AI agents are beginning to research, compare, and purchase products on behalf of consumers. What those agents say about your brand depends on what your existing signals say. If your signals are incoherent, the AI won’t correct them — it will amplify the confusion.
What You’ll Learn
- Why AI agents are misrepresenting brands right now, and what it costs
- What “share of model” means and why it’s replacing share of voice
- How to determine whether your brand needs its own AI agent
- Why the fix for AI misrepresentation is the same fix your brand already needed
- What forward-thinking companies are doing to manage their presence in LLMs
Why Are AI Agents Getting Brands Wrong?
AI agents don’t invent perceptions from nothing. They assemble them from the signals a brand has already scattered across the internet: website copy, product descriptions, social media, reviews, press coverage, third-party databases. When those signals disagree with each other, the AI draws the wrong conclusion.
This is what happened to Ballantine’s Scotch whiskey. In 2024, Gokcen Karaca, the head of digital and design at Pernod Ricard, discovered that Meta’s Llama model was categorizing Ballantine’s as a prestige brand. Ballantine’s is the company’s second-best-selling Scotch worldwide, an affordable mass-market product. The confusion made sense: Ballantine’s also has a prestige line, and the brand’s signals across channels didn’t clearly distinguish between the two. The AI picked up on the ambiguity and got it wrong.
“We have a brand called Ballantine’s. It’s the No. 2 Scotch whisky that we sell in the world. So it’s a product for mass audiences,” Karaca told MIT Technology Review. “However, Llama was identifying it as a prestige product.”
Karaca’s team partnered with a digital marketing agency and began systematically prompting all major LLMs, cataloging how each model described their brands, and then updating website and advertising copy to correct the record. Through careful iteration, Ballantine’s is now correctly identified as an affordable Scotch by the major models.
This wasn’t a technology failure. It was a coherence failure that technology made visible. The signals were already contradicting each other. The AI just showed up with a mirror.
As a general rule, if your brand’s own materials can’t agree on what you are, AI agents won’t agree with you either.
What Is “Share of Model” and Why Does It Matter?
Share of model is the measure of how often and how accurately a brand appears in AI-generated results compared with its competitors. It’s the AI-era equivalent of share of voice, but with a critical difference: it doesn’t just track volume. It tracks whether the AI gets the brand right.
Jellyfish has built a software platform called Share of Model that assesses how different AI systems perceive a brand. Each model (ChatGPT, Gemini, Claude, Llama) draws on different training data, so brand perceptions vary across platforms. The tool prompts each model with structured questions about a brand and analyzes the responses for trends. “It’s similar to a human survey, but the respondents here are large language models,” said Jack Smyth, Jellyfish’s chief solutions officer, in a 2025 MIT Technology Review interview.
The underlying finding matters more than the tool. A July 2025 survey of 750 U.S. consumers by the management consulting firm Kearney found that 60% of shoppers expect to use agentic AI to make purchases within the next 12 months. Two-thirds of Gen Z and more than half of Millennials already use LLMs to research products before buying. Your brand’s AI perception is no longer a technical curiosity. It’s a commercial reality.
And here’s the part that should keep brand managers up at night: AI brand recommendations are wildly inconsistent. Research conducted by SparkToro co-founder Rand Fishkin in late 2025, involving 2,961 prompts across ChatGPT, Claude, and Google AI, found that identical prompts produce different brand recommendation lists more than 99% of the time. There’s less than a 1-in-100 chance that any two responses will recommend the same set of brands in the same order.
The inconsistency means that any single snapshot of where your brand “ranks” in AI results is noise, not signal. What matters is whether your brand shows up at all — and whether it’s described accurately when it does.
Does Your Brand Need Its Own AI Agent?
Not every brand should rush to deploy an AI agent. The decision depends on what your customers want, which depends on the nature of your product, the stakes of the purchase, and how much your customers value human involvement.
A meta-analysis by researchers Bingqing Li, Edward Yuhang Lai, and Xin Wang, published in the Journal of Marketing in 2025 and spanning 287 effect sizes across more than 119,000 participants, identified clear patterns. People are willing to delegate decisions to AI agents in low-stakes, routine contexts with predictable outcomes. They resist delegation in three categories: personally meaningful purchases where identity is involved, emotionally significant interactions like gift-giving, and high-stakes decisions where maintaining control matters.
Amazon understood this early. The company has been automating routine purchasing decisions for nearly a decade, starting with the Dash Button in 2015, expanding through Subscribe & Save in 2019 (used by 23% of U.S. Amazon customers by 2024), and arriving at the current Alexa+ AI. Alexa+ can restock groceries autonomously by checking pantry levels, referencing past orders, and confirming delivery times. For commodity replenishment, AI delegation is a solved problem.
Lamborghini understood it from the opposite direction. CEO Stephan Winkelmann has deliberately kept the company away from autonomous driving technology. “The purpose of a car like a Lamborghini is to drive it, not be driven in it,” he said. The same logic applies to premium shopping experiences where the process of discovery and selection is part of the value.
