0. Introduction

In 2025, serious arguments about the future of intelligence no longer come wrapped in leather-bound books, or even in the respectable blue and grey of an OECD working paper. They come wrapped in soft off-white webpages, sparsely adorned, scroll-driven, and quietly insistent.

You have probably seen them: an austere title at the top of the page – Situational Awareness, The Compendium, A Narrow Path. A few muted buttons. A large-font quote, maybe from a nuclear physicist or a Cold War strategist. And then the scroll begins. The world narrows into sequential screens of cool fonts, muted greens and reds, log-scale graphs, and urgent but clinical prose. If you are patient (and you should be), you might be rewarded with a choose-your-own-adventure ending: Will humanity slow down in time? Will the race for superintelligence consume us?

These AI microsites are not just web design trends. They are a new rhetorical genre: part research briefing, part activist playbook, part intellectual performance. They occupy an uneasy space between the slow caution of peer-reviewed science and the raw urgency of political campaigning. And because they are so visually polished, so editorially deliberate, they quietly shape what readers believe to be reasonable, probable, and inevitable.

In this essay, I want to trace the rise of the AI microsite. I will map its form and aesthetic. I will place it inside the wider patchwork of AI safety thinking where hard fact bleeds into forecasting, normative argument, and speculation. I will suggest that the microsite is not just how AI futures are explained: it is how AI futures are made plausible.

Finally, I will ask what it means for policy, for public discourse, and for memory itself when our most serious arguments are published not in journals or newspapers, but in artifacts as ephemeral, persuasive, and architected as these.

But before we can understand the microsite as a persuasion engine, we have to understand its lineage. These sites did not appear from nowhere. They inherit both the ambitions and the ambiguities of a longer tradition: the grey literature of policy reports, working papers, and white papers – documents designed not to endure in the scholarly canon, but to intervene in the moment.

The AI microsite is grey literature stripped of its bureaucratic drabness and rebuilt for a web-native world. To see how radical that shift is, we have to start with what grey literature used to be and what it was supposed to do.

How to Read a Microsite (in four easy steps)

Admire the minimalism. Pale backgrounds. Sparse fonts. A serious quote. Trust begins here.

Surrender to the scroll. No escape hatches – just you and the unfolding argument.

Absorb inevitability. Graphs will slope upwards. Percentages will climb. Scenarios will narrow. This is not a bug.

Decide, but only between two options. Pause the race, or lose it. Save humanity, or doom it. Choose wisely.

If you feel a sudden desire to sign something, propose legislation, or reallocate $100 billion, congratulations: the microsite has worked.

1. Grey literature with a minimalist skin

Grey literature has always been the nervous system of serious policy work. While books garner prestige and journals accumulate footnotes, it's the reports, working papers, and issue briefs that move more nimbly between them – documents crafted not to settle arguments, but to shape what happens next.

Traditionally, this material embodied physical and stylistic restraint. Government white papers printed on stiff stock, conference papers photocopied and stapled, early web-era PDFs with Times New Roman body text and clip-art flowcharts. This restraint wasn't merely aesthetic; it reflected the institutional culture from which these documents emerged. Bureaucratic forms conveyed procedural legitimacy. Standardised formatting signalled deference to organisational hierarchy. The simple, often drab presentation communicated that these were serious documents meant for serious people making serious decisions.

The form mirrored the function: seriousness was performed through minimal design and maximal documentation. Authority rested in substance and weight: the number of references, the appendices, the committee memberships listed on the cover page.

The new generation of AI microsites inherits the ambition of grey literature – its speed, influence, and position at the intersection of research and advocacy – while discarding almost everything else about its presentation. Gone are the paperweight PDFs and bureaucratic conformity. In their place: lightweight, scroll-driven sites with generous typography, muted color palettes, thoughtful editorial gravity, and deliberate, slow-paced unfolding of ideas.

If old grey literature wore a dusty suit, the new microsite wears a carefully ironed shirt without a tie. The effect is not casual, it’s carefully considered. It signals that the work is serious enough not to require old-fashioned trappings of authority.

This aesthetic transformation goes beyond mere questions of taste. The visual language of the AI microsite performs substantive epistemic work. It preloads an impression of seriousness and shapes how claims feel before the reader can even begin to assess them. Before a single argument is articulated, the reader has already been subtly nudged to expect something important, considered, and true.

