Your Brand Story Doesn’t Work on AI Agents — Your Reputation Record Does
The Assumption Hidden Inside “Brand”
For the last two decades, brand-building rested on one foundational assumption: the buyer is a human who can be moved. Not manipulated — moved. By a story, by trust earned through consistent voice, by the subtle social proof of design and copy that signals “people like you buy this.” Persuasion was the primary medium through which brand communicated value to uncertain buyers.
What agentic commerce does is remove the being-who-can-be-moved from the decisive moment. An AI agent evaluating a purchase on a shopper’s behalf is not susceptible to narrative the way a human is. A 2026 Harvard Business Review study by Sabbah and Acar found that classic persuasion tactics — scarcity messaging, countdown timers, strike-through pricing — did not reliably move AI shopping agents and could actually backfire with stronger models, which grew skeptical of obvious influence attempts. Only star ratings and price consistently mattered across agents tested.
The instinct from there is to conclude that brand is dead and data wins. That conclusion is also wrong.
Agents do evaluate brand — they evaluate its machine-legible proxies. Star ratings, review volume, review recency, the text of what reviewers actually said, return policies, shipping times, fulfillment accuracy. These are not decorative data points. They are the form that trust takes when the buyer cannot be charmed. Your brand reputation does not disappear in an agent-mediated transaction; it gets re-encoded into a structured evidence surface that an automated evaluator can parse without reading your About page.
So brand bifurcates. The story and emotional layer still matters — for the humans who set the agent’s preferences in the first place, and for every purchase that agents never mediate. But for the agent-evaluated transaction, brand must exist as verifiable, structured evidence of reliability. You cannot persuade an agent. You can only qualify for its consideration by having a reputation record it can read.
Agents read your reputation. They cannot feel your story.
Tomorrow’s forecast, in your inbox.
One email. Five minutes. Written for owners, not engineers.
The Moving Target Nobody Warned You About
Even if you build an excellent machine-legible reputation surface, the game is not stable. A peer-reviewed study (arXiv 2508.02630, presented at the ACM Web Conference 2026) examined how AI agents actually make product selections and found something that should make any owner pause: agents exhibit provider-specific position biases, tend toward choice homogeneity — different shoppers using the same model end up buying from the same narrow set of merchants — and product market-shares among agent-selected merchants reshuffle sharply when the underlying model is updated.
There is no stable endpoint where you “win the agent game.” The criteria agents use to evaluate selection are partially opaque, model-dependent, and subject to change without notice every time a model provider ships an update. A business that ranks well for one version of ChatGPT’s shopping evaluation may rank differently after the next training cycle, through no change of its own.
This matters for how you think about the practical work involved. Building a machine-legible reputation surface is an ongoing discipline — more like maintaining a credit score than deploying a website. The score reflects cumulative behavior, degrades if you stop tending it, and is evaluated by systems whose exact criteria you do not fully control.
Where Agents Actually Look — and What Gets You There
A persistent misconception in discussions of agentic commerce is that the answer is adding schema.org markup to your product pages. Independent testing of ChatGPT and Perplexity suggests these agents treat JSON-LD structured data as ordinary page text and parse unstructured prose without difficulty. Missing schema does not make you invisible to agents. The more accurate statement is that missing schema leaves you marginally less legible in edge cases — but it is not the determining factor.
Being understandable by an AI is not enough — your product has to actually exist in the data sources and commerce infrastructure agents pull from. Agents are built on top of specific programs and pipelines; if your products are not present in the structured product feeds and catalogs those programs draw on, you are not in the consideration set, regardless of how good your website is, how clean your markup is, or how compelling your brand story is. Legibility is a prerequisite, not a strategy.
Getting into the feed is the floor. What the agent sees once it is there is your reputation record: ratings, reviews, return and shipping terms, fulfillment history.
The one consequential action this moment calls for: audit where your products and reputation actually exist as machine-readable data. Verify that your star ratings, review corpus, and return and shipping terms are structured and current. Confirm your presence in whatever merchant feeds and catalog infrastructure are relevant to your category. Treat this as a maintained business asset, not a project you close.
The owner who is best at human persuasion — sharpest copy, most carefully tested funnel — carries the largest blind spot here. The skills that earned them their current customers are precisely the skills that do not transfer to being selected by an agent. That expertise gives them a reason not to look.
The Forecast
As agent-mediated buying grows from a fraction of transactions toward a meaningful share, competitive advantage will shift toward businesses with maintained, verifiable reputation surfaces and consistent merchant-feed presence — away from businesses that optimize persuasion alone.
The position: businesses that treat machine-legible reputation as a maintained asset — review volume and quality, ratings recency, fulfillment track record, feed presence — will hold a durable edge in agent-mediated transactions. Businesses that rely primarily on persuasion-layer brand signals will find that edge eroding as the share of agent-mediated demand grows.
How this could be wrong: agent adoption in commerce could plateau well below levels that make this a decisive variable, especially if trust and privacy concerns slow consumer uptake. Platform providers could add translation layers that flatten feed-optimization advantages. Model updates could redistribute selection in ways that are effectively random, rewarding patience over investment. Regulation in the EU or elsewhere could slow agentic checkout deployments significantly. Any of these would extend the timeline or reduce the magnitude of the shift — they would not reverse the direction.
Sources: Salesforce “Cyber Week 2025 Shopping Recap” and holiday shopping data; Adobe Analytics generative-AI shopping traffic data, holiday season 2025; OpenAI Agentic Commerce Protocol announcement, Sept 2025; Google agentic checkout blog post, Nov 2025; Shopify commerce-for-agents infrastructure announcements, 2025; Google – Shopify Universal Commerce Protocol announcement, Jan 2026; Sabbah & Acar, “Traditional marketing doesn’t work on AI shopping agents,” Harvard Business Review, 2026; arXiv 2508.02630 (ACM Web Conference 2026), agent product-selection bias and market-share instability.