Can a Company Stop an Intermediary From Taking Over Its Customer Relationship?

The wrong variable

The intuitive answer is that leverage comes from scarcity. If what you sell is hard to replace, you can dictate terms. If it is easy to replace, you cannot.

Southwest ruins that answer.

A seat from Dallas to Chicago is about as replaceable as a product gets. Four other carriers fly it. Southwest nonetheless sued Kiwi.com for scraping and reselling its fares, won a preliminary injunction in September 2021 on a straightforward breach of its site terms, and obtained a final judgment and permanent injunction on 28 December 2021 barring Kiwi from scraping its site, publishing its fare information anywhere, or selling its flights. Kiwi dropped its appeal on 3 January 2022. That is about as complete a win as this history contains.

What Southwest had was not a better seat. It was 81 percent of 2024 passenger revenue arriving through Southwest.com and the Southwest app, and 82 percent the year before. Fifty years of customers who already knew where to go. Site terms that Kiwi had to accept in order to operate. And a customer for whom the intermediary was not solving much.

We know that last part because Southwest eventually tested it. In 2024 it listed on Google Flights, Kayak and Skyscanner. On 26 February 2025 it went live across Expedia, Travelocity, Hotwire, Orbitz and CheapTickets, the first online travel agency distribution in its history. In March, CEO Bob Jordan told a financial conference that the company saw no share shift to Southwest from having better policies. Almost nobody else in this history ran the experiment, and the one company that did found the intermediary was worth close to nothing.

Put American and Southwest beside each other and the supply story collapses. The airline that controlled every unit of its own supply lost the channel. The airline selling the more substitutable product kept the intermediary out.

Three other questions do the work instead.

Can you control the way in?

Owning the product is not a control point. It feels like one, which is why companies keep spending money on it.

In 2008 and 2009 the film studios tried to keep new releases away from Redbox's dollar kiosks. Universal wanted a 45 day delay, Fox 30, Warner 28. They instructed wholesalers to withhold discs. Redbox responded by sending staff into Walmart, Best Buy and Target to buy the discs at retail, and the stores started limiting purchases to as few as three copies per customer. The studios owned the copyright on every film in the kiosk and could not keep a single one out of it, because the first sale doctrine meant anyone could buy a disc and rent it.

What worked in this history was narrower and more procedural than ownership.

Sometimes it was an agreement the intermediary had to accept to operate. LinkedIn spent five years losing the headline question against hiQ, twice, and then won in November 2022 on the plain fact that hiQ had accepted the user agreement when it opened a corporate account. Southwest won the same way.

Sometimes it was refusal made explicit and then ignored. Craigslist never required 3taps to log in. 3taps was reading public pages. What Craigslist did was send a specific cease and desist and block the IP addresses, and when 3taps routed around both, a court allowed the claim to proceed in August 2013 on the basis that authorization had been withdrawn and 3taps knew it. Craigslist lost its copyright claim over the listings themselves. Its strongest surviving claim rested on the fact that it had said no in a way that could be proved.

Sometimes it was a statutory right in the thing itself, which is rare and decisive. Aereo lost at the Supreme Court on 25 June 2014 and filed for bankruptcy that November. When a company holds an exclusive right by statute, blocking works quickly and completely. Almost no one holds one.

And sometimes it was operational. Ryanair spent years suing online travel agencies and got a jury verdict against Booking.com in July 2024, which the trial judge then set aside in January 2025 on the ground that Ryanair had not proved the required loss. Seven months later, on 26 August 2025, Booking Holdings joined Ryanair's approved agency programme. Ryanair got the customer record and the notification channel: passengers booking through those sites now reach their myRyanair accounts without the extra verification step, and receive flight updates from Ryanair directly.

Read that sequence carefully. The legal position had already collapsed. What Ryanair still held was a verification gate that made the unapproved path worse for the passenger, and that appears to be the leverage the eventual deal was built around.

