A website becoming enshittified is not an aesthetic problem: it is the conversion of an asset into a liability. A degraded corporate page is not neutral, because every day it stays online it destroys value \u2014it drives away customers with genuine purchase intent, erodes the trust that took years to build, and exposes the company to legal and reputational risks that no small or mid-sized business can afford to absorb. The term, coined by Cory Doctorow to describe the cycle in which platforms first treat their users well, then exploit them to favour their business customers, and finally capture all the value for themselves once the forces that used to discipline them \u2014competition, regulation, users’ ability to modify technology, and worker power\u2014 disappear, describes a pattern that is no longer confined to the big platforms. Generative artificial intelligence did not invent enshittification, but it has accelerated and industrialised it: it has made producing junk cheaper, automated manipulation, and begun to hijack the traffic that millions of websites depend on. This article starts from that relationship \u2014AI and enshittification\u2014 and traces what it truly costs your business.<\/p>\n
The first tangible effect of degradation is financial, and it happens long before any sales conversation takes place. The professional buyer now completes most of their research independently, so the website becomes the main salesperson during the most decisive phase of the process. If that website is saturated with artificial friction, invasive forms, pop-ups and design traps, the sale is lost before it begins. And unlike a monopoly platform, which can mistreat captive users because they have no alternative, a conventional B2B company enjoys no such shield: its prospective customer can leave with a single click \u2014and does.<\/p>\n
On top of that voluntary exodus there is now a structural drain caused by AI itself. For two decades, the web’s implicit bargain was simple: search engines indexed your content and, in return, sent you visitors. Generative answers break that contract, because they resolve the query on the results page itself and retain the user without sending them anywhere. The available research is stark: an analysis reported by Ars Technica<\/a> concludes that the presence of AI summaries cuts clicks to websites almost in half. For a business that depends on organic traffic as the entry point of its funnel, this is not a nuisance: it is the amputation of its main acquisition channel, and it turns every historical SEO investment into an asset whose return evaporates.<\/p>\n The cost does not end with the lost sale. Replacing a customer who leaves out of distrust is far more expensive than retaining them, because the direct loss is compounded by the cost of acquiring replacements and the erosion of customer lifetime value. Enshittification also compresses margins cumulatively: every dark pattern that forces a short-term conversion generates cancellations, returns and negative reputation that make all future acquisition more expensive. What looks like conversion optimisation is, in reality, a mortgage on tomorrow’s revenue. There is also an opportunity cost that is rarely accounted for: while a company devotes resources to squeezing every visit, it is not devoting them to solving its customer’s real problem, and that shift of investment from value creation toward value extraction is precisely what, at platform scale, ends up hollowing out the service.<\/p>\n The extractive logic that degrades the experience almost always degrades accessibility too. The very elements that characterise an enshittified website \u2014interstitials that trap you, poor contrasts designed to hide the reject button, visual hierarchies built to confuse, content that loads erratically\u2014 are exactly the ones that shut out people who browse with a screen reader, with a keyboard, or with contrast needs. Exclusion is rarely a deliberate decision; it is collateral damage of prioritising extraction over clarity. But the result for the excluded user is identical: they cannot complete the task, and they leave.<\/p>\n Here it is worth recalling a technical principle that many managers forget: a search engine’s crawler is, to all intents and purposes, a blind user. It perceives the page just as a screen reader does \u2014through semantic structure, alternative text and labels, not visual artifice\u2014 so a website that is poorly structured for accessibility is also poorly structured for ranking. With the arrival of generative models, that principle is amplified: AI systems that synthesise answers extract and cite clear, structured, verifiable content, and discard pages saturated with friction and artifice. Bad accessibility, which once penalised SEO, now also penalises citability by AI.<\/p>\n The cost of this exclusion is an invisible but quantifiable market. A significant share of the population lives with some form of disability, and to that we must add the temporary or situational circumstances \u2014a slow connection, a screen in the sun, a busy hand\u2014 that mean accessibility actually benefits everyone. Every barrier a website adds in the name of extraction is a customer segment discarded in advance. In business terms, accessibility is not a compliance expense: it is the difference between addressing the whole market or voluntarily giving up part of it.