1.1 The microsite aesthetic: quiet gravity

Across the landscape of prominent AI microsites, a consistent visual vocabulary emerges with remarkable coherence. This aesthetic isn't the neon-on-black palette of techno-dystopian design, nor does it embrace the cheerful primary colors of startup culture. Instead, the palette gravitates toward muted off-whites, pale greys, and soft accent colors: gentle blues, dusky greens, bloodless reds. The cumulative effect is editorial rather than cinematic – more reminiscent of The New Yorker than Blade Runner.

Typography follows this same principle of restrained authority. Clean, modern sans-serif fonts dominate body text, occasionally complemented by serious serif headings that add a touch of intellectual weight. Font sizes are consistently generous, creating abundant white space that gives ideas room to breathe. Decorative elements are kept to a minimum: few flourishes, sparse iconography, and almost no ornamental linework to distract from the content.

The layouts themselves are linear and deliberate, guiding users vertically through sections. These sections are often punctuated by large block quotes or sub-chapter headings that provide rhythmic breaks in the narrative flow. Navigation menus, when present, remain spare and unobtrusive; tables of contents are foregrounded. Very few sites encourage lateral browsing or non-linear exploration. The design suggests – almost demands – that the reader proceed through one idea at a time, in the precise order intended by the authors.

Data visualisations appear sparingly throughout these sites, but when deployed, they maintain the same clean, stark aesthetic: bar charts tracking compute growth, timelines projecting AGI forecasts, minimalist tables comparing model capabilities. These visualisations don't present data to invite debate or alternative interpretations, but rather to establish inevitability.

The numbers consistently slope upward, the scenarios progressively narrow, and the outcomes appear to crystallise into predetermined shapes.

Colour itself serves as a signaling mechanism. Green represents optimistic outcomes: scenarios where humanity successfully implements slowdowns, imposes thoughtful regulation, or averts potential disasters. Red signals failure: scenarios where competition accelerates beyond control or safeguards prove inadequate. Even when different microsites offer divergent explicit forecasts, the use of visual coding remains remarkably consistent, embedding an emotional valence into the very architecture of the page.

This carefully calibrated aesthetic performs substantial rhetorical work. It frames even highly speculative claims with an aura of plausibility and projects objectivity without having to explicitly argue for it. The design trains readers to expect seriousness – and consequently, to experience the content as serious, regardless of its actual epistemological status.

In this way, the AI microsite doesn't merely contain an argument – it constitutes an argument in itself. The selection of palette, typeface, layout, and scroll pacing subtly nudges readers toward treating the presented claims not as provocations to be questioned, but as realities to be absorbed and accepted.

This new rhetorical power is not neutral. The same techniques that elevate urgent, careful arguments can just as easily be turned toward factional narratives, selective framing, or premature certainty. As microsites become more sophisticated, the line between civic persuasion and ideological engineering grows thinner.

2. Form follows function: scrollytelling and solemnity

If the aesthetic of the AI microsite sets the mood, its structure disciplines the experience. Unlike traditional articles, where the reader can skim, skip, or double back at will, most microsites lock the reader into a slow, downward drift through sequential argument. One idea per screen, one progression per scroll. Reading a microsite is less like parsing a policy brief and more like moving through a curated exhibition where each room reveals a carefully positioned artifact or insight. Now the effect becomes almost cinematic.

This structure – often called "scrollytelling" in web design – emerged from longform journalism in the early 2010s. The New York Times' Snow Fall is usually credited as the breakthrough: a piece that unfolded its story through full-bleed visuals, triggered animations, and careful vertical pacing. The AI microsite borrows this model, strips it of visual flamboyance, and repurposes it for persuasion. The design choices here are not merely aesthetic but fundamentally rhetorical, transforming how arguments are received and processed.

Each scroll click becomes a small act of consent, a miniature commitment to the unfolding narrative. The reader silently affirms: "Yes, I accept this frame. Yes, I will proceed to the next idea. Yes, I will watch the future narrow." This incremental buy-in creates a momentum of acceptance that would be difficult to achieve in a format allowing more freedom of navigation and selective engagement.

There is a reason so many microsites frame their arguments with chapter structures: "Section 1: Strategic Realism," "Section 2: The Slowdown Scenario," "Section 3: A Narrow Path." The chapterised scroll mirrors the conventions of serious nonfiction, giving the digital experience the rhythm and authority of a hardbound book. It is a rhetorical sleight of hand: the reader feels as though they are moving through something vetted, edited, formal. But in reality, the microsite is web-native, unreviewed, and mutable. It is a living document that can change without notice, erasing its history with each quiet update.