There is a pattern worth naming here. Technical blocks rarely worked as barriers. They were much more useful as evidence. Kiwi obscured its data source. Ticket brokers ran thousands of accounts behind proxies and SIM banks. Perplexity pushed an update circumventing Amazon's block within a day. The block was still worth doing in every one of those cases, because it was the thing that put refusal on the record and made the circumvention provable.

Can customers already reach you?

Direct demand is what makes refusal affordable, and it has to exist before the intermediary arrives.

Southwest could turn down a distribution channel for five decades because most of its customers were already coming to it. Independent hotels could not turn one down for a season. By 2025 the online travel agencies were taking 63.4 percent of independent hotel bookings. The European Commission designated Booking.com a gatekeeper on 13 May 2024, and on 2 December 2024 the company waived the rate parity clauses that had stopped hotels undercutting it on their own sites. Germany's competition authority had banned narrow parity in December 2015. The relief arrived somewhere between ten and twenty years after the channel had already formed, at which point the freedom to price was worth far less than the dependency it had been traded against.

Restaurants had it worse. From 2018 the delivery platforms listed restaurants they had no agreement with, took orders, dispatched couriers, and frequently showed stale menus and wrong prices, and the customer blamed the restaurant. What individual restaurants got from litigating individually is instructive: Grubhub's proposed 2021 class settlement would have covered roughly 150,000 restaurants listed without consent, paid them nothing, and required them to claim their own listing before they could ask to be removed.

There is no threshold in this evidence. One clean case at 81 percent does not tell you where the line sits. What it does tell you is that the number is worth knowing before you need it, because once the intermediary is capturing first touch the figure stops measuring what you want it to measure.

If you vanish, who do they see instead?

This is probably the most important of the three, and it is the one companies think about least.

Craigslist's listings existed nowhere else. Aereo's broadcast signal had no substitute. Both intermediaries died. Every discovery intermediary in this history that became durable could keep the customer's decision going when one supplier disappeared.

That is what Booking.com does. It shows the whole city. A corporate booking tool shows every carrier inside the travel policy. A delivery app shows every restaurant on the block. When a supplier withdraws from one of these, the customer's decision continues without interruption, and the only party inconvenienced is the supplier.

Why waiting costs so much

Every company in this history that successfully blocked an intermediary challenged it early. Craigslist sued within months. LinkedIn's cease and desist went out immediately. Southwest sued in 2021. Amazon wrote to Perplexity before its browser shipped.

None of the companies that let an intermediary become an established part of how customers work later removed it. Not the hotels, not the restaurants, not the banks, not American Airlines against a distribution system the airlines themselves had built and spun off.

Two years is not a law, and the mechanism matters more than the interval. Delay changes the bargaining position because all three conditions decay at once. Control points erode as the intermediary finds another path in. Direct reach erodes as the intermediary takes first touch. The substitute set grows as it signs your competitors. And the customer builds a habit, which is the part that does not reverse.

Losing the fight and still getting paid

Failing to block is not the same as losing everything, and the negotiated outcomes in this history are often better than the fights.

The studios could not keep a disc out of a Redbox kiosk. Lionsgate broke ranks first in August 2009 with a five year supply deal at street date, reportedly worth around $158 million. Warner settled in February 2010 for an estimated $124 million over two years, and got the 28 day window it had been trying to impose by force, plus a clause requiring Redbox to destroy used discs instead of reselling them. The studios sold access at a price that included the thing they actually wanted.

Ryanair is the same shape. So, eventually, are the banks. Financial data aggregators built their businesses on bank data they took for free for roughly twenty five years. In July 2025 JPMorgan sent pricing sheets to those aggregators, saying it received nearly two billion monthly requests for customer data from intermediaries and that more than 90 percent were unrelated to a customer actually using a fintech service. Reporting put the initial figure to Plaid at around $300 million a year, more than three quarters of its 2024 revenue, though the deal the two firms struck in September 2025 was reported at fractions of a cent per data pull and the terms are not public.

Note what JPMorgan recovered and what it did not. It will get paid. The customer still starts inside the fintech app.

