Agentic commerce: pricing when the shopper is an AI agent

A growing share of the visits landing on retail sites this year are not people. Adobe's analytics team measured a 393% year-over-year jump in AI-sourced traffic to US retail sites in the first quarter of 2026, and by March those visits were converting 42% better than traffic from paid search or email. Agentic commerce - shopping done by AI assistants and autonomous agents on a customer's behalf - has moved from conference-keynote material to a line in the channel mix. For a mid-market retailer, that changes something fundamental: a machine reading your prices does not respond to any of the things your pricing was designed to do to a human.

This is a memo about what that means for pricing strategy, and what is worth deciding now rather than in two years. The short version: agents reward retailers whose pricing is structured, consistent, and explainable - and they are indifferent to almost everything else merchants have spent decades perfecting.

What actually changes when the shopper is an agent

A human shopper experiences your prices. An agent parses them. That distinction sounds philosophical, but it has hard commercial consequences.

Consider what a typical assortment relies on to carry a price image: a sharp price on the traffic-driving items near the entrance, wider margins in the aisles, charm endings, anchor products, bundle framing, a loyalty price that makes the shelf price feel like a penalty for strangers. All of it works because a person walks a path through the assortment and forms an impression from the handful of prices they actually check.

An agent does not walk a path. It queries. Asked to find a specific cordless drill, it compares your price on that drill against every other retailer it can read, in milliseconds, with no memory of your brand's value reputation and no goodwill from the last time you were cheapest on milk. The price image you built across 30,000 SKUs collapses to the one row the agent is looking at. In that transaction, you are exactly as cheap or expensive as that single number, plus shipping, minus any discount the agent can verify.

Three properties of your pricing suddenly matter much more than they did:

• Consistency. Agents cross-check. A product priced differently on your site, your app, and a marketplace listing reads as noise at best and as manipulation at worst. Humans rarely notice channel gaps; agents notice nothing else.
• Verifiability. An agent recommending your offer wants specific, checkable facts: the price, the unit price, the shipping cost, the availability. Vague promotional framing ("up to 40% off selected lines") is unparseable and gets ignored. A shopper might be intrigued; an agent moves on.
• Stability. A price that flaps - changing several times a day with no discernible logic - is hard for an agent to cache, quote, and complete a purchase against. Some agent frameworks re-verify the price at checkout and abandon the transaction if it moved. Volatility that was invisible to humans becomes a conversion problem.

Machine readability is now part of price competitiveness

Adobe's content-visibility research adds an uncomfortable detail: retail product pages - the pages where prices live - are the least machine-readable pages retailers publish, scoring 66% on average against 75% for homepages. The best sites score above 82%, the worst near 54%. In other words, there is already a meaningful gap between retailers whose prices agents can reliably read and retailers whose prices they cannot.

This puts a new item on the pricing leader's agenda that used to belong purely to the SEO team. If an agent cannot extract your price, unit price, and stock status cleanly, you are not "expensive" in its comparison - you are absent. Structured product data, consistent identifiers, and prices that mean what they say are becoming the equivalent of being on the shelf at eye level. As one recent industry analysis put it, the first interaction with your brand increasingly happens on an AI platform, not on your homepage.

The practical checklist is unglamorous: schema markup on product pages, one canonical price per product per channel, unit prices computed correctly, shipping costs exposed before checkout, and promotional prices published with their conditions machine-readable rather than buried in a banner. None of this is a pricing algorithm. All of it decides whether your pricing algorithm's output ever gets seen.

The strategic fork: compete on price data quality, not just price level

The reflex reading of agentic commerce is "agents compare prices, so prices race to the bottom." The evidence so far points somewhere more interesting. Agents optimize for the shopper's stated intent - which includes reliability, delivery time, return terms, and whether the recommendation turned out to be correct last time. An agent that sends its user to a retailer whose advertised price did not survive checkout has failed at its one job, and the platforms building these agents penalize that hard.