The solemnity of this structure matters profoundly. Microsites are not bombarding the reader with flashing urgency. They are not calling for immediate outrage. They are building a slow, confident inevitability, with a narrative progression that whispers rather than shouts: "This is how the world works now. These are the scenarios available. Choose wisely – but choose from this menu." The measured pace and structural formality create an atmosphere of deliberation that makes radical claims feel more reasonable.

2.1 Choice architecture: constrained agency

Many AI microsites offer the reader choices, an element that initially feels refreshingly democratic. Unlike a static report or op-ed, the microsite promises interactivity – agency – a role for the reader not just as a spectator but as a decision-maker in the unfolding futures being presented. These choice points often arrive after a substantial scroll, positioned as natural culminations of the preceding narrative.

While only a few microsites (notably AI-2027) present explicit branching interfaces, most constrain futures more subtly. Even without visible decision points, the scroll-driven structure funnels readers toward binary or tightly bounded outcomes. One idea per screen, a stark menu of futures at the end: the experience of choice is honoured, but carefully framed.

Upon closer examination, the choices offered are almost always binary, and they are almost always framed within a narrow space of possibility. You can choose a Slowdown or a Race. You can favour Open Models or Closed Labs. You can side with Deliberate Pause or Unchecked Acceleration. No third futures are offered. No real space exists to reimagine the terrain or question the dichotomy itself.

This constraint is not a design flaw but a deliberate feature of the form. By narrowing the range of imaginable futures, the microsite amplifies its persuasive power. By shaping the menu of options, it shapes the reader's imagination. It does not just argue for a particular outcome; it defines what outcomes are thinkable in the first place, establishing the boundaries of reasonable debate.

In behavioural economics, this tactic is called choice architecture – structuring decisions in a way that nudges individuals toward certain outcomes without explicitly forcing them. The AI microsite deploys a narrative version of this: a menu with two dishes, both prepared in the same kitchen, both seasoned with the same implicit priors. The reader's agency is simultaneously honoured and circumscribed, creating an experience of freedom within carefully drawn lines.

The result is a soft coercion, a gentle steering of the imagination. The reader believes they are choosing freely, and in a limited sense, they are – but only among futures already selected and curated by the authors.

This is what gives the microsite its peculiar persuasive strength. It feels participatory, but it is directed. It feels open, but it is narrowed. It feels urgent, but it is staged. Where the old white paper attempted to overwhelm through sheer volume – hundreds of pages, dozens of appendices, exhaustive citations – the microsite persuades through choreography. It guides the reader through a carefully timed sequence: absorb the evidence, decide between constrained options, and arrive at a predetermined conclusion. This three-act structure transforms complex policy debates into streamlined narrative experiences with clear moral arcs.

3. The epistemic patchwork: fact, forecast, fiction

Microsites do not merely deliver arguments. They deliver realities – or at least, the feeling of reality. They create immersive worlds of plausibility that envelope the reader in a coherent narrative. But when examined closely, the claims they make rest on a wide and uneven epistemic terrain – a patchwork of different knowledge types with varying degrees of certainty and verifiability.

At the foundation of this patchwork are hard facts – the empirical bedrock upon which microsites build their more speculative structures. These include benchmarked capabilities of large language models, documented compute scaling trends, chip manufacturing constraints, and recorded incidents or near-misses in biosafety, cybersecurity, or model deployment. These claims are empirical and falsifiable. They can be cited, replicated, challenged. They provide the solid floor beneath the microsite's epistemic edifice.

One layer up are forecasts – projections about future developments based on current trends and expert judgment. These include probabilistic estimates of AGI arrival dates, modelled timelines for automation thresholds, and scenario analyses for geopolitical race dynamics. Forecasts are disciplined guesses, not certainties. Some are based on structured expert elicitation with formal methodologies. Others emerge from simple extrapolation of past curves. They are not facts, but they are not mere fictions either. They are bets, weighted by current information, offering plausible projections of what might come.

Above forecasts sit normative arguments – claims about what ought to be done in response to the facts and forecasts. These include assertions about why we should slow AI progress, why we should regulate compute clusters, or why open access must be curtailed or defended. These claims are not descriptive but prescriptive. They reflect values, priorities, and competing visions of a good (or at least survivable) future. They cannot be falsified in the same way as empirical claims, but they can be scrutinised for consistency, feasibility, and alignment with broader ethical frameworks.

