Ten reasons never to read another AI listicle in your life

Today we look at the format that closes this anti-clickbait section, and I'll start by confessing that this article's title is deliberately rigged. I've spent twenty pages criticizing ten-point listicles and yet here I come offering a ten-point listicle. The trap is the point. The promise is that, if you make it to the end, you'll come out with antibodies against any artificial-intelligence listicle you cross this week. My take is biased: I've produced bad listicles in other professional lives and I know the mechanics from the inside. Form your own. Read the ten reasons and judge whether the format comes out defensible or not.

I've kept a notebook for years where I jot down what the AI listicles passing through my feed have in common. The observations have gradually organized themselves, and, since the format demands ten points, I'm going to deliver them in ten bullets so the irony can do its job. But —and here's the trick— each bullet will include what the average listicle never includes: context, nuance and the burden of proof. If at the end the reader decides the listicle is still a useful format, that'll be their informed choice. If they decide it isn't, this article will have done its job.

One. Ten is arbitrary and almost always wrong

On no serious artificial-intelligence topic are there exactly ten relevant elements. For technical problems —architectures, methods, benchmarks, trade-offs— the realistic numbers range between three and thirty. For ethical problems —biases, concentration, capture, labor impact— there are four or five structural risks and dozens of derivatives. The number ten isn't determined by the topic; it's determined by the content-distribution algorithm. The listicle chose ten because ten works in search and on social media. The number precedes the content. That alone should stop the reader.

Two. To reach ten you have to pad

If the topic has six important elements, the ten-point listicle will include four more to fill the format. Those four extra bullets are always the weakest, the most generic, or the most recycled from previous listicles. The reader who reaches bullet seven usually finds an observation so vague it couldn't apply specifically to AI: "artificial intelligence is changing the world" or "you have to stay up to date." Those bullets don't inform; they take up space. And since the average reader reads listicles with decreasing attention, the weak bullets at the end pass without scrutiny.

Three. For the bullets to look equivalent you have to flatten

An honest taxonomy of AI's problems is hierarchical: there are grave structural risks —corporate concentration, input capture— and secondary risks —algorithmic bias in high-risk applications— and speculative risks —singularity, artificial consciousness—. The ten-point listicle presents all three levels as bullets of similar length and visual weight, with no explicit hierarchical distinction. That equivalence is fictitious. Mixing corporate concentration with the singularity is mixing concrete present with hypothetical future, and mixing them as if they were equivalent problems disorients the reader.

Four. The listicle loses the causal connection

The technical and political reality of AI is made of causal chains: the concentration of compute produces dependence on a few companies, which produces capture of citizen input, which produces a permanent competitive advantage, which produces concentration of compute in the next cycle. Each link explains the next. The listicle, by splitting reality into independent bullets, breaks the chain. The reader comes away with loose pieces and no frame for reconstructing how the system works. The consequence is being able to say "there are ten problems" without understanding how they're generated or how they reinforce one another.

Five. The listicle allows no internal contradiction

A serious idea about AI almost always contains internal tension. The tool is useful and, simultaneously, can atrophy skills. The open model democratizes access and, simultaneously, makes malicious uses easier. Automation increases productivity and, simultaneously, intensifies work. Each of those tensions is the reality that serious analysis holds without resolving. The listicle bullet can't hold the tension: the format demands clean, closed statements. That's why, in every listicle, one of the two sides of the tension falls away. The reader gets half the picture and doesn't know the other half existed.

Six. The listicle rewards false certainty

"The ten problems" sounds closed, exhaustive, complete. The reality is that no serious list about AI can be exhaustive, because the field evolves on the scale of months and new problems emerge continuously. Two years ago it wasn't a priority to talk about model collapse, three years ago it wasn't a priority to talk about electoral deepfakes, four years ago it wasn't a priority to talk about the chip embargo on China. A listicle written today will be out of date in twelve months. The format lies by presuming exhaustiveness: it presents as a finished map what is, at most, a snapshot of an instant.

Seven. The listicle favors the cognitive shortcut

The cognitive literature on reading formats is consistent. Research by Maryanne Wolf, gathered in Reader, Come Home (Harper, 2018), documents that fragmented reading produces lower semantic retention than sustained reading. The listicle is the paradigmatic example of fragmented reading: the reader jumps from bullet to bullet, without building the argument, without remembering the previous bullet when reaching the next. The sense of having understood is high —because the format is comfortable— but retention a few days later is extremely low. This is measured, not opinion.

Eight. The listicle alienates the demanding reader

There's a selection effect worth naming. The reader who wants to understand AI seriously stops, within a matter of months, reading listicles. They recognize them and skip them. The consequence is that the format ends up serving, exclusively, the reader who doesn't demand depth, which confirms the editor in the decision to keep producing listicles. The selection reinforces the cycle. The market for serious outreach in Spanish erodes because the reader who would sustain it can't find content and leaves for English or leaves for books.

Nine. The listicle blocks the space for analysis

Here comes the most operational piece for anyone producing content. When the listicle format dominates the SERP for a keyword —say "dangers of AI," "advantages of AI," "how ChatGPT works"— the in-depth analytical article can't compete on position. However technically superior it is, however better its sources, however much historical context it offers, it stays off Google's first page because the algorithm rewards metrics the listicle satisfies better (high click rate, retention to the end, shares on social media). The analyst loses because their format doesn't fit the metric. This is exactly what you're reading in 0061 and what sustains the mediocre ecosystem.

Ten. The meta reason

If any of the previous nine reasons were, on its own, enough to abandon the listicle format, then listing ten reasons is excessive. Which confirms reason one —ten is arbitrary—. Which confirms the format's own defect. And closes the loop.

