The Honest Guide to AI Claims in B2B Commerce

Separating AI substance from AI marketing is harder than it should be. The claims are everywhere, the pressure to act is real, and the gap between what’s being sold and what’s been proven is wider than most vendors will admit.

This article looks at three claims circulating loudly in B2B right now:

• that “AI-powered” products deliver what they promise
• that agentic commerce is ready for manufacturers and distributors
• and that LLMs are reshaping how buyers actually buy

We looked at what the research says about each one, brought in commerce analyst Heather Hershey’s perspective, and focused on what deserves your attention versus your budget.

1. AI Washing: When ‘AI-Powered’ Means Very Little

The first filter you need to apply is simple: is this actually AI, or is it just automation with a higher price tag?

AI washing has become the greenwashing of the tech sector. Vendors frequently label basic rule-based scripts as ‘AI-powered’ to inflate their value. Recently, this has evolved into agent washing, where standard chatbots are rebranded as autonomous agents.

The cost of falling for this is real. Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs and unclear business value. Most of what’s being pitched as agentic today, the firm notes, are “early-stage experiments driven by hype and often misapplied.”

Part of why this keeps working is that AI is genuinely hard to pin down. Heather Hershey, commerce analyst and former AI researcher, put it directly on the B2B Commerce UnCut podcast: “AI is a big, nebulous umbrella that can mean essentially everything and nothing all at once. That’s why it’s horrible as a marketing term.”

She’s right. Under that umbrella you’ll find:

• machine learning (ML) models that train on historical data to make predictions
• deep learning systems that handle more complex pattern recognition
• large language models (LLMs) built specifically to process and generate language
• and genuinely agentic systems that reason and act with minimal human direction.

These are fundamentally different technologies with different cost profiles and different fits for different problems. A workflow engine that routes orders based on rules you wrote is not an AI agent. A chatbot that retrieves answers from a knowledge base is not the same as a system that reasons, decides, and executes autonomously.

Where AI actually earns its keep

Traditional ML has a well-documented track record in operations: demand forecasting, order routing, predictive maintenance, pricing optimization. Measurable, proven, no agentic required.

LLMs work best on specific problems. The AI use cases that hold up for manufacturers and distributors:

• Converting POs and RFPs that arrive as PDFs or emails into structured data
• Natural language interfaces for ERP and commerce-tools analytics — useful for anyone who isn’t a power user
• Content at scale: product descriptions, spec sheets, RFP responses
• Customer self-service for product discovery, search, and configuration

Knowing what each type of AI is actually good for makes the vendor conversation clearer.

Three questions that cut through vendor noise:

• What outcome does this produce and how is it measured? Vague answers signal vague value.
• What is the AI actually deciding versus executing rules you already defined? Faster automation is still automation.
• Where does it fail and what’s the fallback? If a vendor can’t answer this, they haven’t thought seriously about production.

2. Agentic Commerce: Real Concept, Wrong Timeline for B2B

Agent washing gets easier to spot once you know what genuine agentic AI is supposed to do. Which brings us to the claim generating the most noise in commerce right now.

IDC analyst Heather Hershey defines agentic commerce by three conditions: autonomous purchase, autonomous payment, and autonomous fulfillment. The AI makes the decisions — not just assists with them. Most things currently marketed as agentic commerce don’t meet all three. Many don’t meet any.

For B2B specifically, the gap between the pitch and the reality is even wider. B2B commerce runs on EDI, audit trails, and contractual data exchange — systems that persist precisely because they’re traceable, standardized, and reliable. Agentic AI is probabilistic by design. Bridging those requires new validation layers, governance structures, and fallback logic.

The path forward requires AI that’s built into the same layer handling pricing rules, buyer permissions, and approval workflows, not added on top. When AI has access to that business logic natively, adding more capable AI becomes an architectural evolution.

But the timeline question misses a different pressure. Even if the infrastructure isn’t ready, are buyers already moving? That’s where the pressure is actually coming from.

3. LLMs and B2B Buying: What’s Changing in 2026

Two questions are worth separating here because the market keeps treating them as one.

Are LLMs changing how B2B buyers research and shortlist vendors? Yes, and meaningfully. One in three B2B buyers now use AI tools to surface options. That number is climbing.

Are they purchasing through those platforms? That’s a different story.

LLMs are taking share in informational search: comparisons, category research, early-stage evaluation. Purchase-intent queries sit below 2.1% across all platforms. Buyers are using AI to build their shortlist, then doing what they’ve always done: verifying, aligning internally, and deciding through existing channels.

But LLM platforms are actively building toward that purchasing side too.

• OpenAI launched its Agentic Commerce Protocol in late 2025.
• Google followed with the Universal Commerce Protocol in January 2026.

Merchants should pay attention to what participation in these ecosystems historically costs. Hether Hershey put it plainly: “If enough merchants participate, any early advantage disappears. Eventually it commoditizes the front end of commerce and turns most merchants into dropshippers for these LLM platforms.”

Amazon Business offers a useful reference point. Sellers who joined early for the reach watched fees climb, customer data flow to the platform, and Amazon launch competing products using their own sales data.

What to Actually Take Away

AI in B2B commerce is neither the revolution being sold nor the distraction skeptics claim. The gap is between where the technology is and where the marketing says it is. And that gap is expensive if you miscalibrate.

A few things worth holding onto:

• Not every problem needs an LLM. Demand forecasting, pricing optimization, order routing – traditional ML and workflow automation have been delivering on these for years, quietly. Match the tool to the problem, not the other way around.
• Native AI beats bolted-on AI. If you’re evaluating AI for commerce, look at whether it’s part of the platform or wired in from outside. Native means connected to your pricing logic, buyer permissions, and business rules from day one, resulting in better control, better guardrails, and lower cost.
• Start narrow. Use AI where the ROI is obvious first, like repetitive workflows, manual data entry, and unstructured document processing. Prove the value in the specific case before expanding.
• Ask vendors one question: what specific outcome does this produce, and how is it measured? If the answer wanders, so will the results.
• Take security seriously before it becomes urgent. Most companies are building AI capabilities faster than AI defenses. That order should probably be reversed.

Hype creates urgency. Evidence creates results. Know which one you’re buying.