The Thought Leader's Playbook for the AI Era

The day explanation became cheap

In late 2022, a peculiar economic event occurred in public. A task that had once consumed billable hours, editorial budgets and a fair amount of professional prestige — producing a tidy, competent essay on almost any subject — collapsed in price towards zero.

Ask a modern language model for a summary of Basel III, a primer on menopause in the workplace, or a respectable overview of battery supply chains, and it will usually deliver something fluent, organised and broadly adequate in seconds. Not definitive. Not always reliable. But good enough for the casual reader, the busy executive, the conference producer, or the junior marketer staring at a content calendar.

That shift has not eliminated expertise. It has, however, demolished a particular business model of expertise: being the person who explains familiar things clearly in public. For years, much of thought leadership rested on that premise. If you could synthesise the trade press, smooth the jargon and publish with consistency, you could accumulate an audience. Clarity was scarce. Distribution was manageable. The internet rewarded the articulate intermediary.

Now the intermediary faces a machine that can imitate the form of expertise at industrial scale.

The consequence is not that all expert content becomes irrelevant. It is that the centre of gravity moves. The value migrates away from competent synthesis and towards what cannot be cheaply inferred from the public corpus. The winners are not producing more content in panic. They are doing something subtler and, to many incumbents, deeply counter-intuitive.

They are publishing what models cannot know, taking positions that could be wrong, and building bodies of work that compound over time.

The collapse in scarcity

To understand why this matters, it helps to be clear about what AI has actually commoditised.

Large language models are prediction engines trained on immense quantities of existing text, code and image data. Their great commercial trick is not omniscience but plausibility. They are exceptionally good at producing the kind of language that looks like the next sensible sentence. In practice, that means they are increasingly adept at generating:

• overviews

n- listicles

• executive summaries
• explainer posts
• first-draft white papers
• comparison tables
• interview prep notes
• conference abstracts
• routine marketing copy

This is why firms from Microsoft to Google have embedded generative tools directly into productivity software, and why software companies across categories now promise AI-assisted drafting as a standard feature. It is also why traffic-dependent publishers have spent the past two years in a collective identity crisis. If search users can obtain an instant summary at the top of the results page, or generate one themselves, the old rewards for producing one more interchangeable explainer begin to disappear.

One need not subscribe to every extravagant claim about AI to see the effect. The internet is already being flooded with synthetic prose. Researchers at Stanford and Georgetown have documented the accelerating presence of AI-generated and AI-assisted content in public discourse. Educators report a dramatic increase in machine-written assignments. Platforms are adjusting policies to cope with synthetic media and disclosure issues. The practical point is simple: competent text is becoming abundant enough to lose signalling power.

When abundance rises, scarcity changes shape.

Move one: publish what a model cannot know

The first move in the new playbook is the most important: become a primary source.

A language model can remix what is already legible in its training data or prompt context. It cannot directly know what happened in your customer interviews last month, what conversion pattern you observed across 600 sales calls, what your laboratory discovered on Tuesday, or what your supply chain team learned after switching a crucial component supplier in Vietnam. It cannot independently originate first-hand field evidence. That remains the human expert’s strongest moat.

This is why the most durable intellectual brands increasingly look less like commentators and more like research organisations.

Consider the firms and analysts that still reliably shape debate in saturated fields. Mary Meeker’s internet reports mattered not because she could write neatly about technology, but because she assembled original market evidence into a coherent interpretation. Benedict Evans remains widely read because he combines synthesis with an analytically distinctive frame built over years, not because he merely summarises headlines. Stratechery by Ben Thompson works for the same reason: the attraction is not access to facts, but a disciplined explanatory model applied repeatedly to unfolding events.