Finally, at the apex of the patchwork, are speculative futures – scenarios involving developments that have not yet occurred and cannot be directly observed or tested. These include narratives about self-improving AGIs, civilisational collapse scenarios, and posthuman transitions. These claims are not empirically grounded yet, because the systems they describe do not exist. But they are not necessarily unscientific either. At their best, they are plausible extrapolations from incentives, physics, and observed capabilities. At their worst, they are science fiction dressed in the language of inevitability.

The strength of the microsite – and simultaneously its danger – is that it stitches all four layers together into a seamless scroll. The reader moves from a benchmarked scaling law to a modelled race dynamic to a normative prescription to a speculative disaster, without necessarily noticing the epistemic shifts underfoot. The transitions between knowledge types are smoothed over through consistent visual design, narrative pacing, and rhetorical techniques that blur the boundaries between what is known, what is projected, what is advocated, and what is imagined.

The result is a peculiar feeling: the possible becomes probable, the probable becomes inevitable. To understand the power of the microsite, we need to understand its method. It does not ask readers to assess each claim separately, with its own appropriate standards of evidence and reasoning. Instead, it invites them to move smoothly across an epistemic terrain where facts, forecasts, normative arguments, and speculations are stitched into a single, coherent narrative progression.

Weaving with confidence

Microsite narratives often weave together empirical data, forecast, value judgement, and speculation into a continuous flow:

Hard Fact Layer

• Benchmarks
• Scaling laws
• Chip supply data

↓ seamless narrative transition ↓

Forecast Layer

• AGI timelines
• Race dynamics
• Automation scenarios

↓ keep scrolling ↓

Normative Layer

• Slowdown arguments
• Regulation proposals
• Openness debates

↓ almost there! ↓

Speculative Layer

• FOOM
• Civilisational risk
• Posthuman futures

Each layer has its own standards of proof. Hard facts can be falsified through experimental evidence. Forecasts can be scored against eventual outcomes. Normative arguments can be debated through ethical and political discourse. Speculative futures can be assessed for internal consistency and plausibility. But the microsite makes these shifts feel frictionless. The statistical law about scaling becomes the prediction about AGI timelines, becomes the imperative to regulate, becomes the warning about civilisational collapse – all within a few scrolls, all rendered in the same typographic style, with the same visual weight.

It is not that the connections are necessarily wrong or deceptive. In many cases, they are carefully argued and reasonably drawn.

It is that the form itself – the soft palette, the scroll pace, the single-column certainty – makes every step feel equally warranted, equally solid, equally deserving of the reader's confidence.

3.1 Blending without breaking: how microsites smooth epistemic shifts

The most sophisticated AI microsites understand that too sharp a jump between empirical fact and speculative forecast risks breaking the reader's immersion. If a benchmark graph abruptly gives way to a civilisation-ending prediction, the reader notices the seam. Doubt creeps in. Persuasion falters.

To avoid this, microsites smooth transitions carefully, using four main techniques.

First, they anchor speculation in near-term trends. Forecasts are paired with empirical scaling laws, chip price curves, or capability incidents, creating the feeling that the speculative is merely an extension of the observed.

Second, they soften normative leaps. Calls to regulate frontier AI emerge naturally after a sequence of data, forecasts, and geopolitical framing. Advocacy feels less like ideology and more like inevitable consequence.

Third, they frame speculation as responsible foresight rather than wild prophecy. Phrases like "it is plausible that" or "if trends continue" offer gentle hedges, maintaining plausibility without encouraging active doubt.

Finally, they enforce visual and tonal consistency. A speculative timeline and an empirical benchmark share the same font, colour palette, and scroll pace. This aesthetic unity lulls the reader into treating all claims – whether factual, forecasted, or imagined – with similar seriousness.

This smoothing is not necessarily deceptive. Many microsites are transparent about uncertainties. But the form itself – slow scroll, muted design, carefully curated options – is optimised for persuasion as much as for information. It encourages the reader to carry confidence in facts forward into far more speculative terrain.

4. Microsites as campaigns, not just briefings

Microsites do not only shape belief. Increasingly, they shape action.

Almost every serious microsite ends with a call to action. Sometimes it is explicit: a petition to sign, a conference to attend, a policymaker to email. Sometimes it is implicit: a strategic alignment of your imagination with the author's preferred future. But the structure is clear. Microsites are not there just to make you know. They are there to make you move.

