A federal judge blocked Perplexity's AI shopping bot from accessing Amazon on March 9, 2026. The same week, Amazon expanded its own Buy for Me feature to cover tens of millions of products. If you think this is about protecting customers from bots, you are reading it wrong. Amazon is locking the discovery stack for itself, and if your listings are not built for how AI agents read them, you are already losing ground.
Key Takeaways
- Amazon sued Perplexity in November 2025 and won a preliminary injunction in March 2026 blocking its Comet agent from shopping on its platform
- The same week, Amazon expanded Buy for Me through third-party product feeds to tens of millions of products from 400,000+ merchants
- Rufus and Buy for Me read structured specs, Q&A, and review content, not just title keywords
- Sellers built for keyword search but not agent-layer discovery are being skipped at the recommendation stage
- The fix is not a new keyword strategy. It is structural: listing completeness, Q&A hygiene, and spec-level accuracy
What Amazon actually did, and why the timing matters
The Perplexity case is straightforward on the surface. Amazon sued Perplexity in November 2025, alleging that its Comet AI agent disguised itself as a regular Chrome session to shop on behalf of users without Amazon's authorization. Amazon had issued five prior warnings starting in November 2024, blocked the bot in August 2025, and Perplexity bypassed the block within 24 hours. The judge found strong evidence of unauthorized access and granted Amazon a preliminary injunction, per reporting from Search Engine Land.
But the ruling was not issued in a vacuum. Amazon has consistently worked to block third-party AI agents from browsing and transacting on its platform, as Digital Commerce 360 reported, while simultaneously building its own agent-layer discovery tools. When Amazon blocks Perplexity's agent and expands Buy for Me in the same news cycle, the message is clear: discovery runs through Amazon's infrastructure, or it does not run at all.
This is the same playbook Amazon ran with Buy with Prime, with Rufus, and with A+ Content. Amazon builds the tool, sets the rules, and determines whose products surface. Sellers who understand this structure can work with it. Sellers who ignore it get filtered out.
How Buy for Me actually works for the customer
When a customer searches Amazon for a product that is not sold directly in Amazon's store, they may see Shop Direct results. These are products from external merchant websites pulled into Amazon's interface through third-party feeds from services like Feedonomics, Salsify, and CEDCommerce, as confirmed by Amazon's official announcement.
The customer has two options: click Shop Direct to go to the merchant's site, or click Buy for Me and let Amazon's AI agent complete the purchase on their behalf using their saved payment and shipping details. The customer confirms the order, but they never leave Amazon's environment in any meaningful sense. They do not browse your product page. They do not read your bullets. Amazon's agent reads the listing, confirms it matches the customer's query, and completes the transaction.
Shop Direct currently covers over 100 million products from more than 400,000 merchants, with tens of millions of those products eligible for the Buy for Me purchase flow, according to Amazon. If your product is in this system and a customer uses Buy for Me, the sale happens because Amazon's agent determined your listing was the right match. The customer's experience of your brand starts after the purchase, not during it.
What the agent layer reads, and what it skips
Rufus is Amazon's AI shopping assistant, and it powers both the conversational search experience and the recommendation logic behind Buy for Me. Based on Amazon's own technical documentation and published research, Rufus draws from four main sources to answer customer queries.
Structured product specs: battery life, materials, dimensions, certifications, compatibility. These are the technical attributes in your product detail page. If they are missing or vague, Rufus cannot confidently answer "is this waterproof?" or "will this fit a 2022 model?"
The Q&A section on your product page is not just a customer service tool. It is machine-readable structured data. Questions and answers about real usage scenarios are exactly what Rufus surfaces when a customer asks a conversational query about your product.
Rufus also extracts patterns from review content, in particular around use cases, complaints, and recurring observations. Review sentiment and detail level affect how Rufus characterizes your product in recommendations.
Category attributes round out the picture: the product type taxonomy, browse node, and category-specific fields you filled out (or skipped) during listing setup. These fields tell Amazon what your product is at a categorical level, separate from your title keywords.
What Rufus does not rely on: keyword density in your title, the order of your bullet points, or backend search terms. Those still matter for traditional A9 search. But an agent making a purchase recommendation on behalf of a customer is working from a different input set entirely.
The invisibility problem sellers are not talking about yet
Most sellers running solid PPC and ranking well on traditional search have not touched their listing structure with agent-layer readability in mind. Their titles are keyword-heavy but spec-light. Their Q&A sections have a handful of outdated questions with no systematic answers. Their product attributes have enough filled in to pass Amazon's listing quality check, but not enough to satisfy a conversational query from an AI.
The result is an invisible wall. A customer asks Rufus "what's a good protein shaker that won't leak and fits in a car cupholder?" and Rufus surfaces products with clean dimensional specs, confirmed leak-test answers in their Q&A, and verified cupholder-compatible size attributes. Your product, which ranks fine for "protein shaker BPA free 28oz," does not make the cut because the agent cannot confidently answer the dimensional and use-case questions the customer actually asked.
