AI Agent Traffic: Should Retailers Block or Allow It?

AI Agent Traffic: The Retail Decision Behind the Robots File

AI agent traffic has gone from a curiosity in the server logs to a board-level retail question in under a year. Adobe expects AI-assisted traffic to US retail sites to rise around 130 percent this holiday season compared with last year. OpenAI has signed deals letting shoppers buy from Target, Instacart, and DoorDash inside ChatGPT, Walmart is working with both Google and OpenAI, and Shopify added Meta as an AI sales channel in September before shipping browser tools that let agents complete checkout. Amazon has gone the other way, reportedly blocking Meta's shopping agent and continuing to resist outside agents on its marketplace. Every retailer now has to make the same call, whether they make it deliberately or by default: what happens when a machine, not a person, comes to shop?

Block, Allow, or Verify

There are really three postures, and each trades something a retailer values for something else it values.

Why AI Agent Traffic Forces a Decision Now

For years, the only automated visitors most retailers worried about were search crawlers they wanted and scrapers they did not. The new traffic is different because it carries purchase intent. An agent browsing on behalf of a shopper is not indexing your catalogue for later. It is comparing your price, stock, delivery date, and returns policy against three competitors right now, and it may complete the purchase without a person ever seeing your homepage.

That changes the economics of a visit. Many retailers spent a decade building revenue around what happens while a human is on the site: sponsored product slots, cross-sell, loyalty sign-ups, and impulse additions. An agent skips almost all of it.

What AI Agent Traffic Looks Like in Your Logs

It helps to separate the categories, because they deserve different policies. Training crawlers collect content to build models. Retrieval fetchers pull a page to answer a specific question. Browsing agents operate a real browser on a shopper's behalf, sometimes indistinguishable from a person. And protocol-based agents use structured checkout integrations agreed with the platform. A single robots rule cannot handle all four sensibly, and robots rules were never enforceable in the first place.

What Amazon and Walmart Are Each Betting On

The two largest US retailers have taken opposite positions, and both make sense on their own terms. Amazon is a destination. People start product searches there, and its advertising business depends on those searches happening on its own pages. Letting an outside agent shop on Amazon would hand that starting point, and the data that comes with it, to someone else. Amazon took Perplexity to court over agent shopping last year, and the reported block on Meta's agent fits the same logic.

Walmart and Target are betting that the starting point is moving whether they like it or not, and that being the easiest retailer for agents to buy from is worth more than defending a front door shoppers may stop using. Neither bet is obviously wrong. The mistake would be copying either one without checking which kind of business you actually are.

The Retail Media Problem

Retail media has become one of the most profitable lines in modern retail, and it rests on a simple premise: a person scrolls a results page and sees sponsored placements. An agent does not scroll, does not see banners, and does not reward a brand for paying to be in slot one. If a meaningful share of shopping moves to agents, the advertising income attached to on-site search shrinks with it.

That is why the verified-agent posture is likely to involve commercial terms rather than just technical ones. Retailers will want something in exchange for letting an agent bypass the storefront: a share of the transaction, access to aggregated intent data, or a sponsored signal the agent is allowed to weigh. Expect those negotiations to look a lot like the early fights between publishers and search engines.

Six Questions Before You Set a Policy

1. Are you a destination or a supplier to destinations? If shoppers start their search with you, protecting the front door matters more.
2. How much of your margin comes from retail media and on-site upsell? The higher the share, the more an agent visit costs you.
3. Can you tell agents apart today? Check whether your bot management separates verified agents, browsing agents, and scrapers.
4. What does a fraudulent agent purchase look like? Model the chargeback, account takeover, and inventory-hoarding scenarios before you open up.
5. Is your product data accurate enough to be read by machines? Agents act on what they find, including stale prices and wrong stock.
6. Who owns the customer after an agent purchase? Decide what data you need back to handle returns, warranty, and service.

Pros and Cons of Letting Agents In

• Pro: new demand. Shoppers who would never have visited your site can now buy from you through an assistant.
• Pro: holiday reach. With agent-assisted traffic growing fastest in peak season, being purchasable matters most when volume is highest.
• Pro: lower acquisition cost. A strong offer can win the agent's recommendation without paid search spend.
• Con: thinner relationship. You may never learn who the customer is, which weakens loyalty and lifetime value.
• Con: price pressure. Agents compare tirelessly, which rewards the lowest delivered price above everything else.
• Con: new abuse patterns. Agents can be used to snap up limited stock or test stolen payment details at scale.

Real Scenarios Worth Thinking Through

These scenarios are illustrative, showing how AI agent traffic plays out in practice rather than presented as verified case studies.

A mid-sized home goods retailer blocks all automated traffic during peak season to protect against scalping. Its products disappear from agent-led gift recommendations in the weeks that matter most, while a competitor that allowed verified agents picks up the demand. The block worked technically and failed commercially.

An electronics retailer allows agents openly and sees order volume rise. Three months later, finance notices that retail media income per order has fallen sharply and returns from agent purchases run higher than average, because agents optimised for price over fit. The net effect is close to zero.

A fashion retailer opens only to agents that authenticate and agree to pass back order and contact details. Volume grows more slowly, but the retailer keeps the customer record and can handle service and returns properly. It is a slower path that preserves the relationship.

“After testing multiple AI models on coding, research, and business prompts, combined outputs produced more reliable results than any single model.” Internal multi-model evaluation, Talkory research team.

Agents Act on What Models Believe

Whatever access policy a retailer picks, one thing is easy to overlook. Shopping agents do not only read your site at the moment of purchase. They decide which retailers to consider based on what the underlying model already believes about you: your price position, your delivery speed, your returns window, whether you stock a brand. If a model thinks your returns window is fourteen days when it is thirty, an agent may steer a cautious shopper elsewhere before it ever fetches a page.

Those beliefs differ between models, and they are often out of date. Checking them is part of agent readiness, alongside the product feed and checkout work we covered in agentic commerce. Payment networks and bot management vendors are also building ways for agents to prove who they are and who they act for, which makes the verified posture far more practical than it was a year ago.

Why Talkory Wins

Retailers deciding on agent policy need two things: a clear view of how AI models currently describe them, and a sound read on a fast-moving question. Talkory runs the same prompt across GPT, Claude, Gemini, Grok, Perplexity Sonar, and Kimi K3 in one pass. Ask what each model believes about your returns policy, delivery times, or price position, and errors show up immediately as disagreement. Ask the strategic question, block, allow, or verify, with your own numbers, and you see which arguments hold across six independent models and which depend on one model's assumptions. It is a decision aid for a policy that will shape your next several peak seasons.

Final Verdict

AI agent traffic is not a technical nuisance to be handled in the robots file. It is a channel decision with consequences for data, margin, retail media, and fraud. Blocking makes sense for true destinations with large advertising businesses. Open access suits retailers who are already competing mostly on price and availability. For most, the answer will be verified access on negotiated terms, with careful measurement of what each agent channel actually returns. Whichever path you choose, check what the models already believe about you, because agents will act on it.