IBM holds the top spot on the Spanish SERP for almost every critical search about AI. "10 dangers", "risks", "disadvantages", "pros and cons", "ethics", "bias", "privacy". Type any variant and there, in the first link, is IBM. It's been doing this for years. The problem isn't that it writes badly — it writes correctly, soberly, professionally, almost always with a blue illustrative image and a button at the end. The problem is that in 2026 IBM is not in the race. The technical frontier is set by others, and by different others. Yet IBM still rules Google. By inertia, not by relevance.
Today we're talking about intellectual authority. About why we grant it to people and companies that probably no longer deserve it. And about one emblematic case — IBM's — to understand the mechanism.
The trick of inherited authority
When you read something signed by IBM, your brain runs a silent operation you'd struggle to describe if you had to put it in writing. It recognises the name. It associates the name with "big company, technological, serious, founded 1911". It concludes, without thinking, that what that company says carries more weight than what someone you don't know says. You assign it authority. Not for what it's saying now, but for what that company did at some point between 1950 and 2010.
It's a useful heuristic in many contexts. If IBM publishes a guide on mainframe architecture, it's probably reliable: it's been making mainframes for sixty years. If IBM publishes a manual on relational databases, the same: it helped invent them. Domain authority in mainframes and relational databases is, up to a point, earned.
What fails is when that authority transfers, almost automatically, to a new domain where the company is no longer a protagonist. And here's the case. IBM, in 2026, is not a relevant player in generative AI. Not for lack of trying — Watson has had a decade of marketing campaigns — but for lack of results. Its models don't show up in the rankings that matter. Its watsonx product line is peripheral, marginal in public evaluations, practically invisible on Chatbot Arena, the platform where users vote blind on which model answers best. The models that are actually at the top of the Arena in 2026 are published by OpenAI, Anthropic, Google DeepMind, xAI and the Chinese labs like DeepSeek and Alibaba. IBM doesn't compete on that list. It isn't even in the conversation.
And yet, when a Spanish-speaking user searches for "risks of AI", the first link leads to IBM. And there, with inherited authority in their head, they read what a company that lost the technical battle years ago has decided to tell them about the sector it no longer dominates.
Watson, the case study
It's worth looking slowly at the Watson case, because it's the cleanest example of the mechanism. Watson was born in 2011 as the system that won the game show Jeopardy! against two human champions. It was, in its moment, a legitimate milestone. It demonstrated natural-language processing under real competition conditions. IBM turned that moment into a corporate banner.
Then came years of escalating promises. Watson was going to cure cancer. Watson was going to transform the financial sector. Watson was going to optimise the supply chain. Watson was going to personalise education. Each announcement came with intensive advertising campaigns, conference presence, carefully narrated case studies. And, in parallel, a series of quiet withdrawals. MD Anderson, the famous Houston cancer centre, cancelled its Watson project in 2017 after spending 62 million dollars and getting no clinical use. Hospitals in other countries did the same. The "Watson Health" line was sold in 2022 to a private-equity firm for a fraction of what IBM had invested. The operation, in corporate terms, was a failure. The market promise was cashed in for a decade in advertising without the product backing it.
When the generative moment arrived — GPT-3 in 2020, ChatGPT in 2022 — IBM was already out of position. The company launched watsonx as a response, a platform offering access to models — some its own, called Granite; many open source brought in from outside — and to enterprise deployment tools. It's a correct product for specific corporate niches. It's not a frontier competitor. It doesn't appear in serious technical discussions about the state of the art. No independent researcher cites Granite as a reference.
And yet IBM is still Google's first result when someone searches for criticism of AI. The reason isn't technical. It's about domain.
How a domain holds when it no longer deserves to
Here comes the least discussed part of the SEO phenomenon. Google doesn't rank by current technical relevance. It ranks by a complex mix of signals — historical domain authority, inbound link profile, content age, user behaviour metrics, editorial freshness and dozens of other factors. A domain that's been ranked for twenty years, with thousands of inbound links accumulated over those twenty years and encyclopedic content published between 2018 and 2023, starts with an enormous structural advantage over any new domain, no matter how much better the new one's content is.
IBM.com is a domain with very high authority. It's been active for decades. It has millions of inbound links from universities, press, technical blogs, academic papers — links earned, in many cases, back when IBM really was the frontier. Its informational content about AI is published, mostly, between 2020 and 2024, with cosmetic updates after that. To Google, that content carries every quality seal. To the sector, it's conceptually outdated and disconnected from what's happening today.