That creates an opening for mid-market retailers that cannot win a pure price war against the largest players: become the retailer whose prices are always exactly what they claim to be. Accurate stock, honest promotions, stable prices, no drip-priced surcharges. For a decade, that discipline earned a modest trust premium with humans. With agents, it compounds - because agent platforms remember, at scale, which merchants' data can be trusted, and route demand accordingly.

The inverse is also true. Every pricing shortcut that quietly worked on humans - the inflated reference price, the "was" price that never was, the fee revealed at the last step - is now being audited continuously by software with perfect recall. Regulators needed years to catch reference-price abuse. Agents notice in one crawl cycle.

Where agentic commerce lands first

Agentic commerce will not arrive evenly across the assortment, and the pricing response should not be even either. Agents are strongest where the purchase is specification-driven: the customer knows the product, or can describe it precisely, and the job is to find the best verified offer. That describes consumer electronics, DIY and tools, auto parts, pet food, supplements, and most replenishment purchases - categories where the barcode is the decision and the retailer is interchangeable.

It describes far less of fashion, fresh food, and anything bought on inspiration, where the human still browses and the agent at most assists. A sensible posture is to treat your specification-driven categories as the beachhead: that is where your prices are already being compared programmatically today, where machine readability pays back first, and where a competitor with cleaner price data can quietly take share without ever out-discounting you. Adobe's consumer research found 39% of shoppers already using AI in their online shopping journey - concentrated exactly in these research-heavy, comparable purchases.

What to decide this quarter

None of this requires betting the company on agents. It requires a handful of decisions that are cheap now and expensive to retrofit:

1. Audit what agents can see. Run your top 500 traffic and revenue products through a machine-readability check: is the price in structured data, is the unit price right, is availability truthful, do promotional conditions parse? Score yourself the way an agent platform would.
2. Decide your consistency policy. Which price differences across channels and markets are deliberate strategy, and which are drift? Write the deliberate ones down as rules. Kill the drift. An agent will surface every inconsistency you did not choose.
3. Put stability into the pricing cadence. Decide how often prices may change in agent-exposed categories and hold to it. Repricing that reacts to every competitor tick was already of questionable value against humans; against agents that re-verify at checkout, it actively costs conversions.
4. Make every price explainable. When an agent platform, a marketplace, or a regulator asks why a price was what it was on a given day, the answer needs to come from a rule you can show, not from a model nobody can interrogate. Explainable, auditable, rules-based pricing was good governance before agents; it is now also a commercial requirement, because the systems reading your prices keep records.

The common thread: agentic commerce punishes improvisation and rewards structure. Retailers running pricing from a patchwork of spreadsheets and instinct will not fail loudly - they will just quietly stop appearing in recommendations, and the analytics will show a soft decline in a channel nobody owns. Structured pricing - rules you can state, prices you can trace, data an agent can trust - is what keeps you in the comparison.

What this does not change

Honesty about scale: AI-sourced traffic is growing at triple-digit rates from a small base. For most mid-market retailers it is low single digits of sessions today, and store revenue dwarfs it. Nobody should be rebuilding their pricing organization around a channel this size - the argument is about slope, not intercept, and the slope is unusually steep.

Agents also do not repeal the fundamentals. Your cost base still sets the floor. Key-value items still shape human price perception in stores, which is still where most of the money is. Assortment, availability, and service still decide loyalty. And the retailers that do well against agents will mostly be the ones that were already disciplined: clean data, deliberate rules, prices that survive scrutiny. Agentic commerce does not introduce a new virtue - it raises the price of not having the old ones.

The uncomfortable question for a leadership team is not "do we have an AI agent strategy?" It is simpler: if a machine audited every price you publish tomorrow morning, would the results read like a strategy or like an accident? If the honest answer is the second one, that is fixable in a quarter - and it is worth fixing before the machines doing the auditing are the ones bringing you customers. If you want to see what your pricing looks like when every rule is explicit and every price traceable, we are happy to walk through it with your own data.