The paradox of borrowed authority

Posted on: 22 May 2026

Everyone is now talking to AI, and a new industry is emerging at a pace that, to someone who has watched advertising cycles for four decades, feels uncannily familiar. It is called Generative Engine Optimization, or Answer Engine Optimization, depending on which agency is selling the package. American enterprise CMOs are spending between five and fifty thousand dollars a month to restructure their websites, rewrite schema markup, optimise textual chunks and calibrate meta-tags so that Claude, ChatGPT, Perplexity and Gemini will cite them when a user asks a question. Ninety-four per cent of marketing directors surveyed say they intend to increase investment in 2026, and ninety-seven per cent report positive impact. The numbers, taken in isolation, tell a clear story of success.

Then you look at another data set and the story falls apart. AirOps analysed over twenty-one thousand brand citations across ChatGPT, Claude and Perplexity, and the result is the kind that should keep anyone signing one of those fifteen-thousand-a-month retainers awake at night. Eighty-five per cent of brand mentions in AI responses originate from third-party sources, while only thirteen per cent come from the brand's own domain. Brands are cited through external sources at a rate 6.5 times higher than from their own websites. Nearly ninety per cent of those third-party citations come from listicles, comparison pieces and review sites, and to give a sense of scale, Reddit alone appears in roughly one AI response out of five, while forty-eight per cent of citations come from community platforms such as Reddit and YouTube.

The picture that emerges is grotesque in its coherence. Companies are spending growing sums to optimise precisely the place where AI systems do not go looking for them.

It is worth pausing here, because the phenomenon is more subtle than simple technical incompetence. The companies signing these contracts are not stupid, and the conversion data on AI traffic is striking: a visit arriving from ChatGPT converts on average at 14.2 per cent, against 2.8 per cent for Google organic traffic, and in certain segments the multiplier reaches twenty-three times. Gartner estimates that traditional search volume will decline by twenty-five per cent by the end of 2026. Anyone investing in AEO is chasing a channel that genuinely performs. The problem is not the decision to enter the game, it is the part of the pitch they have chosen to position themselves on.

There is a structural mechanism here that deserves slow examination, because it does not concern only this particular industry. It concerns an epistemological error that companies have been repeating in different forms for decades, and it is the pattern of self-declared authority.

A few examples may help. A bank that spends millions on an institutional campaign declaring itself trustworthy, while its actual reputation forms on forums where customers describe what really happened when they applied for a mortgage. A university that publishes glossy brochures about the quality of its teaching, while the choices of incoming students are guided by what former students write in private Telegram groups. A company that produces two-hundred-page ESG reports to certify its sustainability, while specialist analysts look elsewhere to understand what it actually does. In every case the pattern is identical: the actor invests significant resources in constructing its own narrative, ignoring or underestimating the fact that real corroboration forms outside its perimeter of control.

AI systems have automated this principle and have made it brutally operational. When an AI must answer a question about a product, a company or a service, it does exactly what an experienced decision-maker would do: it seeks external corroboration before believing the self-declaration, not out of algorithmic malice but for a structural reason that deserves to be named properly. A brand's self-declaration about itself is not falsifiable, which is to say it carries no discriminating information. What others say, particularly when they agree from independent sources, is the kind of evidence that survives a minimum test of reliability. AI systems have rebuilt, inside their retrieval procedures, the basic epistemological intuition that any competent human being would naturally apply: do not trust the one who speaks of himself, look at who speaks of him when he is not in the room.

A series of consequences follows from this, and if taken seriously they should redesign the entire marketing architecture currently emerging around the topic.

The first is that much of the AEO and GEO infrastructure being sold today is theatre of optimisation, not optimisation itself. Rewriting your site to be "AI friendly", adding structured schema markup, restructuring FAQs into extractable chunks, calibrating entities and topic clusters: these are all activities that produce measurable work, regular billing and convincing dashboards, but they operate on the thirteen per cent of citations while leaving the remaining eighty-five per cent entirely uncovered. It is like obsessively polishing the windows of your house while the garden, where visitors actually stop, is left abandoned.

The second is that the real game of authority in the AI era looks very much like what serious people in communications have always known. It is called earned media. It is built over time, through coverage in independent publications, presence in independent comparisons, authentic reviews, citations in industry research, real conversations on communities frequented by real users. It is precisely the kind of activity that traditional marketing director metrics have always struggled to quantify, because it does not produce dashboards, it cannot be attributed to a specific campaign, and it cannot be bought by the linear metre.