The result is perverse. A company that no longer competes technically occupies the informational shop window. The user who arrives grants it inherited authority. And the cycle feeds itself: because the user arrives, Google keeps ranking; because Google keeps ranking, more users arrive; because more users arrive, the behaviour metrics stay acceptable and the ranking holds. Domain authority has become a rent. A rent the reader pays in time and in judgement formed out of obsolete information.
What you miss while reading IBM
While a Spanish-speaking reader spends four minutes on IBM's listicle about the ten dangers of AI, what do they not read? Here's the operational question.
They don't read, for instance, Stanford HAI's AI Index Report 2026, published in April, with aggregate figures for the sector, analysis of global adoption, data on the geographic concentration of frontier models, evolution of capabilities, documented incidents. It's the most complete free reference document on the state of AI in existence, several hundred pages of aggregated data. It's not translated into Spanish. It has no SEO team. It doesn't compete with IBM on the SERP.
They don't read Ben Thompson at Stratechery, who publishes deep strategic analysis of the sector every week. Casey Newton at Platformer, who covers the big firms' internal decisions and their consequences. The long reports at 404 Media on the dark side of model training. The investigations at MIT Technology Review into precarious annotators in Kenya, into military uses of AI, into systemic failures.
They don't read Kate Crawford, who in Atlas of AI takes apart the metaphor of the cloud and shows the real materiality of the infrastructure. Shoshana Zuboff, who in Surveillance Capitalism explains the business model of which generative AI is just the latest chapter. Timnit Gebru, who in the Stochastic Parrots paper anticipated many of the critiques that later reached the mainstream.
And they don't read, simply, the frontier players themselves explaining themselves. Sam Altman's annual letters. Dario Amodei's essays. Anthropic's technical papers on interpretability. DeepMind's publications on safety. Meta AI's technical blog. They write them with a party's bias — they have it — but at least they're inside the sector they claim to explain. IBM, in 2026, is not.
The pattern enlarged: the fauna of the displaced
IBM isn't the only case. The pattern repeats with several big brands of the earlier tech sector that keep their Spanish SEO domain without holding current technical relevance in generative AI.
Oracle occupies high positions in searches about enterprise AI without being a frontier player. SAP the same. Slack publishes posts about AI productivity that rank higher than serious sector analysis. Avast and Malwarebytes occupy the "AI risks" SERP from the cybersecurity angle. Cisco writes about AI and networks. Dell publishes white papers that rank on general queries. All these companies share a common pattern: they're veterans of the tech sector, they have domains with accumulated historical authority, they keep content-marketing teams of industrial size, and they produce informational articles about AI even though their real contribution to the state of the art is marginal or nil.
None of them lies. Their articles are, technically, correct. The problem isn't the lie. The problem is the selection. Each one picks the angles that connect with its product lines, ignores the ones that don't, and presents what's left as if it were the whole conversation. A reader who takes that set as their main source ends up with a picture of the sector that resembles a coordinated corporate shop window more than an honest representation of what's happening.
A word about generosity
It's worth stopping for a moment to qualify, because this isn't about piling on. IBM has done invaluable things in its history. It invented key concepts of modern computing. It funded basic research for decades. It still employs thousands of excellent engineers. Its quantum-computing line is one of the few serious industrial bets in the sector. Its contribution to open standards is real. None of that disappears because today it isn't a protagonist in generative AI.
What disappears is the automatic equivalence between having been a leader in computing in the late twentieth century and being a legitimate authority on what happens in AI in 2026. Technical authority expires when the player stops competing. The name doesn't. The name keeps ranking.
The criticism isn't against IBM as a company. It's against the reader who confuses name with currency. And, above all, against the mechanism — Google, the SEO ecosystem, corporate content teams — that keeps that confusion alive because it suits every player in the mechanism.
What the sector looks like when IBM isn't in it
If you take IBM out of the conversation for ten minutes and look at the sector with the players who are actually building it, what you see is another map. You see a handful of technical-frontier labs — OpenAI, Anthropic, Google DeepMind, xAI, DeepSeek — with different philosophical bets and models in production that millions of people use every day, plus players of varying weight depending on the moment, like Meta or the Chinese labs at Alibaba. You see a layer of compute providers — Nvidia, AMD, Intel, TSMC, ASML — on which the whole frontier depends and which is, geopolitically, the real bottleneck. You see a handful of application companies — Cursor, Replit, Glean, Perplexity, Harvey — building complete vertical stacks on top of the base models. You see academia — Stanford HAI, MIT CSAIL, Berkeley, CMU — producing evaluations, benchmarks, safety papers. You see the specialised Anglo press — Stratechery, Platformer, MIT Tech Review, 404 Media, The Information — covering the sector with deep journalism. And you see the structural critical voices — Crawford, Zuboff, Gebru, Hinton, O'Neil, Bender — addressing the ecosystem from outside.