Restaurants are the cleaner version. California's Fair Food Delivery Act was signed in September 2020 and took effect on 1 January 2021, requiring consent before a platform can list a restaurant. Chicago settled with Uber for $10 million. The Federal Trade Commission and the Illinois Attorney General settled with Grubhub for $25 million on terms including consent based listing. Restaurants won. And the customer still opens the delivery app when deciding what to eat.

This is the most useful thing in the whole history. Money is easier to recover than position. Economics, consent rights, customer records, notification, timing windows: all of these have been clawed back years after the fact, through courts, regulators and legislatures. Nobody in this record ever clawed back the place where the customer makes the decision.

What changed

Someone new at the door

Every intermediary discussed so far was acting for itself. 3taps, hiQ, Kiwi, Booking.com and the ticket brokers were all principals, and the defence that worked against them reduces to one sentence: you are not allowed in here.

Amazon said that sentence to Perplexity. It told the company before the Comet browser shipped that its AI products were not authorized to interact with the Amazon store. Perplexity declined to identify its assistant in a way that would let Amazon block it. Amazon sent a cease and desist on 31 October 2025, Perplexity published a post titled "Bullying is not innovation," and Amazon sued on 4 November. On 9 March 2026 a district judge granted a preliminary injunction covering Amazon's password protected pages, finding that the browser had the user's permission but not Amazon's authorization.

On 4 August 2026 the Ninth Circuit vacated it, and the reasoning is the part that matters. The panel held that on the record before it, Perplexity does not access Amazon's computers at all within the meaning of the Computer Fraud and Abuse Act. The user accesses Amazon, using the assistant as a tool. The court described the assistant as software in service of the person running it rather than a person in the statute's terms, and applied the same reasoning to the California equivalent.

Read carefully, because the ruling is narrower than the headlines. The panel limited its holding to the meaning of access under that statute, on those facts, and said explicitly that it was not establishing a legal regime for agentic AI. It is one circuit. The case has been sent back. And it says nothing about contract, which is the theory that actually decided Craigslist, LinkedIn and Southwest, and which Amazon did not lead with.

Still, something real has changed. The question is no longer only whether the intermediary is allowed in. It is whether the party coming in is the intermediary at all, or your own customer holding a tool.

What this suggests about agents

Most of the corporate attention on AI agents right now is on the transaction. Who completes checkout, who holds the payment credential, who is merchant of record, what the take rate is.

This history suggests that is the second question.

OpenAI launched Instant Checkout inside ChatGPT in September 2025 and scaled it back in March 2026, telling merchants they could use their own checkout experiences while it focused on product discovery. Roughly thirty merchants were live at the point of retreat. Walmart said it would route its own assistant into ChatGPT rather than transact inside it. Merchants kept the transaction, which is the outcome the industry wanted.

They kept the part that has historically been recoverable and conceded the part that has not. If the customer's shopping decision moves into a conversation with an assistant, the merchant is in the position of the hotel on Booking.com and the restaurant on DoorDash, whatever the checkout arrangement says. Both of those groups eventually got legal relief on economics and consent. Neither got the customer back.

Amazon appears to understand this. Its most effective move against agents so far was not the lawsuit. It was updating robots.txt to keep its catalogue out of ChatGPT's shopping results, which took no litigation and worked immediately.

The four questions

Nothing in fourteen cases proves a law, and the honest reading of this history is a pattern rather than a rule. But the pattern is consistent enough to act on, and it reduces to four questions worth asking before deciding how to respond to an intermediary that has started transacting with your customers without asking.

Can you control the way in, through accepted terms, refusal that has been made explicit and then ignored, a statutory right, or an operational gate inside the transaction that the intermediary needs in order to serve the customer properly? Owning the product is not one of these.

Can your customers already reach you without it, and do you know that number today rather than after the fact?

If you disappear from it, can it simply show them someone else?

And are you honestly better off with it there, because if you are, the resistance will be theatre.