This is a sharp break from the tradition of grey literature. The working paper, the policy memo, even the governmental white paper – these were designed to inform decision-makers, or at most to nudge them gently. They assumed a separation between information and mobilisation. The microsite discards that separation.

In its structure, the microsite borrows heavily from modern campaign design: frictionless funnels that convert belief to action in a single scroll; embedded sign-up forms for calendars, mailing lists, and event registrations; preformatted policy templates that provide draft legislation, regulatory frameworks, and open letters.

The goal is not merely to persuade you intellectually. The goal is to create a situation where agreeing with the argument and taking action become, psychologically, the same decision.

This is no accident. The people who build these sites understand that attention is scarce, that conviction fades quickly without outlet, that the difference between an informed reader and an engaged advocate is often just one button. And the design reflects it. You scroll through a scenario of imminent danger. You reach the end. There is no "congratulations, you finished reading." There is an invitation: Join us. Support us. Act now. Sometimes you are even presented with multiple pathways: slow, medium, urgent. But standing still is never one of them.

The AI microsite, then, is not just an argument. It is a mobilisation device. This fact is neither good nor bad in itself. It depends on the quality of the arguments, the soundness of the forecasts, the care of the normative commitments. But it changes the terms of engagement. The reader is no longer a distant observer. The reader is a latent activist – and the scroll is the activation sequence.

5. Speed versus certainty: the trade-off

One of the great strengths of the AI microsite is speed. It allows small teams to move faster than formal academic publishing cycles. It allows ideas to surface, circulate, and evolve in weeks, not years. It allows policymakers, journalists, and concerned citizens to engage with emerging threats before they crystallise into disasters.

But speed has a price. And the microsite makes that price unusually visible. Because so much of the content – particularly in AI safety – depends on forecasting rather than measurement, the microsite must ask the reader to act before full certainty is available. It must turn scenarios into strategies, incomplete evidence into mobilisation.

This is not a defect of the medium. It is a feature of the terrain. In fields like nuclear policy or pandemic response, we accept that decisions must often be made under conditions of radical uncertainty. The risk of waiting for complete information is often higher than the risk of acting prematurely. The microsite applies this same logic to AGI futures: better to anticipate, better to steer, better to build institutions and regulations while we still can.

But the compressed timeline – the seamless scroll from benchmark to forecast to policy prescription – carries dangers. It creates a false sense of epistemic homogeneity. Hard empirical claims and high-variance forecasts sit side-by-side in the same typographic style, the same muted palette, the same slow scroll. The reader is subtly encouraged to treat them with the same degree of certainty.

It risks premature consensus. The polished authority of the microsite aesthetic can make speculative futures feel already widely agreed upon as the default scenario. When in reality, profound uncertainties remain about timelines, capability thresholds, governance strategies.

It also raises problems of institutional memory. Microsites are lightweight, web-native, and editable or deletable at a whim. Unlike academic journals, they have no formal correction mechanisms, no preserved versions, no enduring record of what was claimed, when, and by whom. An idea launched in May 2025 may have vanished – or been quietly rewritten – by May 2026.

The AI microsite turns the strengths of grey literature – speed, influence, flexibility – into an even sharper tool. But the same qualities that give it power also create vulnerabilities in our collective reasoning about the future.

Microsites placed in context

While microsites are a distinct rhetorical innovation, they are part of a broader ecosystem of 'new-form' AI discourse — alongside video essays (e.g., Robert Miles’ YouTube series), podcasts (Dwarkesh Patel’s episodes on AI governance), and interactive demos (e.g., NVIDIA's AI Playground). Each format shapes the argument differently: the microsite's strength lies in its blending of narrative control, aesthetic authority, and mobilisation capacity.

6. The real microsite aesthetic: muted authority

At first glance, AI microsites feel almost self-effacing. Soft backgrounds. Spare typography. Linear layouts with generous white space. Compared to the noisy color palettes of Silicon Valley tech websites – or the cinematic gloom of dystopian sci-fi – they seem understated, even plain. But this minimalism is not neutral. It is doing serious rhetorical work.

• Pale colors: Most AI microsites favor off-white or very pale grey backgrounds. They reject the "dark mode" aesthetic common in tech culture. The effect is subtle but powerful: a pale canvas projects editorial seriousness. It evokes old books, government reports, the printed page. It reassures the reader: this is not hype; this is history unfolding.

