The Ban That Built Amazon's Bait Shop

Based on the research ofXia, Cai and Gu, "Disintermediation Governance and Complementor Innovation: An Empirical Look at Amazon.com," Management Science, 2026

Xia Han, Gaoyang Cai, and Grace Gu study a change to Amazon's Communication Guideline that banned sellers from sending buyers external website links, phone numbers, or email addresses through Amazon's buyer-seller messaging system. The policy is classic anti-disintermediation governance: platforms lose money every time a transaction that could have happened on-platform happens instead through a side channel the seller opened up. Cut off the side channel, the theory goes, and complementors have no choice but to keep transacting, and innovating, where the platform can see it and take its cut. Using a difference-in-differences design that compares sellers who already had a direct-selling channel against those who didn't, the authors find the opposite. Product innovation on Amazon fell, measured both by the count of new products launched and by how innovative those products were. The rule did not stop sellers from having an exit. It just changed which products get to use it, and which ones stay behind.

The Guideline That Sealed the Chat Window

Amazon's own Communication Guidelines are specific about what the ban covers. Permitted messages, the policy states, may not include "external links unless they are secure working links... necessary for order completion or links to Amazon," and may not include "email addresses" or "telephone numbers except those related to warranties, shipping providers, or manufacturers." A seller who violates this risks having proactive messages restricted to Amazon's own templates, or, in Amazon's words, "a suspension of selling privileges in Amazon stores." This sits inside a broader anti-circumvention stance the company has held for years: its seller code of conduct states plainly that a seller "may not attempt to circumvent the Amazon sales process or divert Amazon customers to another website." The Communication Guideline update the paper studies took that general principle and made it enforceable at the level of a single chat message, closing off the most direct route a seller had to hand a buyer their own contact information.

That specificity is what makes the finding sharp. This isn't a story about vague platform pressure. It's a policy shock with a clean before and after, aimed at a single, well-defined behavior, and it still produced a result opposite to its stated purpose. Complementors with an existing direct-selling channel, the group most able to act on the policy, cut back their Amazon-based innovation the most.

Why Locking the Chat Didn't Lock the Relationship

The mechanism is where the paper gets interesting, because it shows the ban solved the wrong problem. The Communication Guideline governs one channel: the in-app buyer-seller message. It says nothing about a QR code printed on a box, or a "follow us" line on a thank-you card, or a social media handle stitched into the product packaging itself. Guidance for sellers confirms this asymmetry directly: a QR code linking to a seller's social accounts is treated differently from the same link sent in a chat message, and sellers are told they "can, however, request that information outside of Amazon via social media." One industry guide summarizing Amazon's rules for sellers puts the contact-info ban and the packaging allowance side by side without apparent contradiction, because from Amazon's perspective they are different surfaces governed by different rules.

That gap is the whole workaround. A seller no longer needs to type a phone number into a chat window to build a relationship that lives outside Amazon. They just need the buyer to scan a code once, on the box, after the sale is already locked in. The paper's data shows sellers doing exactly this: complementors with direct-selling channels significantly increased the number of new products they launched through those outside channels after the policy hit, the effect the authors label the switching effect. The chat ban didn't need to be defeated. It was simply irrelevant to how the relationship actually gets built once a package is already in a customer's hands.

The rule didn't stop sellers from having an escape hatch. It just changed which products get to use it first.

The Bait Effect: Cheap Stays, Good Leaves

The second finding is the one that should worry anyone running a marketplace. Sellers did not just move products off Amazon at random. They sorted them. The paper documents what it calls the bait effect: complementors strategically keep their less innovative, price-competitive products on Amazon, where they're cheap enough to pull in Amazon's traffic, while diverting their innovative and high-value new products to their own channels before those customers ever see them on Amazon at all (see the exhibit). Amazon's storefront keeps its price-competitive draw. It just stops being where the interesting product shows up.

This is a coherent seller strategy once you see the incentives. A price-competitive, low-innovation item is exactly the kind of thing you want ranked by Amazon's algorithm and bought impulsively by an Amazon shopper: it needs volume, not a relationship. An innovative, higher-margin item is exactly the kind of thing a seller wants to sell direct, where they keep the full margin, the customer data, and the repeat-purchase relationship, and where Amazon's 30-day, order-ID-tagged messaging rules never applied to begin with. The Communication Guideline didn't create this incentive. It just removed sellers' remaining reason to launch the good stuff on Amazon first, since the on-platform relationship-building that used to make that worthwhile was the exact thing the policy shut down.

Who Absorbs the Hit, and Who Doesn't

The moderating results sharpen the picture further. The innovation decline is smaller for sellers with strong Amazon reputation and for sellers who already multihome across platforms. That sounds like good news, evidence the policy mostly bites the sellers with the least to lose. Read the other direction, it says the opposite: the sellers most exposed to a shrinking Amazon innovation pipeline are the ones who are newer, less established, and single-homed, precisely the complementors Amazon's marketplace depends on to keep discovering products it doesn't yet carry. The academic literature on multihoming complements has long shown that a complementor's option to operate elsewhere changes how it invests in quality on any one platform; Cennamo, Ozalp, and Kretschmer's 2018 study of multihoming in platform ecosystems found the same logic in a different setting, that a complement's presence on rival platforms reshapes the quality trade-offs it's willing to make on any single one. Amazon's ban didn't remove that option for sellers. It just made the option more attractive for exactly the sellers sophisticated enough to already have it, while the sellers without a ready alternative had nowhere to put their innovation but nowhere to keep it either.

The deeper lesson is not specific to Amazon or to e-commerce. Any platform governance policy premised on cutting off a channel assumes the channel is the relationship. It usually isn't; the channel is just the fastest way to start one. Sellers on Amazon didn't need Amazon's chat window to keep a customer after the sale. They needed one scan, one follow, one card in the box, and the ban left every one of those open. A platform that wants to keep its complementors' best work has to make staying the better deal, not just make leaving marginally more inconvenient in the one channel it happens to control.