<\/p>\n For years, dark patterns inhabited a grey area: annoying, but rarely sanctioned. That time is over. Regulators around the world now treat deceptive design as an actionable unfair practice, and they have begun to coordinate to document and sanction it. The Federal Trade Commission, together with international networks of consumer and privacy authorities<\/a> has published reviews confirming the widespread use of dark patterns in subscription services and privacy management, and has warned that these practices breach consumer protection law. What was a conversion tactic yesterday is today a legal risk with a name and a case file.<\/p>\n AI has opened a new front on this terrain. Dark patterns are no longer aimed solely at deceiving people, but also at manipulating the AI systems that mediate between the company and the customer. Tactics have emerged such as recommendation poisoning \u2014hidden instructions on a page so that an assistant prioritises the products of whoever inserts them\u2014 or the mass creation of fake accounts to inflate the mentions that models pick up as a signal of authority. It is the same old extractive logic, but automated and aimed at the machine; and it is just as actionable, because deceiving the algorithmic intermediary to mislead the consumer fits squarely within the definition of unfair practice that regulators already monitor.<\/p>\n The danger is profoundly asymmetric. A large platform can afford to absorb a fine, litigate for years and dilute reputational damage across millions of captive users. An SME cannot. It inherits the same regulatory framework and the same scrutiny, but without the financial capacity to take the sanction or the user base that would let it survive a loss of trust. Copying the big platforms’ patterns \u2014because they “work”\u2014 means, for a small or mid-sized company, importing a risk it cannot afford and giving up the one asset that sets it apart: trust.<\/p>\n Reputation works like compound interest: it accumulates slowly, over years of honest interactions, and is destroyed in an instant. Every dark pattern, every piece of data used without clear consent, every obstacle placed to retain someone who wants to leave, is a withdrawal from that trust account. And unlike money, lost trust is not recovered with a campaign: it requires proving consistent behaviour again, over a long time. In an environment where the user distrusts by default, honest design has stopped being a soft virtue and become a measurable competitive advantage.<\/p>\n To that erosion is added a new phenomenon that aggravates the whole web’s crisis of trust: so-called AI slop, the flood of mass-generated AI content with no verification or real value. The problem is not marginal: WIRED has documented cases in which the search engine ranked AI-generated spam above original reporting<\/a>, burying the signal under the noise. For an honest company, this has a double consequence: on one hand, its quality content competes against an unlimited volume of cheap imitations; on the other, the user’s generalised distrust of anything that looks automated also contaminates the perception of its brand. In an ecosystem saturated with junk, appearing human and trustworthy becomes a scarce differentiator.<\/p>\n AI’s own mediation introduces an additional layer of distrust. When a model summarises, recommends or cites, it rarely explains why, and far too often it gets it wrong: a Tow Center study reported by Nieman Lab<\/a> found that AI search engines failed to produce correct citations in more than 60% of tests. For a company, this means its reputation can end up in the hands of an opaque intermediary that describes it badly, attributes it to the wrong source or mixes it with third-party content with no possible control. The only structural defence is to build a presence so clear, verifiable and consistent that it is hard to misrepresent \u2014for a human and for a machine alike.<\/p>\n It is worth looking one step further, because AI is not only an agent that accelerates the web’s enshittification: it is a candidate to suffer its own cycle of degradation. Today, to a large extent, AI tools are in their friendly phase, the “good for users” one, subsidised by capital that seeks growth before profitability. But that phase is, by design, transitory. The pressure to recoup colossal investments will arrive, and with it the temptation to squeeze a user base that, by then, will already be locked in.<\/p>\n The pattern is predictable because we have seen it before. A WIRED analysis of whether AI can escape the enshittification trap<\/a> clearly describes the likely sequence: once users are hooked, recommendations may start to tilt toward whoever pays for visibility, prices may rise to maintain the same answer quality, and users’ conversations may end up feeding the training of new models simply because companies can do it. The high cost of developing cutting-edge models also guarantees that only a few companies will dominate the space, which multiplies dependence and reduces the user’s room to leave.