On that map, IBM appears as a footnote in the chapter on enterprise providers. That's no insult. It's honest geographic placement. The thing is that placement doesn't match its presence on the Spanish SERP, where it appears at the centre, occupying the space that on an honest map would belong to the AI Index, Stratechery or Crawford.
SEO is not journalism
There will come a moment, perhaps five years from now, when the informational content about AI produced in 2020-2023 is so outdated that not even Google can hold it up top. There will come a moment when new domains — independent media, sector blogs with their own voice, serious translations of the Anglo sources — accumulate authority and displace the SaaS brands from the top positions. It will be a slow process because domain inertia is enormous. But it will happen.
Meanwhile, it's worth knowing. When you go to IBM.com and read its listicle on the ten dangers of AI, you aren't reading a player in the sector explaining what it sees from inside. You're reading the content department of a company that decided a decade ago it was going to compete in AI, didn't make it, and now lives off the domain rent to stay visible. That isn't journalism. It isn't science communication. It isn't criticism. It's institutional marketing with SEO. SEO is not journalism, even if it sometimes looks like it.
Knowing it doesn't solve the problem. But it stops you believing you've already understood just because you read Google's first result. And that, in this sector and at this moment, is the first step to understanding anything.
Quick definitions
- Domain authority: a composite metric Google uses to evaluate the historical reliability of a web domain. It includes age, inbound link profile, editorial consistency. It's highly persistent and poorly correlated with the broadcaster's current technical relevance on each specific topic.
- Frontier model: an AI model sitting at the state of the art of technical capability at a given moment. In 2026 the frontier models are published by OpenAI (the GPT-5.x family), Anthropic (Claude Opus 4.x), Google DeepMind (Gemini 3.x), xAI (Grok 4.x) and the Chinese labs DeepSeek and Alibaba, among others. IBM (Granite) is not in this category.
- Chatbot Arena: a public evaluation platform where users vote blind on which model answers their question best. It produces a continuously updated Elo ranking. It's the least manipulable measure of the sector's real state.
- Domain rent: the competitive advantage a web domain accumulated over decades keeps on the SERP even though its current content no longer reflects the technical frontier of its sector.
Referencias
- Stanford HAI (2026). AI Index Report 2026 (published April 2026). https://hai.stanford.edu/ai-index/2026-ai-index-report — technical-performance chapter: https://hai.stanford.edu/ai-index/2026-ai-index-report/technical-performance
- Chatbot Arena / LMArena (operated by LMSYS). Leaderboard. Empirical Elo rankings by blind voting, continuously updated. https://lmarena.ai/leaderboard
- Epoch AI. Notable AI Models. Historical registry of frontier models.
- Christensen, C. (1997). The Innovator's Dilemma. Harvard Business Review Press.
- Lehdonvirta, V. (2022). Cloud Empires. MIT Press.
- STAT News (2017). Account of the MD Anderson project with Watson, its halt with an audited cost of around 62 million dollars and no clinical use. https://www.statnews.com/2017/09/05/watson-ibm-cancer/
- IBM (2022). Sale of the Watson Health assets to Francisco Partners, which went on to operate under the Merative brand. https://newsroom.ibm.com/2022-01-21-Francisco-Partners-to-Acquire-IBMs-Healthcare-Data-and-Analytics-Assets
- IBM (2024-2025). 10 AI dangers and risks. https://www.ibm.com/think/insights/10-ai-dangers-and-risks-and-how-to-manage-them
- IBM. Granite (IBM's family of open enterprise AI models, integrated into watsonx; versions 3.0 in October 2024 to 4.0 in October 2025). https://www.ibm.com/granite
- Crawford, K. (2021). Atlas of AI. Yale University Press.
- Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.
- Bender, E., Gebru, T., McMillan-Major, A., Shmitchell, S. (2021). On the Dangers of Stochastic Parrots. FAccT '21.
También te interesa
- Why the critiques of AI aren't made by real experts
- How did the public conversation about AI break?
- The bullets UNESCO wrote about AI
No comments yet
No comments yet. Be the first.
Leave a comment