretive rituals to rival the Templars. Some sommeliers even have divine revelation (John 2:1-11, if you don’t believe me).
Each story has evolved to attach itself to a brand in such a way that the two are hard to separate.
The uncomfortable reality in 2026 is that good stories don’t matter to LLMs (Large Language Models).
The way brand discovery happens has changed pretty profoundly. Search has leaped from a list of links to AIs that want to tell their own story, based on an opaque curation, summarisation, and recommendation systems. This move will challenge drinks businesses, and the brand discovery strategies they have relied on for the last several decades.
That’s because being a brand of distinction to an AI bot means something very different than what it means to their temporary human overlords.
Prompting insights
Jordan Brannon of Coalition Technologies sets out how to make your business and the brands and services you sell relevant to AI Search platforms
We people still initiate the whole thing with our prompts.
But prompts are messier than keywords. They are longer (averaging over 20 words), more personal, and more situational. Given the nature of popular AI usage behaviours, they also tend to ask for more creative input. A good ‘ole Google search could have been “premium mezcal supplier UK”.
A Google Gemini prompt could ask the LLM to “create a credible premium mezcal offering with a meaningful sustainability story that works for a high end bar in London”.
That one prompt may also rope in several additional, hidden searches via a technique referred to as query fan out - or QFO. (Not to be mistaken with the more colourful GTFO).
The AI may look for premium mezcal brands, sustainability claims, hotel-bar fit, UK distribution, trade credibility, reviews, press coverage, and competitor comparison before providing a single answer.
Traditional keyword research was built around measures of demand. We could see terms, the search volume, related queries, and measures of competition.
But today’s AIs either don’t have that data, or they don’t share it. ChatGPT’s reporting is all internalised and only shared with select publishing partners. Gemini visibility is equally opaque, and even experiences on Google.com, like AI Overviews, tend to have muddied transparency in Google Search Console or Analytics.
Step one then becomes rebuilding your keyword research efforts for prompts, and initiating what tracking mechanisms you can.
A prompt library
Build a prompt library around real buying situations faced by your customer. Aim for 25 to 50 prompts that reflect how different audiences (whether B2B or D2C) might ask for recommendations. Don’t forget the propensity for “create for me” type prompts.
Group those by category, geography, use case, social proofs, comparisons, and audiences.
It’ll be impossible to track every possible phrase, so consider those that have been most economically vital to your brand. Think about how existing keywords (for which you have substantial data) may evolved into new LLM prompts.
From there, dial in an evaluation on whether prompts are branded, category driven or a decision-making prompt. Many a premium brand is like Narcissus, overly focused on brand visibility. That’ll only tell you if someone who already knows you, can find you. Which isn’t usually the problem.
We recommend tracking three buckets separately: branded prompts; category prompts; and decision prompts.
The growth happens in the category and decision visibility.
You’ll also need to shift your focus from ranking your own URLs, to understanding what citations show up in AI answers. Coalition Tech is doing a lot of testing on AI rank tracking tools, and most allow you to get an idea of mention rate, citation rate, primary recommendation rate, and sentiment.
Multi platforms
Brands and businesses need to be monitoring a number of AI search engines to see how relevant and visible they are to them. Picture istock Gerenme
Finally, as far as prompt tracking goes, remember to cast a wide net.
ChatGPT has market share and recognition, but Gemini is the steady grower. Claude’s shot up in the last few months. Google’s integrated AI experiences, AI Overviews and AI Mode are also worth tracking.
Using more than one rank tracking tool (or working with an agency who does), can be helpful here. Many tend to prioritise one or two key LLMs over reliable measurements across the market broadly.
As you evaluate your “rankings”, don’t forget to monitor sentiment.
LLM companies are responding to earlier criticisms of being overly upbeat by training and reinforcing their models to lean more critical.
Coalition’s recent study in Forbes showed that Grok was the most negative of major LLMs, with ChatGPT being next up. ChatGPT tended to insert some kind of brand or product negative in three out of four conversations.
And as ecommerce interest through LLMs increases, we’re seeing more skepticism from chatbots about promotional language. In one recent test we did for a brand, ChatGPT described their product claims as “gimmicky”, largely based on a five year old criticism found in an archived Reddit thread.
Tracking the prompt will help you know how you rank. But tracking sentiment will help you understand where your brand story isn’t persuasive enough for AIs or AI agents.
Think beyond your website
This is critical tactical step to take.
Many of these prompt responses are formulated from websites beyond branded ones. In a recent Search Engine Land publication, leaning on Peec AI’s analysis, Reddit, YouTube, LinkedIn, Wikipedia, Forbes, Yelp, and G2 were found to be big contributors to AI recommendations.