In B2B markets the pattern is even clearer. HubSpot built authority partly by publishing benchmark data drawn from its own ecosystem. Gartner’s influence depends less on elegant prose than on privileged access, method, taxonomies and accumulated client intelligence. McKinsey, when it is at its best, commands attention through surveys, sector datasets and operator interviews rather than generic management bromides. In software, companies such as Cloudflare or Shopify attract serious readership when they reveal what they can uniquely see: network-level trends, merchant behaviour, system performance at scale.

The same logic governs journalism. The New York Times, the Financial Times and Reuters have all had to confront AI’s ability to summarise public information. Their future defensibility lies not in being paraphrased, but in original reporting, source networks and institutional judgement. Summary can be automated; reporting cannot be conjured from the void.

From content engine to evidence engine

For executives, consultants and independent experts, this implies a strategic inversion. The question is no longer, “What should we post this week?” It is, “What can we observe that nobody else can?”

That observation might come from:

AI did not erase expertise; it erased the premium on generic explanation.

• proprietary data from products, platforms or operations
• first-party surveys with a clear method
• repeatable experiments and measured results
• field notes from client work, suitably anonymised
• interviews with practitioners at decision level
• internal process failures and the lessons drawn from them
• longitudinal observation across a specific niche

This need not be grand. A recruiter with real salary data has an edge over a hundred generic LinkedIn commentators. A GP writing about what patients actually ask in consultations can outperform a polished health copywriter. A founder with six years of pricing experiments has material no model can genuinely invent.

The key is not merely possessing proprietary experience; it is rendering it legible. Too many organisations sit on insights locked inside service teams, support tickets, analytics dashboards and project retrospectives, while their external content remains interchangeable. The editorial opportunity is to convert operational reality into publishable evidence.

Why evidence now travels further

There is another, subtler effect. In an age of synthetic abundance, audiences have become more sensitive to the texture of reality. Concrete detail stands out. The meeting that went wrong. The chart with a surprising break point. The implementation constraint nobody mentions on stage. The quote from a chief risk officer who changed her mind after deployment.

Readers may not articulate it this way, but they are increasingly scanning for signs that a piece was written from contact with the world rather than from contact with the internet.

Move two: take a position that could be wrong

The second move is harder for institutions because it carries reputational risk. It is also what makes a human voice recognisable.

Models are optimised, by design and by pressure, towards balance, safety and probabilistic middle-ground. Their outputs tend to flatten conflict. They hedge. They enumerate trade-offs. They often sound sensible and bloodless at once. This is useful in many contexts. It is also forgettable.

Authority, by contrast, is often built through selective asymmetry: seeing one fact as more important than others and saying so plainly.

A real point of view has three properties. It privileges. It excludes. And it can be falsified.

That is why the most memorable public thinkers rarely speak in fully de-risked prose. Aswath Damodaran is not read because he describes valuation techniques in neutral corporate language; he is read because he applies a distinct judgement to contested companies and is willing to be challenged. Ed Conway on economics, Zeynep Tufekci on technology and society, or Tyler Cowen on culture and growth all retain readership because they make arguments, not merely compilations. Even when readers disagree, they know where the writer stands.

The disappearance of the bland middle

In the pre-AI content economy, there was room for a vast middle tier of “useful but generic” thought leadership. Search rewarded it. Social algorithms occasionally amplified it. Sales teams appreciated having it. Now machines can generate an endless supply of that category.

This leaves a harsher editorial landscape with two poles:

• low-value generic content, increasingly automated
• high-value distinctive argument, grounded in evidence and personality

The middle is being hollowed out.

This is not licence for theatrical contrarianism. Much online “hot take” culture is merely noise in expensive clothing. The point is not to be provocative for the sake of it, but to make a claim strong enough that a serious person could dispute it.

“The future of work is changing rapidly” is machine fodder.

“Most firms do not have a return-to-office problem; they have a manager capability problem disguised as a property question” is at least an arguable thesis.

“AI will transform healthcare” is vapour.

“AI’s near-term impact in healthcare will be largest in documentation, coding and workflow orchestration, not diagnosis” is a claim one can examine against deployment patterns, reimbursement structures and regulatory realities.