• Restrained typography: Microsites almost universally use clean, modern sans-serif fonts for body text – often geometric or humanist in style. Some deploy serious serif headings to add a touch of old-world gravitas. Sizes are generous. Line spacing is ample. The goal is to create a feeling of oxygen – intellectual breathing space – rather than to cram or rush the eye. There is no font-shifting, no flashing animations, no frantic data visualisations. The consistency of type enforces the consistency of tone. And tone, in turn, stabilises trust.

• Carefully paced layout: Microsites almost never allow readers to browse laterally or jump freely. Navigation is vertical, controlled, chaptered. Each full-screen section presents one, and only one, idea at a time. This "one-thought-per-viewport" rule encourages cognitive absorption. The reader is rarely juggling multiple claims at once. Instead, they move through a sequence of carefully arranged moments, each building on top of the last. The cumulative effect is a feeling of inevitability, as arguments are walked, calmly and patiently, into the reader's mind.

Visual risk: the velvet hammer

The microsite aesthetic is a velvet hammer. It hits softly but leaves a deep impression. Readers must be careful: just because a site looks serious does not mean every claim within it is equally serious. Minimalism is not a substitute for epistemic caution.

7. Engineering inevitability: subtle narrative design

If the aesthetic of the microsite sets the tone, and its structure disciplines the reader's attention, its narrative techniques do something even more powerful: they make the future feel foreclosed. The microsite does not argue that one future is possible. It suggests that only a narrow band of futures remain.

It achieves this subtle determinism through three main techniques.

i) Scenario narrowing

Many microsites present a stark menu of futures: Slowdown or Race. Governance or Collapse. Deliberation or Disaster. These are framed not as speculative sketches, but as the last exits on a narrowing highway. The "choose your ending" panels found on several sites simulate agency – but the choice is binary, and heavily weighted by the preceding scroll.

There is rarely a third path, a muddle-through, a slow evolution, a partial success. The reader is offered radical success or radical failure, with nothing in between.

This narrowing performs two functions: It heightens urgency (action must be taken soon, before the choice disappears) and it simplifies coalition-building (it's easier to agree on a binary than to navigate shades of grey).

ii) Forecast curves as moral accelerants

Growth curves are a favorite motif: compute doubling curves, capability scaling curves, talent concentration curves. These curves are often presented cleanly, without heavy annotation. Their shapes – steep, inevitable, unsympathetic – do most of the rhetorical work. They suggest that the forces driving toward AGI or existential risk are physical, not political: a kind of natural phenomenon rather than a contested social choice.

This visual determinism discourages resistance. The future arrives like a train, not like a debate.

iii) Historical anchoring

Microsites frequently insert well-chosen historical echoes. Quotes from nuclear physicists confronting atomic inevitability. Timelines paralleling the speed of past technological revolutions. Allusions to 20th-century arms control efforts. These references frame today's choices as part of a longer human pattern – one where scientific discovery outruns governance, and tragedy looms when foresight fails.

The message is subtle but clear: We have seen this movie before. And unless we act, we know how it ends.

The microsite is a persuasion machine tuned for the age of exponential technologies. It does not shout. It does not plead. It gently and quietly engineers inevitability – until the reader no longer asks whether these futures are likely, but only how, and when, and what little might still be done.

8. Citation and academic mimicry

The AI microsite is not just a narrative structure or a visual language. It is also an epistemic performance. And like any good performance, it borrows its props from trusted traditions.

One of the most striking features of modern microsites is how carefully they mimic the apparatus of academic research: inline citations, often styled subtly like footnotes; full references sections, sometimes downloadable in BibTeX format; embedded datasets and model cards; occasional formal disclaimers about forecast uncertainty or methodological limits.

This academic mimicry serves multiple purposes. First, it signals seriousness. In a field crowded with speculation and hype, visible citation practices anchor the microsite in the intellectual traditions of rigour and reproducibility. It tells the reader: We know the rules. We respect them. You can trust us.

Second, it deflects criticism. By offering sources – even if many are blog posts, internal documents, or other grey literature – the microsite inoculates itself against accusations of pure opinion. It creates a plausible defense: We're not guessing. We're building on a structured body of thought.

Third, it enables quick academic uptake. By offering ready-made citation formats, the microsite smooths its own path into formal discourse. Researchers, policymakers, and journalists can reference microsite arguments almost as easily as they would cite a journal article.

