Should you block AI shopping agents on your store?

How much of your traffic is now AI agents?

Close to half of it. Akamai reported that AI bots account for 47.9% of commerce traffic, with agentic AI traffic growing 7,851% year over year (Akamai¹). That means the question is no longer whether AI agents visit your store but what mix of them you are already receiving, and whether your current setup treats a buying agent the same as a scraping one. For most merchants, the honest answer is that they have no policy at all, so every agent is handled identically by default.

This traffic is not one thing, and that is the crux. Some of it is the search-indexing bots that let ChatGPT, Perplexity, and Google AI surface your products to shoppers, which you want. Some is agents fetching your pages on behalf of a specific human shopper, which is often a real buyer in a new wrapper. And some is abusive automation: scraping your catalog and prices, testing stolen cards, or creating fake accounts. Lumping these together is what leads to bad decisions in both directions.

The size of the shift is why a default of no policy is now a risk. When half your traffic is automated and rising fast, treating it as background noise means you are neither capturing the buyers it contains nor stopping the abuse it hides. The rest of this guide sorts the traffic into categories you can actually make decisions about, starting with the three kinds that matter.

What are the three kinds of AI agent traffic?

The three kinds are search-indexing bots, user-fetch agents, and buy-for-me agents, and each deserves a different policy because each does something different to your business. Search-indexing bots crawl your store so AI answer engines can recommend your products; blocking them makes you invisible in AI search. User-fetch agents retrieve a specific page because a real shopper asked their assistant about it; these are often real buyers in a new wrapper. Buy-for-me agents actually complete purchases on a shopper’s behalf, which is genuine revenue if the agent is legitimate and fraud if it is not.

The distinction that matters most is legitimate versus abusive, and it cuts across the three types. A user-fetch agent retrieving your page for a real customer is welcome; a scraper harvesting your catalog and prices to undercut you is not, even though both may look similar at the network level. A buy-for-me agent completing an authorized purchase is revenue; the same shaped traffic testing stolen cards is fraud. The policy you want is not per-type but per-intent, and that requires telling the two apart.

Which AI agents actually send you buyers?

The agents that send you buyers are the search-indexing bots that make you discoverable and the user-fetch agents retrieving your pages for real shoppers, so blocking those is blocking demand. When a shopper asks ChatGPT or Perplexity for a product recommendation and your store surfaces, that visibility depends on having let the indexing bot crawl you. When their assistant then fetches your page to compare or buy, that fetch represents a real person with intent. Treating either as noise to block is turning away customers arriving through the fastest-growing discovery channel in commerce.

This is the strategic reason not to blanket-block. Amazon can afford to lock down AI agents and build its own assistant, Rufus, because it is a destination shoppers seek out directly (CNBC³). A smaller store that blocks AI discovery does not gain independence; it just disappears from the AI answers where a growing share of shoppers now start. For most merchants, being findable and buyable by legitimate agents is a channel to protect, not a threat to eliminate.

The nuance is that not every agent is a buyer, which is why you allow-and-watch rather than allow-and-ignore. A user-fetch agent is probably a real shopper; a high-velocity scraper hitting every product page is not. Allowing the traffic that represents demand while watching for the patterns that represent abuse is the balance, and it depends entirely on being able to tell them apart, which the next section covers.

How do you tell a real AI shopper from a bot?

You tell them apart by behavior and identity: legitimate agents act like a single shopper with a purpose, while abusive bots act like automation at scale, and increasingly, legitimate agents can prove who they are. A real user-fetch or buy-for-me agent retrieves a handful of relevant pages and may complete one purchase; a scraper hits hundreds of pages fast, and a card tester attempts many transactions in a short window. Velocity, breadth, and purchase behavior separate a shopper from a harvester even when both are automated.

Identity is the stronger signal, and it is where agentic commerce is heading. A legitimate agent that can present a verifiable identity, proving which platform it represents and that it is acting for a real, authorized shopper, is one you can allow with confidence, while an agent that conceals or spoofs its identity is one you should not. This is exactly the issue in the Amazon case: Amazon alleged Perplexity disguised Comet as a regular Chrome browser and refused to identify it (GeekWire²), which is the difference between an agent you can trust and one you cannot.

The durable answer is verifiable agent identity and attestation, not an ever-growing blocklist of user agents you recognize. A blocklist fails the moment a bot spoofs a known-good agent string, which abusive bots routinely do. A system that requires an agent to prove its identity and its authorization, rather than trusting a self-reported label, is what lets you allow legitimate AI buyers and block abuse without playing whack-a-mole. That is precisely what RankShield’s verifiable AI security is built to provide.

What does the Amazon versus Perplexity case mean for you?

The Amazon versus Perplexity case means the ground rules for agentic commerce are being set in court right now, and identity is at the center of them. In March 2026, a federal judge issued a preliminary injunction blocking Perplexity’s Comet shopping agent from Amazon, after Amazon argued the agent accessed its platform without authorization and concealed its identity (PYMNTS⁴). The early signal is that agents which hide what they are and act without authorization can be blocked, while the door stays open for agents that identify themselves and act for authorized shoppers.

For your store, the practical read is that you have both the right and a growing set of tools to insist on identity. You do not have to choose between blocking all AI and allowing anonymous automation; the emerging norm is to allow identified, authorized agents and refuse concealed ones. Amazon is enforcing that as a destination with leverage, but the same principle scales down: require agents to be what they claim, and you can welcome legitimate AI buyers while excluding the ones operating in disguise.

The other signal is that cooperation is coming. While Amazon litigated, Walmart and Target were testing ways to work with AI shopping platforms rather than block them outright (CNBC³). The likely future is not a wall but a handshake: agents that prove their identity and authorization get to shop, and merchants that can verify that identity get the sales. Positioning your store to verify rather than to block is how you end up on the winning side of that shift.

Should you block AI agents in robots.txt?

Do not block search and indexing AI bots in robots.txt, because that makes your store invisible in AI search, which is where a growing share of shoppers now discover products. There is a three-tier distinction that matters: training crawlers can be blocked if you do not want your content used for model training, but search and indexing bots like OAI-SearchBot, PerplexityBot, and Google’s crawlers should be allowed so your products can be recommended, and user-retrieval agents should be allowed so a shopper’s assistant can fetch your page. Blocking the wrong tier is how stores accidentally disappear from AI answers.

robots.txt is also the wrong tool for stopping abuse, because it is a request, not an enforcement mechanism. Well-behaved crawlers honor it; abusive bots and spoofed agents ignore it entirely, which is exactly the behavior at issue in the Amazon case. So robots.txt is where you express your discovery policy, allow the indexing and retrieval bots you want, but it cannot be your defense against scrapers, card testers, or agents operating in disguise. Those require enforcement at the edge and identity verification, not a politely worded file.

The right setup is therefore two layers: use robots.txt to stay visible to the AI search and retrieval bots that bring you buyers, and use behavioral limits plus verifiable agent identity to block the abuse that ignores robots.txt anyway. That combination keeps you discoverable and buyable by legitimate AI while stopping the automation that would scrape or defraud you, which is the balance every merchant needs as agentic commerce grows.