A point of view that could be wrong is often the clearest signal that genuine judgement is at work.

Why institutions struggle with this

Most corporate content is produced by committee under conditions designed to remove risk. Legal reviews sand off edge; brand teams standardise tone; executives fear alienating potential buyers. The result is text that offends nobody and moves nobody.

This problem will worsen, not improve, with AI. If an organisation already tends towards safe consensus language, AI systems will accelerate that tendency by producing polished drafts in precisely the same register.

The antidote is procedural as much as stylistic. Give real operators authorship. Let named experts sign pieces. Separate thought leadership from product copy. Build an editorial process that asks, before publication, what is the non-obvious claim here, and what evidence supports it?

A point of view that could be wrong is often the clearest signal that genuine judgement is at work.

There is a governance lesson in that. The most credible uses of AI increasingly depend on decisions made before execution, not after. In Society OS terms, F-ACT — the Framework for Agent Conformance & Trust — begins with ASDAR: Authority, Scope, Data, Audit, Revocation. The principle is plain: govern before execution — not after. Editorial systems would benefit from the same discipline. Who is authorised to speak? On what scope? With what evidence? How is the argument auditable? And if a claim proves wrong, how is it revised or withdrawn? In an era of machine-amplified publication, thought leadership needs not less conviction, but better governed conviction.

Move three: compound a body of work

The third move is the least glamorous and perhaps the most decisive. Individual posts are now brutally easy to imitate. A body of work is not.

A serious intellectual reputation is rarely built on a single viral essay. It emerges from repeated publication that deepens a framework over time. The audience does not merely consume isolated pieces; it learns a vocabulary, an angle of attack, a way of seeing.

This is why the strongest thought leaders increasingly resemble institutions, even when they are individuals. Their work interlocks. Their essays refer back to earlier concepts. Their examples accumulate. Their forecasts can be checked against prior claims. Over time, readers come not only for answers but for orientation.

Stratechery is again a useful example. So is Azeem Azhar’s Exponential View. In economics, Paul Krugman’s long-running columns have mattered less because each one is uniquely revelatory than because they form a durable archive of argument. In business, Michael Porter’s influence was not any single article in Harvard Business Review, but an architecture of ideas around competitive strategy that executives could inhabit for decades.

The moat is coherence

This matters because AI can generate a piece in your style; it cannot easily generate the historical continuity of your mind.

A compounding body of work creates several advantages at once:

• it gives new readers a path deeper into your thinking
• it raises the cost of imitation, because imitators can copy outputs but not lineage
• it allows concepts to be refined rather than endlessly restated
• it makes disagreement productive, because critics must engage a framework rather than a slogan
• it converts content from a stream into an asset

This is especially powerful in professional services and founder-led businesses, where trust depends on repeated demonstrations of judgement. A consultant who has built a decade-long public thesis on pricing strategy is harder to replace than one who posts generic advice every Tuesday. A biotech founder who steadily explains regulatory design, trial trade-offs and failed assumptions builds a different class of authority from a peer who merely recycles headlines from Nature and STAT.

The architecture of accumulation

To compound a body of work, experts need an architecture, not just a posting habit.

That usually means:

• a small set of enduring themes
• named concepts or frameworks
• canonical essays worth revisiting
• periodic updates to prior theses
• consistent examples and case comparisons
• a searchable home rather than dependency on one platform

The point is not to become jargon-heavy. It is to create intellectual continuity. Readers should be able to say, with some precision, what your worldview is and how it has evolved.

In the AI era, that coherence becomes a form of provenance.

The false comfort of volume

Many brands are responding to generative AI in exactly the wrong way. Seeing content become easier to produce, they assume the rational response is to produce more of it.

This is understandable. For two decades, digital strategy often rewarded volume. More pages meant more search surface area. More posts meant more chances to catch an algorithmic wave. More newsletters meant more touchpoints.