There is a scene, recounted to me by the head of communications at a large European company, that captures the point in an almost moving way. He had spent years defending in board meetings the importance of qualified editorial coverage, of relationships with specialist journalists, of the patient construction of credibility with third-party sources. Every year his budget was cut in favour of more "measurable", more "optimised", more "ROI-driven" investments. He had effectively lost that battle. Now, however, as the AI numbers begin to confirm empirically that this work was exactly the right thing to do, he must sit through the same board voting five-figure monthly investments in AEO, designed to optimise their website to be "read better by AI systems", without anyone realising that the machine is looking for precisely the kind of coverage that had been cut in previous years.

The third consequence is the one that, to someone who has watched marketing move from press advertising to display banners, from display banners to SEO, from SEO to paid social, from social to influencer marketing, sounds familiar in a specific way. Every technological transition produces, in its early years, a service industry that sells the illusion of optimising the old system for the new one. It is a transitional mechanism that serves a function: it gives companies time to adjust to change without having to confront its deeper implications immediately. The problem is that, once the transition phase ends, the new system reveals itself to operate on completely different logic, and those companies that spent the early years optimising the optimisable find themselves behind the ones that, from the beginning, understood that the real game was elsewhere.

Classical SEO worked for a very specific reason, which is that Google was a relatively unintelligent machine reading surface signals such as backlinks, keyword density and internal site structure. The game was mechanical enough to support an industry of technical optimisation. AI systems, by contrast, read, comprehend, evaluate and corroborate. They are not machines stupider than Google, they are machines that embody an operational approximation of qualified human judgement. And qualified human judgement, always and everywhere, gives more weight to what independent third parties say than to what the interested party declares.

The structurally most interesting aspect is that this creates a long-term asymmetry. Companies that built real authority through decades of genuine work, contributing to define their own category through original research, feeding their sector with substance, letting others write about them on the basis of actual merit, now start with a structural advantage that cannot be bought with any monthly retainer. Those that built their positioning through paid amplification of self-produced messages are discovering that AI systems, quite simply, do not believe what they say.

For the second group, the remedy is not AEO. The remedy would be to rebuild ten years of legitimacy they have never genuinely accumulated. That requires time, real work, and above all a patience that the quarterly corporate reporting cycle does not easily afford.

This is where the real mechanism lies, the one that should interest anyone looking at the matter with clarity. AI systems are doing, inside the digital ecosystem, the same work that activist funds do inside listed companies. They are separating, at speed exceeding what the market previously allowed, signal from noise. They are rewarding those who had accumulated something real and penalising those who had accumulated narrative. They are doing it in automated fashion, at scale, and with a brutal transparency the advertising market had never known.

So to the initial question, whether it makes sense to build websites more for bots than for humans, the clinical answer is that the question is poorly framed. It is not a matter of choosing between humans and bots. It is a matter of understanding that even the most sophisticated bots, when they have to say something about you, go and ask the others. And if the others have never had reason to talk about you, no amount of optimised chunking will change that.

A question remains, one that deserves to be asked without diplomacy, because the piece is worth nothing if I do not ask it. For decades marketing has sold itself as the discipline that builds brand authority. For decades it has struggled structurally to demonstrate this, and has defended itself by arguing that authority is intangible, that it requires time, that short-term metrics do not capture it. That was also true. But AI systems have now done something that changes the terms of the comparison. They have automated an external judgement, at industrial scale, that separates with clinical precision the brand others cite from the brand that only talks about itself. They have made falsifiable, for the first time, the difference between accumulated authority and declared authority.

Months ago, on these pages, I argued that AI is not taking away jobs but is making evident which jobs were only pretending to exist. The middle management whose role was to pass information between hierarchical levels. The bureaucracy whose role was to justify its own existence. The organisational theatre that produced dashboards instead of value. These were jobs that, in retrospect, produced less than they claimed. AI has not killed them, it has only removed the cover that protected them.

Marketing now finds itself in exactly that same condition, and much of the industry has not yet realised it is there. The question is not what to do with AEO. The question is how much of the work done, billed, presented in board rooms, awarded at festivals over the last twenty years would have passed the test that AI systems apply today. How much produced autonomous external corroboration, and how much was instead paid amplification of self-produced messages that nobody else would ever have repeated if we had not paid them to do so. And perhaps, going deeper still, it is worth asking whether the very metrics by which that work was evaluated were not, from the beginning, the symptom of the problem rather than its solution.

AI systems are now presenting that bill. It is arriving, and it will arrive over the next two or three budget cycles. At that point every marketer will have to do what until now nobody had the courage to make them do: open a blank page, write their own name at the top, and underneath answer honestly a single question. Has what I did over the last twenty years built real authority for someone, or has it built presentations that no independent external system would ever have validated? Was it real work, or just performance pretending to be work?

The AI will not answer this question. Each will answer it, in private, in front of himself or herself, and it will be one of the most difficult conversations the profession has ever had to have.