This gap is going to widen. As Buy for Me handles more transactions and Rufus handles more of the consideration phase, the traditional keyword-ranking playbook accounts for less of the customer journey. The brands that catch this early and build listings structured for machine readability will have a durable edge. The ones that realize it a year from now will be playing catch-up.
At ALFI, we started auditing client listings for Rufus and agent-layer gaps after early signals in 2025 showed mismatches between keyword ranking and conversion on AI-driven traffic. The pattern is consistent: listings built for crawlers outperform listings built purely for keyword matching when discovery runs through Rufus. If you want to know where your top 10 SKUs stand against this framework, schedule a listing audit.
What you can do about it now
Start with your five highest-revenue ASINs. These are the ones worth fixing first, and the work is transferable to the rest of your catalog.
Structured specs first. Pull up your product detail page and read through the technical specifications section. For every claim a customer might make in a conversational query about your product, there should be a corresponding spec attribute. "Fits in a car cupholder" needs a diameter measurement. "Waterproof" needs an IP rating or explicit waterproof certification. "Works with iPhone 15" needs a compatibility field, not just a mention in the title.
Q&A next. Review the existing questions on your listing. For any question without a brand answer, answer it now. Then look at questions being asked about your top competitors' products in your category. If customers are asking those questions about competitors, they will ask them about you too. Pre-populate answers before the question is asked.
Then mine your reviews. Read your most recent 100 reviews and pull out the phrases customers use when describing how they use your product. The language customers use in reviews is often the same language they will use in a Rufus query. If your reviews say "fits perfectly in my gym bag" and "held up through daily outdoor use," those are attribute signals you can reinforce in your Q&A and specs.
This post is part of a broader look at how AI search is reshaping Amazon visibility. For the strategic context, see our piece on the AEO gap no Amazon agency is talking about.
What is Amazon's Buy for Me feature?
Buy for Me is an AI-powered purchasing tool that lets Amazon's agent complete a purchase on a customer's behalf on a merchant's external website. The customer confirms order details but does not manually browse or check out. It works in conjunction with Shop Direct, which surfaces products from outside Amazon's own catalog when customers search for them.
Did Amazon really block Perplexity from shopping on its platform?
Yes. Amazon sued Perplexity in November 2025, alleging Perplexity's Comet AI agent accessed Amazon's platform without authorization to shop on behalf of users. A federal judge granted Amazon a preliminary injunction on March 9, 2026. As of late March, an appeals court had temporarily allowed Perplexity's agents to continue operating pending a longer review. The case is ongoing.
Does my listing need to be on Shop Direct for Buy for Me to matter to me?
Not necessarily. Even within Amazon's standard catalog, Rufus drives product recommendations and surfaces listings in its shopping assistant interface. Whether or not you are in the Shop Direct feed, the principles are the same: structured specs, complete Q&A, and category attributes are what Rufus reads when a customer asks a conversational question.
What is the difference between building for Rufus versus traditional Amazon SEO?
Traditional Amazon SEO focuses on keyword placement in titles, bullets, and backend search terms to rank in A9 search results. Building for Rufus and Buy for Me focuses on structured data completeness, Q&A content, and spec accuracy so AI agents can confidently answer conversational queries about your product. Both matter. A listing that ranks well for keywords but has thin specs and no Q&A content is increasingly vulnerable as more of the customer journey runs through AI interfaces.
How fast is this shifting?
Faster than most sellers realize. Amazon's Rufus handles millions of daily queries, and Buy for Me is being rapidly expanded through third-party feeds to cover more of the catalog. The sellers building for this now are setting up a structural advantage. Waiting until the signals show up in your data means you are already three to six months behind.
Can I opt out of Buy for Me?
If your products are in Amazon's standard catalog through Seller Central or Vendor Central, your products are already eligible to surface through Rufus recommendations. For Shop Direct, Amazon has indicated merchants can opt out by contacting them directly. But opting out of Rufus-driven discovery is not practical or advisable. The right response is to make your listings readable for the agent, not to hide from it.
What to do this week
- Pull your top five revenue ASINs and audit the technical specification fields for completeness. Flag every attribute that is blank or uses vague language instead of a specific measurement or certification.
- Review the Q&A section on each ASIN. Answer any unanswered questions. Add pre-emptive answers to the top three use-case questions in your category based on competitor Q&A.
- Mine your most recent 100 reviews for the phrases customers use to describe how they use your product. Add those use cases as Q&A answers if they are not already covered.
- Check your product's category attribute fields in Seller Central, particularly the technical details and product features tabs. These fields feed Rufus's structured data layer directly.
- Search for your top product in Rufus or the Amazon app using a conversational query a customer would actually ask. See what comes up and whether your listing answers the question being asked.
- If you are already in or considering Shop Direct, sync your catalog through a supported feed provider (Feedonomics, Salsify, or CEDCommerce) to ensure your real-time pricing and inventory are accurate. An agent making a purchase on a customer's behalf with stale pricing data is a problem.
If you want a full audit of where your listings stand against the agent-layer framework, book a call with ALFI. We will walk you through the gaps in your top SKUs and what to prioritize.