But when the marginal cost of production collapses across the market, volume ceases to differentiate and starts to devalue the whole category.

The web is already showing signs of this. Google’s search results have become cluttered with SEO-shaped content designed to answer queries in broadly similar language. Google’s own response, including its repeated emphasis on E-E-A-T — experience, expertise, authoritativeness and trustworthiness — is effectively an admission that the platform must distinguish between pages that merely resemble knowledge and those that arise from it. The system is imperfect, but the direction is revealing: provenance matters more when cheap text is everywhere.

A parallel can be seen on LinkedIn, where formulaic AI-assisted posting has become easy to spot. The tell is not usually poor grammar. It is generic confidence unbacked by concrete experience. The posts read as though they were generated from a shadow idea of professional insight. They often perform briefly and then vanish, because they leave no residue in the reader’s mind.

What this means for experts, firms and media brands

The practical implications differ slightly by category, though the governing logic is the same.

For individual experts

The moat is not any single post; it is the accumulated worldview.

Stop trying to win by being comprehensive. Win by being close to reality.

Publish fewer pieces, each with more original reporting, stronger examples and a clearer thesis. Keep notebooks on recurring anomalies in your work. Turn client questions into research agendas. If you are wrong, say so in public and explain why. Readers trust visible revision more than hidden certainty.

For professional services firms

Mine the firm’s actual operating knowledge.

The richest material is usually trapped in delivery teams, post-project reviews, analyst models and partner conversations. Build editorial systems that can surface this without breaching confidentiality. Treat proprietary pattern recognition as a strategic asset. If your public content could have been produced by an intern with a model and a browser, you are leaking prestige.

For corporate brands

Distinguish between content marketing and institutional thinking.

The former can and should use automation where appropriate. Product descriptions, FAQs, routine drafts and campaign variants are fair terrain for AI assistance. The latter — the ideas by which a company wants to be known — requires senior judgement, named ownership and evidentiary discipline.

For media organisations

Double down on original reporting, analysis and archives.

The cheapest part of the value chain is now summary. The most defensible is source access, verification, context and a trusted editorial tradition. Subscription businesses that survive will do so because readers value the institution’s judgement, not because it can paraphrase yesterday’s press release faster than a model can.

The deeper cultural shift

There is a larger reason this playbook matters. Thought leadership is not only a marketing tactic; it is part of how societies decide whom to trust when knowledge is distributed unevenly.

In that sense, generative AI is not merely changing content production. It is changing the social signals by which expertise is recognised. Fluency used to imply effort. Effort often implied competence. That chain has broken. Fluency now implies almost nothing on its own.

What replaces it are older, sterner markers of authority: direct evidence, accountable judgement, consistency over time, and a willingness to be specific enough to fail.

This may prove healthy. The internet’s long boom rewarded charisma, speed and relentless output. The next phase may favour people and institutions that can show their workings, disclose their basis for belief, and build trust through accumulated contact with reality.

The playbook, stated plainly

All three moves share the same logic: stop competing with the machine on volume and speed, and compete on the things the machine structurally cannot do.

Those scarce assets are not mystical. They are concrete.

• proprietary experience
• original evidence
• lived judgement
• defensible interpretation
• a coherent worldview that compounds

Everything else is becoming cheaper by the quarter.

That does not mean experts should reject AI. On the contrary, they should use it aggressively for transcription, drafting, comparison, summarisation, editing and research support. The trap is allowing the machine to occupy the highest-value layer of the stack: deciding what is worth saying.

After the flood

The AI era will not abolish thought leadership. It will simply make lazy thought leadership intolerable.

When anyone can generate a passable essay on any topic, generic authority evaporates. What remains is rarer and more demanding: the discipline to observe something first-hand, the courage to advance a claim sharper than consensus, and the patience to build a body of work that outlasts the feed.

The experts who will still command attention are not the loudest publishers. They are the clearest owners of reality.

In a world of infinite competent prose, that is what authority looks like.