Fourth, it frames policy advocacy as scholarship. When a microsite presents a legislative proposal alongside structured citations and forecast graphs, the line between political lobbying and academic contribution blurs. The advocacy is still there, urgent and deliberate, but it wears the calm mask of the research note.

Academic mimicry is not the same as academic rigour. Many microsite citations are preprints, blog posts, or internal documents. Forecasts are sometimes framed as empirical extrapolations when they are, in reality, structured guesses. Speculative scenarios sit alongside benchmark results in identical visual wrappers.

This is not a flaw unique to microsites: they meet a real need for faster, more agile policy thinking. But it places a burden on readers to distinguish between the appearance of rigour and its substance – to ask where evidence ends and assumption begins.

Who’s behind these AI microsites?

The visual uniformity of microsites suggests a shared rhetorical style, but the authorship spans a range of actors: academic research centres (e.g., Center for AI Safety), advocacy organisations (e.g., Apollo Research), and independent labs or collectives (e.g., Conjecture, FAR AI). Funding often comes from EA-adjacent philanthropies, corporate grants, or internal research budgets – a mix that both empowers agility and raises questions about transparency and agenda-setting.

9. Microsites, memory and media influence

Microsites occupy a precarious position in our intellectual infrastructure. They move fast, respond quickly to emerging challenges, and adapt to shifting evidence. But this same agility makes them fundamentally ephemeral. Unlike books, journals, or formal reports, microsites rarely have an archival plan. They live on rented web servers, under fragile domain names, reliant on Stripe accounts and GitHub deployments and whoever still remembers the password.

This ephemerality is not an accident. It reflects the same values that make microsites so powerful: agility, responsiveness, lightness. But it raises serious questions for intellectual memory. If microsites become a major way that key arguments about AI governance, regulation or existential risk are made, then the loss of those sites would mean the loss of primary historical evidence.

Yet even when microsites survive, their influence on real-world decision-making is uncertain. Rapid media amplification can create the illusion of impact –citations, headlines, fleeting political mentions – without delivering lasting institutional change. The speed with which microsites move through the media ecosystem contrasts sharply with the slowness of governance itself.

What did leading figures believe in 2025? How did arguments about model openness or compute monitoring evolve? Which forecasts proved prescient, and which fell away? Without stable archives, these questions will be harder to answer. And without a clear understanding of how microsites actually shape outcomes, the role they play in AI governance debates may be misunderstood by future scholars, policymakers, and citizens alike.

9.1 Fragility by design

Web-native grey literature is inherently precarious. Domains expire when renewal payments fail. Hosting platforms change their terms or shut down entirely. Third-party embeds – data visualisations, interactive models – rot as dependencies break. Sites are quietly edited or retracted without public notice, erasing their earlier claims and commitments.

Microsites are living documents, but living things die. And unlike academic publications or government reports, there is often no institutional commitment to preserve them beyond their immediate utility.

There are heroic efforts to combat this decay. The Internet Archive crawls millions of pages per day, creating imperfect snapshots of the web's constantly shifting surface. Individual researchers download and save critical reports, building personal archives of endangered knowledge. But these efforts are piecemeal and incomplete.

The emergence of projects like the AI Microsite Museum – a web-based archive curating influential AI safety microsites – reflects growing recognition of this fragility. By cataloguing these artifacts before they disappear, such efforts acknowledge that microsites are not just ephemeral advocacy tools, but part of the historical record of AI governance debates. Yet the Museum itself is lightweight, informal, and apparently reliant on public goodwill. It highlights both the cultural significance of microsites and the precarity of their preservation.

There is no systematic preservation infrastructure yet for AI-policy microsites. No formal library deposit system. No routine archiving at scale. For fields that may be defining the trajectory of human futures, this is a serious oversight. The arguments reshaping our technological landscape are being written in disappearing ink.

9.2 Memory and accountability

Archival fragility matters not just for historians, but for accountability. If ideas harden into policy – if microsite arguments help shape legislation, treaties, funding – then we will one day need to trace those chains of influence. Without preserved versions, without timestamped snapshots, it will be easier for bad faith actors to rewrite their own past arguments, to claim prescience after the fact, to disown failed forecasts without evidence.

The memory problem also creates serious vulnerabilities in regulatory and governance contexts. Without a reliable audit trail of claims, promises, and forecasts, it becomes difficult to hold organisations accountable for their prior positions. Did a lab predict its system would be safe? Did an advocacy group forecast specific harms? The answers may disappear with a domain renewal lapse or a quiet website update.