What's left after the loop

With the listicle closed, it's worth asking what one does upon abandoning it. Here are four brief operational habits.

The first habit is reading medium- or long-form articles instead of listicles. A well-structured piece of 2,000 to 3,000 words leaves more retained operational knowledge in the reader's brain than five 800-word listicles. The time investment is similar; the return is far higher.

The second habit is preferring content signed by a concrete person over content signed by an "editorial team." The first format puts forward an author with a biography, a public point of view, a sustained presence. The second is a marketing product. The individual byline is a signal of editorial commitment.

The third habit is reading at least one source that challenges the reader's preferred thesis. If you tend to read criticism of AI, also read serious pro-adoption analysis. If you tend toward enthusiasm, read the structural critiques. Plurality isn't relativism; it's mental hygiene.

The fourth habit is reading books, not just articles. A book on AI —Crawford's, Russell's, Karen Hao's— consolidates analytical frameworks no listicle or isolated article can replace. Three books a year deeply change the quality of a reader's thinking. Thirty listicles a year don't.

The political question

This is personal opinion, but the media ecosystem's own evolution supports it. The listicle isn't a traditional journalistic genre that was discovered one day; it's a product of the BuzzFeed era and of the social-media optimization of the early 2010s. Before 2008, practically nobody wrote listicles in serious press. After 2012, almost all digital media adopted it as a stable genre. The standardization was fast and readers accepted it without protest.

The aggregate consequence, measured in terms of the quality of public conversation, is modest and measurable. Complex topics —and AI is one of them— are understood worse when most of the information about them travels in listicle format. People form opinions on the basis of simplified bullets rather than structured analysis. The political, professional and personal decisions resting on that opinion inherit the simplification.

What I'd ask for —again with limited realism— is that serious media set an explicit editorial policy against the listicle except when the format specifically fits the content. There are topics where a numbered list is the right form —installation instructions, a list of people affected in an incident, a calendar of events—. There are topics where it's deliberately misleading —"AI's problems," "advantages of adopting X"—. Distinguishing between the two cases isn't complicated; all it requires is editorial priority and a willingness to lose some traffic in exchange for credibility.

In the meantime, the individual reader has the simplest decision. When they see a headline "The X tips for Y" or "The X reasons why Z," they can simply skip to the next result. The SERP usually has ten results on the first page; at least one or two of them are usually serious analysis if you search patiently. Patience is the only requirement.

The concrete figure to close on. According to the Reuters Institute's Digital News Report 2025, "numbered or list" formats account for around 23% of the most-read content in the technology sections of the main European digital media, while in-depth analytical content —pieces over 1,500 words with verifiable citations— accounts for less than 6% of measured consumption. That proportion describes the ecosystem better than any opinion. And it also describes why this article, deliberately designed in listicle format only to mock the format, is doomed to be read by fewer people than any conventional listicle on the same topic. That is exactly the trap of the system I've been describing.

Definitions

Listicle: an article structured as a numbered list with a headline of the type "the X reasons" or "the X tips." A genre that originated on the BuzzFeed website in the late 2000s and was adopted en masse by generalist digital media and corporate blogs.

Semantic retention: the percentage of presented information the reader remembers and can retrieve some time after reading. A more informative metric of a piece's quality than the simple click metric.

Cognitive shortcut: a mental strategy by which a conclusion is reached without traversing the full chain of reasoning that justifies it. Listicles favor the use of cognitive shortcuts over structured reasoning.

Social-media optimization: the tuning of content to the metrics rewarded by social-media algorithms —short retention, high sharing, emotional polarization—. The listicle is one of the genres most optimized for these metrics.

References

Reuters Institute for the Study of Journalism, Digital News Report 2025 (Oxford University, June 2025). Figures on the consumption and production of numbered formats versus in-depth analysis.

Maryanne Wolf, Reader, Come Home: The Reading Brain in a Digital World (Harper, 2018). Neurocognitive research on the difference between sustained reading and fragmented reading.

Nicholas Carr, The Shallows: What the Internet Is Doing to Our Brains (W. W. Norton, 2010). An analysis of how the web format shapes attention and retention.

Pew Research Center, surveys on format and retention of online news (2020-2025). Data on the proportion of readers who remember content a month after reading it.

Edson Tandoc Jr., Tell Me Who Your Sources Are: Perceptions of News Credibility on Social Media (Journalism Practice, 2019). An academic analysis of how format shapes the perception of credibility.

Neil Postman, Amusing Ourselves to Death (Viking, 1985). A classic text on how format determines the transmissible content. A necessary close to the five mentions the batch allows for this author.

Karen Hao, Empire of AI: Dreams and Nightmares in Sam Altman's OpenAI (Penguin Press, 2025). An example of the extended analysis the listicle replaces.

Further reading

Cal Newport, Digital Minimalism: Choosing a Focused Life in a Noisy World (Portfolio, 2019). An operational manual for cutting digital consumption without losing functionality.

Ryan Holiday, Trust Me, I'm Lying: Confessions of a Media Manipulator (Portfolio, 2012, revised edition 2017). An insider's manual on how viral content is designed, the listicle included.

Maria Konnikova, Mastermind: How to Think Like Sherlock Holmes (Viking, 2013). An accessible framework on how sustained reasoning is built versus quick jumps.

Robert Cialdini, Influence: The Psychology of Persuasion (Harper Business, 1984; revised edition 2021). The canonical text on the persuasion mechanics the listicle systematically exploits.

You might also like

• The 10 dangers of AI and a heap of nonsense
• Five advantages of AI, depending on who's selling it to you
• OpenAI's press release is not journalism
• Three months without reading headlines to understand AI

Elsewhere

• Nieman Journalism Lab (Harvard)
• Digital News Report 2025 — Spain (Reuters Institute)