The decision doesn’t have to be binary. AG1, the global nutrition company, deployed AI agents to handle routine customer inquiries while keeping community-building interactions strictly human. The company’s team personally responds to every customer review. Since launching in 2024, their AI agents achieved perfect scores in 99% of interactions, and the company logged a double-digit percentage shift from email to AI-agent interactions. The efficiency gains freed human representatives to focus on complex issues requiring empathy.
If your product is routine, AI agents make sense. If your product is premium or emotionally significant, human connection is the value proposition. Most brands will land somewhere in between, running a hybrid model that delegates routine tasks and preserves human involvement where it matters.
How Do You Compete When Consumers Use Their Own AI Agents?
Even brands that build their own agents face a structural problem: consumers have strong reasons to prefer independent AI agents like ChatGPT, Claude, or Gemini. Independent agents are perceived as unbiased advocates acting in the user’s interest. Brand agents are perceived as serving the company.
A Salesforce survey conducted in mid-2024 found that only 42% of consumers trust companies to use AI ethically — down from 58% in 2023. And 72% of respondents wanted transparency about when they’re interacting with AI rather than a person.
Consumer agents also hold a data advantage. They collect information across all brands and domains a person interacts with, building a comprehensive profile over time. ChatGPT’s memory function, for example, retains user preferences from past conversations, enabling recommendations that draw on a person’s full purchase history rather than just their interactions with one company.
Brands can counter this with two capabilities independent agents can’t replicate. First, proprietary product knowledge. Sephora’s AI system draws on a product catalog with detailed shade and formula taxonomies, Color IQ technology that differentiates 140,000 skin tones, and first-party profiles from more than 34 million Beauty Insider members. Customers using these tools are three times more likely to complete purchases, and product returns dropped by 30%. No general-purpose agent has access to that depth of product intelligence.
Second, human escalation. ServiceNow’s AI agent resolves 80% of incoming queries autonomously and escalates the remaining 20% to human specialists. This hybrid model reduced resolution time for complex cases by 52%. Consumer agents can’t offer this — the only person involved is the consumer.
Responsible AI practices also close the trust gap. Research by Oguz Acar and colleagues, involving 3,268 participants from the UK in large-scale discrete-choice experiments, found that when responsible AI features like privacy, auditability, and transparency were embedded into product design, predicted adoption rates jumped from 2.4% to 63.2%.
The most reliable approach for brands competing with consumer agents is to combine proprietary knowledge with human oversight and transparent AI practices. Then make those advantages visible to the customer.
How Do You Optimize Your Brand for AI Agents You Don’t Control?
For consumers who rely on independent AI agents, brands need a different strategy: ensuring those agents describe and recommend them accurately. This is where the coherence question becomes a technical requirement.
OpenAI is actively building the infrastructure for AI-mediated commerce. Through its Agentic Commerce Protocol, developed with Stripe, and integrations with Shopify, ChatGPT users can now research and purchase products directly in conversation. Shopify reports that AI-attributed orders grew 11x between January 2025 and January 2026. This is not a future scenario. It’s a current revenue channel.
Instacart provides a model for integration. The company built both Ask Instacart (a ChatGPT-powered search tool within its own app) and a ChatGPT plugin that lets users add ingredients to their cart during conversations. When OpenAI introduced custom GPTs, Instacart launched its own GPT to maintain visibility on the platform. The strategy: be present wherever AI conversations about food and shopping happen, regardless of where they start.
Danone regularly monitors how LLMs portray its brands and makes targeted marketing adjustments when discrepancies arise, tracking measurable improvements in how AI agents describe and recommend its products.
For brands with less infrastructure, the fundamentals still apply. Research from Carnegie Mellon demonstrates that even subtle changes in search wording can alter brand recommendations by as much as 78.3%. Knowing how consumers formulate their queries, and ensuring your product information performs well across prompt variations, should be the foundation for content optimization.
Some brands are adopting llms.txt, a machine-readable format designed specifically for LLMs. Cloudflare, Stripe, and others have implemented it. The format allows brands to structure product information in ways AI agents can parse and prioritize. Adoption has exceeded 844,000 websites, growing at more than 500% year-over-year, though measurable citation impact remains early-stage.
The most common mistake here is treating AI optimization as a one-time technical project. Prompt-based optimization is ongoing work. AI systems evolve, consumer query patterns shift, and competitive positioning changes. Brands that treat this as a set-and-forget exercise will fall behind.
Conclusion
The rise of agentic AI hasn’t created a new problem for brands. It has made an existing problem measurable. Incoherent brand signals have always cost something — lost customers, confused positioning, wasted marketing spend. Now they cost share of model.
The Ballantine’s story is instructive not because AI made a mistake, but because the AI was working from the information available. The signals disagreed, so the perception was wrong. Fixing it required the same work that brand clarity has always required: deciding what you are, saying it consistently, and making sure every channel agrees.
AI agents are new intermediaries, not new problems. Brands that build coherent signal systems will be described accurately by AI agents because those agents will have clear information to work from. Strategy, language, design, and experience reinforcing the same meaning is what coherence looks like. Brands that don’t will discover their incoherence reflected back at them, at scale, by systems that don’t know any better.
Reputation builds the same way it erodes: one signal at a time. AI just made the signals louder.