<\/p>\n For a company, the lesson is strategic: building the entire business on a third-party AI platform reproduces the same vulnerability that led so many publishers to depend on a single search engine’s traffic. When the friendly phase ends, the rules will change unilaterally, and whoever has handed their customer relationship to the intermediary will discover they no longer control their own channel. The defence is to use AI as a tool without handing it the essential asset \u2014the direct relationship and the customer’s trust\u2014 so that a change in the platform’s terms does not amount to losing the business.<\/p>\n The good news is that the defence against all of the above coincides with what ranking already rewards. Search engines have spent years rewarding useful, reliable, people-first content, and penalising precisely the practices that characterise the degraded web. Designing against enshittification is, therefore, designing for SEO. Reference documentation on ethical design, such as the material compiled by Smashing Magazine<\/a>, shows that decisions that respect the user \u2014clarity, no traps, honest hierarchy, real accessibility\u2014 tend to align with those that improve organic performance. There is no need to choose between ethics and visibility: they are, increasingly, the same thing.<\/p>\n In the era of generative engines and AI optimisation (what is already called GEO)<\/a>, that alignment becomes even more decisive. Models that synthesise answers extract and cite clear, structured, verifiable content, and the temptation to manipulate them with the tactics described earlier \u2014recommendation poisoning, mention farms\u2014 is a shortcut that regulators and the platforms themselves already pursue and penalise. The sustainable path is the opposite: structure content in clear semantic blocks, answer explicit questions, keep information honest and verifiable, and make the website itself easy to understand for a person and for a model alike. Being citable by an AI and respecting the user are not competing goals, but the same job done well.<\/p>\n Users’ growing rejection and the nostalgia for a smaller, community-based, honest internet are not sentimentality without commercial weight, but a market signal that shows where value is shifting. And the underlying thesis is hopeful: enshittification is reversible, because it is the consequence of decisions, not of immutable laws. At platform scale, the solution runs through restoring competition, regulation, interoperability and worker power; at the scale of a corporate website, the logic is identical and far more actionable, because every company can decide, today, to stop applying the tactics that degrade the experience and thus recover the trust the rest of the ecosystem is throwing overboard. Every form, every cookie banner and every countdown timer is a vote for or against that degradation, and refusing to take part in it has stopped being an ethical stance to become the most profitable strategic decision in the medium and long term. For companies looking to build or recover a website that treats customers as people \u2014and that, precisely for that reason, performs better before search engines and AI\u2014 adopting a B2B marketing approach built on honesty, clarity and strategy<\/a> is now a competitive advantage ahead of extraction.<\/p>\n<\/div><\/div><\/div>The real cost of distrust: acquisition, retention and customer lifetime value<\/h3>\n
Exclusion consequence: accessibility as market amputation<\/h2>\n
When degrading the experience degrades accessibility<\/h3>\n
The crawler (and now the AI) as a “blind user”: bad accessibility is bad visibility<\/h3>\n
The invisible market: how much business is lost by excluding<\/h3>\n
Regulatory consequence: from deceptive design \u2014human and algorithmic\u2014 to sanction<\/h2>\n
From grey area to regulatory target: the authorities’ shift<\/h3>\n
Dark patterns aimed at machines: manipulating AI and “poisoning” recommendations<\/h3>\n
Asymmetric risk: why the SME inherits the danger without the shield<\/h3>\n
Reputational consequence: trust capital in the age of “AI slop”<\/h2>\n
Trust as compound interest<\/h3>\n
AI slop: the AI-generated noise that buries the signal<\/h3>\n
AI without explainability or reliable citations: the underlying distrust<\/h3>\n
The enshittification of AI itself: the pattern repeats<\/h2>\n
The “good for users” phase: why it won’t last under profitability pressure<\/h3>\n
Lock-in and extraction: paid recommendations, price hikes and use of your data<\/h3>\n
What it means for a company that builds on AI platforms<\/h3>\n
SEO\/GEO: the coherence between respecting the user and being visible to humans and AI<\/h2>\n
People-first content: designing against enshittification is designing for SEO<\/h3>\n
Being citable by an AI without resorting to manipulation: legibility, honesty and structure<\/h3>\n
Conclusion: restoring trust is the most profitable decision<\/h2>\n