Your brand’s most important audience can’t see your logo.

That’s not a metaphor. Gartner reckons that by 2028, 90% of B2B purchases will be handled by AI agents: autonomous software with budgets, preferences, and decision criteria, routing more than $15 trillion through automated exchanges. These agents don’t respond to colour palettes. They don’t feel your vibe. They parse specifications, structured metadata, pricing feeds, and machine-readable reviews. And if your brand isn’t legible to them, you don’t exist.

The HBR piece that prompted this — Acar and Schweidel’s “Preparing Your Brand for Agentic AI” in the March 2026 issue — opens with a revealing anecdote. Pernod Ricard’s head of digital discovered that two-thirds of Gen Z consumers were already using LLMs to research products. So he teamed up with Jellyfish to audit how the major AI models actually represented his brands. The results were grim. One model categorised Ballantine’s, an affordable mass-market Scotch, as a prestige product. The brand’s entire positioning, years of careful market segmentation, overridden by a hallucinating algorithm that couldn’t tell the difference.

And that’s the bit most marketing teams haven’t caught up with yet. Your brand isn’t just what you say it is any more. It’s what the models say it is. And the models are increasingly the ones doing the buying.

The fork in brand strategy

This isn’t the death of brand. It’s a fork. You need a human brand and an agent brand, and they operate on completely different logic.

Your human brand still matters — and for some businesses, it matters more than ever. People still buy emotionally, still respond to storytelling, still choose the wine with the label that catches their eye. A hand-thrown pot from a ceramicist in Stoke-on-Trent isn’t competing on structured data. A boutique strategy consultancy doesn’t win work because its schema markup is cleaner than McKinsey’s. These brands live or die on reputation, relationships, and the kind of trust that doesn’t fit in a JSON field. That’s not going away. But a growing slice of purchasing, starting in B2B and bleeding fast into consumer markets, is being mediated by software that doesn’t care about any of it.

An AI procurement agent evaluating suppliers doesn’t browse your website the way a human does. It looks for structured product data in schema.org format. It reads your API documentation. It checks machine-readable reviews and pricing feeds. It compares your specifications against a rubric that was set by a human, then executes autonomously. Google launched its Universal Commerce Protocol in January 2026 specifically to enable this — an open standard for agentic commerce built on structured product data.

The marketing discipline has spent decades getting brilliant at emotional differentiation. Tone of voice, brand purpose, visual identity, experiential campaigns. All of it designed for an audience with eyes, ears, and feelings. An agent parsing JSON couldn’t care less.

What agents actually see

Jellyfish coined the term "Share of Model" to describe how brands appear inside LLMs — and they built a platform to measure it. The concept is a useful reframe. Just as Share of Voice measured how visible your brand was in traditional media, Share of Model measures how an AI represents you when someone asks it for a recommendation.

The data is striking. Schema-compliant pages get cited roughly three times more often in AI-generated overviews. Three times. That’s not a marginal SEO advantage — it’s the difference between existing and being invisible to the fastest-growing discovery channel in commerce. In content marketing terms, AI exposed which content was built on substance; the same reckoning is now arriving for product information.

So what do agents actually parse? Product specifications with consistent, structured markup. Transparent pricing (not “contact us for a quote,” which is invisible to an automated buyer). Machine-readable reviews aggregated with proper schema. API endpoints that let an agent pull real-time inventory or configuration options. Clear, unambiguous categorisation that a model can reason about.

If you’ve spent your marketing budget on brand films and experiential activations but your product pages are unstructured HTML with creative copy and no schema markup, you’ve built a beautiful shop with no door that an agent can find.

This is already happening in B2B

The shift isn’t theoretical. I’ve written before about how agentic loops are becoming the operating system for business. Autonomous decision cycles that observe, orient, decide, and act without waiting for a human to push the next button. Procurement is one of the first domains where those loops are closing entirely.

Korn Ferry’s 2026 report found that more than half of leaders plan to add autonomous agents to their teams this year. Forrester predicts procurement teams will deploy agents capable of negotiating across hundreds of suppliers simultaneously. That’s not a copilot helping a buyer shortlist vendors. That’s software making purchasing decisions at scale, at speed, against criteria that are entirely programmatic.

The brands that win in that environment? Cleanest data. Not best tagline.

I spent years in marketing watching companies pour budget into brand awareness campaigns while their product data sat in inconsistent spreadsheets that even their own sales team couldn’t navigate. The irony of it. You’d spend a fortune making people feel something about your brand, then lose the deal because your spec sheet didn’t match what was on the website. Agents have zero patience for that. They won’t call your sales team to clarify. They’ll just move to the next supplier in the feed.

The consumer market is next

B2B is the leading edge, but consumer markets aren’t far behind. Two-thirds of Gen Z already use LLMs to research purchases. That number isn’t going down. And the gap between "I asked ChatGPT what running shoes to buy" and "my AI assistant ordered running shoes based on my preferences and budget" is narrowing fast.

When that happens — and it’s when, not if — the brands that win will be the ones whose product information is structured, accurate, and machine-readable. Instagram followers won’t matter much. The ones where an agent can confirm that this specific shoe, in this size, at this price, meets the criteria the human set — and complete the purchase without friction.