Drink marketers now need to think more as distributors. Your story being published isn’t enough. It now has to be carried by machine-trusted sources.
Premium drink brands need to look at trade publications, distributor pages, venue partners, buyer interviews, commentary, educational content, YouTube tastings, LinkedIn founder articles, credible reviews, retailer listings, restaurant menus and even community discussion as part of how they market.
This isn’t link building in the old sense. It’s citation building.
Normal links provided a path to your site and some “link juice”. Citations get repurposed by LLMs to provide evidence that your brand belongs in the answers.
The nature of today’s AI answer engines is not to drive traffic first to your site. It’s to occupy and interact with your consumer with an off page click being a near final step.
Brands have to look at how they can utilise their relationships and marketing connections to tell a whole story, from top of funnel to bottom, assuming that their website becomes relevant only near the bottom.
Sameness wins?
Many a marketer loves the opportunity to tell different stories in different channels to highlight different facets of the brands they represent.
But AIs appreciate sameness. It’s in their very foundation.
To understand entities, AI systems need consistency and clear, repeatable connections that help make brand messages the “next right word”.
Entities are concepts like brands, people, or products. The human mind does magical things to make connections that help round out our understanding of any one entity, and computer systems have long been trying to replicate that.
You can help a machine smooth over some of the challenge by ensuring that attributes for a given entity are consistently represented around the web. Ensure that particular production methods are named in the same way across retailers.
Try not to get too creative with adjectives that may have greater importance in your desired prompt outcomes. For example, switching a product description from luxury to craft to sustainable to disruptive could backfire, and the LLM may consider the product as being none of those things.
But so does substance.
And while you’re pursuing a consistent way of representing the entities, don’t forget that AIs want your descriptions to have a bit more substance to them.
Describing something as premium or sustainable doesn’t land well when the LLM wants to reinterpret the content by itself (and especially when the LLM wants to call into question your marketing-speak). Define and reinforce what makes something premium or sustainable in material ways, instead of just making the claim. Speak to the exclusivity in measurable terms. Document the sustainable chops of the brand and its practices.
As you do so, you’ll find ChatGPT and others beginning to creatively describe your various entities, while still leaning heavily onto whatever detail you offered them.
In the previously mentioned brand demonstration that highlighted the dated reddit post, we found that offering detailed write ups of specific processes of the company’s production process (including some independently verified test outcomes), easily replaced the “gimmick” language that ChatGPT relied on to question the product promotion.
If robots dream of sheep, what do they drink?
AI agents are now shopping online to determine which sites are performing best for AI search
For now, most drink discovery involves people. Someone out there is doing the shopping, even if they’re using AI to do it. But agents are already moving out of research experiences and into comparison and transaction.
Harvard Business Review talked about this shift, noting a growing share of shoppers are not human, but AIs researching, comparing, and even purchasing on behalf of consumers. Its article specifically pointed to ChatGPT moving deeper into product discovery and merchant transactions, Google launching its Universal Commerce Protocol and Amazon releasing tools to let agents shop other retailer’s sites.
The early research around agentic shopping should give marketers a bit of pause. Secondary summaries of the HBR piece reported that eight common promotional mechanisms and found that only one behaved consistently the way marketers would expect of human buyers. Tactics like scarcity cues, countdown timers, strike-through pricing, and vouchers failed to interest the agent and sometimes even backfired.
Of course, robot shoppers don’t mean we stop marketing. It means that we have to ensure our brand and marketing are legible to agents as well as desirable to people.
An AI agent may not be seduced by your bottle photography or vague scarcity language. It’s more likely to respond to verified attributes, trusted sources, clear pricing and even (purportedly) human reviews.
Drink brands are not ill positioned to succeed here, but they do need to take it seriously. All the years of product testing and distribution suddenly becomes usable data for AI agents to consume and consider as part of their designated activities.
One final note here. Agents don’t browse the web looking at the UX in the manner people do. They’re looking through the code to the possible presentation. You’ll want to ensure that you understand what technical blockers you may have incidentally sowed in their navigation experience too.
The End?
Premium drink brands that address these aspects of their marketing in the AI era will live to fight another day, and to tell another story. While we don’t fully understand how that story will end, we do know that AIs will increasingly be a big part of it.
* Jordan Brannon is the president and COO of Coalition Technologies. He is a search marketing leader pioneering AI search strategy, focused on helping brands transition from traditional SEO to secure strong visibility and favourable sentiment inside AI-driven discovery environments.