The Pernod Ricard story is a preview. Ballantine’s had decades of carefully managed brand positioning, and a single LLM wiped it out by miscategorising the product. Now imagine that at scale, across thousands of purchasing decisions a day, with no human in the loop to catch the error. That’s the world we’re heading into.

But doesn’t this just commoditise everything?

It’s a fair question, and one I keep circling back to. If every brand’s product data is equally clean, equally structured, equally machine-readable — doesn’t the agent just sort by price and pick the cheapest? Doesn’t the whole thing collapse into a procurement spreadsheet?

For commodity products, probably yes. If you’re selling printer paper or bulk fasteners, an agent comparing structured feeds will optimise ruthlessly on price, availability, and delivery terms. That’s not new — procurement has worked that way for decades. Agents just make it faster.

But most products and services aren’t commodities, and the rubric an agent uses is set by a human. If I tell my purchasing agent “find me a brand strategist with experience in regulated industries and a track record with companies under 50 employees,” that agent isn’t sorting by hourly rate. It’s looking for structured evidence of expertise, client testimonials with proper markup, case studies that a model can parse, and credentials it can verify. The human decided what matters. The agent finds who matches.

This is where services get interesting. A law firm, a design agency, a specialist consultancy — these can’t compete on a spec sheet the way a widget manufacturer can. But they can make their expertise legible. Structured case studies. Machine-readable client outcomes. Schema-marked service descriptions that an agent can actually reason about, rather than a homepage full of stock photography and the phrase “we deliver bespoke solutions.”

For boutique brands, the calculation shifts further. A heritage watchmaker isn’t worried about an agent comparing quartz movements on price. Their buyer set a brief that includes “Swiss-made, independent brand, hand-finished movement” — and the agent needs structured data to confirm those attributes. The intrinsic value is still there. The agent just needs to be able to read it.

The risk isn’t that structured data commoditises your brand. The risk is that without it, an agent can’t see what makes you different — and defaults to the competitor it can parse.

What to actually do about it

This isn’t a panic piece. The practical implications are concrete, and most of them are things marketing teams should be doing anyway. They’ve just never had this level of urgency behind them.

Audit your structured data. Not your brand messaging, not your positioning statement. Your actual product markup. Is it schema.org compliant? Is it consistent across platforms? Would an agent parsing your product page get an accurate, complete picture of what you sell, at what price, with what specifications? If your marketing team can’t answer that question, the answer is probably no.

Build your agent brand alongside your human brand. Two parallel tracks. The human track is storytelling, emotional resonance, visual identity — everything you already know how to do. The agent track is structured data, API access, transparent pricing, machine-readable reviews, and consistent product taxonomy. Most companies are investing heavily in the first track and barely acknowledging the second exists.

Monitor your Share of Model. Ask the major LLMs about your products and see what comes back. If they’re miscategorising you, attributing features you don’t have, or recommending competitors instead, that’s a problem you can address — but only if you’re looking.

Make your pricing machine-accessible. “Request a quote” is a conversion mechanism designed for humans. For an agent, it’s a wall. If your competitor’s pricing is structured and yours requires a human interaction, the agent will route the purchase to your competitor every time. The five levels of agentic maturity apply to your customers’ buying systems too — and the ones at level three and above won’t wait for your sales team to respond.

Treat your product data as a brand asset, not a back-office function that the marketing team delegates to an intern. Your product data is increasingly your primary brand expression to a growing segment of buyers. Fund it accordingly.

The question nobody wants to answer

Marketing has always been about making people feel something. That’s not changing, but it’s no longer sufficient. The discipline now has to serve two audiences with fundamentally different needs, and the one that’s growing fastest doesn’t experience feelings at all.

The brands that recognise this fork early will build for both. Everyone else will keep creating beautiful campaigns that land perfectly with humans and wonder why their pipeline is drying up, as agent-mediated purchasing quietly routes spend to competitors with cleaner data.

Your brand still needs a soul. But it also needs a schema. The first gets you loved. The second gets you found.

Quick answers

What is an agent-first brand?

An agent-first brand is one that’s been optimised for discovery and selection by AI agents, not just human buyers. It means structured product data, machine-readable pricing, schema.org markup, and API-accessible information — everything an autonomous purchasing agent needs to evaluate and buy your product without human intervention.

Why does brand positioning fail with AI agents?

AI agents don’t process emotional messaging, visual identity, or brand storytelling. They parse structured data, specifications, and metadata. A brand that relies purely on emotional differentiation is invisible to agent-mediated purchasing, which Gartner predicts will handle 90% of B2B transactions by 2028.

What is Share of Model?

Share of Model is a concept and measurement platform developed by Jellyfish that tracks how AI language models represent and recommend brands. It’s the AI equivalent of Share of Voice — instead of measuring media visibility, it measures how accurately and favourably LLMs describe your products when consumers ask for recommendations.

How do I make my brand visible to AI agents?

Start with structured data. Audit your product pages for schema.org compliance, ensure pricing is transparent and machine-readable, aggregate reviews with proper markup, and provide API endpoints for real-time product information. Monitor what major LLMs say about your brand and correct inaccuracies through better structured content. Treat product data as a brand asset, not a back-office task.