Good governance demands good memory. And good memory, in the age of the microsite, will not happen by accident. It will require deliberate effort: automated crawlers for at-risk policy domains; institutional archives that treat key microsites as first-class publications; norms of timestamped publication and version control.

We are building futures on scrollable sand. If we want to take the AI microsite seriously as an intellectual artifact, and as a force shaping real-world outcomes, then we must also take seriously the need to preserve it. What is agile enough to shape tomorrow must be stable enough to remember yesterday.

But even when microsites are preserved, their real-world influence is not guaranteed. To understand their impact, we must also examine how they circulate through media ecosystems – and where that circulation falls short.

9.3 Microsites, media, and the limits of influence

AI microsites are increasingly embedded within the feedback loops of mainstream and tech-focused media. A well-produced microsite can generate a predictable sequence: launch announcement on social media, rapid circulation among AI safety commentators, coverage in outlets like Forbes, New York Times or Noema, and occasional citations in political briefings or parliamentary questions. Microsites like Situational Awareness have become minor landmarks in the AI governance conversation, referenced as signals of growing elite concern.

Yet visibility does not equal impact. While microsites often succeed in setting the agenda within media narratives – defining which risks, timelines, and governance challenges are seen as salient – there is little evidence that they directly drive institutional change. Their arguments may shape how journalists frame AI issues, but the path from scrollable persuasion to policy adoption remains tenuous.

Situational Awareness, for instance, achieved significant media pickup across technology newsletters and policy commentary circles. But despite its careful construction of strategic risk scenarios, it has not yet translated into major shifts in national AI strategy documents or international governance frameworks. Microsites excel at compressing complex debates into accessible narratives, but legislation, regulation, and coalition-building still operate on slower, more deliberative timescales, responding to a broader set of pressures than media coverage alone.

The microsite, then, is best understood as a catalyst rather than a guarantee. It can accelerate awareness, concentrate attention, and provide ready-made frames for public and policy discourse. But it cannot substitute for the grinding work of institutional alignment, stakeholder negotiation, and political compromise. In the best cases, microsites seed the terrain for future action. In the worst, they offer the satisfying illusion of progress without its substance.

10. Closing thoughts: The double-edged future

The AI microsite is a new instrument in the machinery of serious discourse. It is fast, elegant, and persuasive. It collapses the distance between research and advocacy, between informing and mobilising, between narrative and action. It is a beautiful machine. And like all beautiful machines, it can be used for many ends.

Two futures suggest themselves.

10.1 The optimistic future: microsites as civic accelerators

In this world, the microsite becomes a tool for better democratic deliberation. It makes complex technical arguments accessible without dumbing them down. It helps policymakers, journalists, and concerned citizens engage with fast-moving fields at the speed those fields demand.

Microsites lower the barrier to participation. They allow small teams – often outside traditional institutional structures – to inject serious ideas into public discourse. They keep the AI governance conversation from being monopolised by the biggest labs, the richest governments, the loudest media brands. Teams with limited resources but important insights can command attention through thoughtful design and clear argumentation, rather than institutional positioning.

In this future, microsites are part of a broader civic literacy project: making technical futures legible, navigable, and crucially, debatable. They become not just persuasion engines but platforms for genuine exchange, spaces where different visions of the future can be articulated, compared, and evaluated. The scroll becomes not just a path to predetermined conclusions but a journey through genuinely open questions.

10.2 The pessimistic future: microsites as narrative weapons

In this world, the microsite becomes a tool for factional dominance. It is optimised not for clarity, but for emotional capture. It blends facts, forecasts, values, and speculations into seamless scrolls that demand action without serious reflection.

Microsites are deployed like campaigns, not conversations. Different groups build rival microsites, each framing the future in mutually incompatible ways. Policymakers are caught in a war of aesthetic persuasion, where the smoothest scroll wins. The medium's strengths – its visual coherence, its narrative momentum, its ability to simplify complexity – become its greatest weaknesses. The serious work of governance is reduced to competing spectacles of inevitability.

In this future, the microsite is no longer a tool for civic education. It is a weapon in a narrative arms race. Competing factions deploy scrollable manifestos, each engineered for maximum persuasion, minimum reflection. Serious, uncertain, high-stakes futures are flattened into aesthetic contests – where the smoothest scroll, not the soundest argument, wins.

The medium's strength becomes its greatest weakness: it captures attention faster than thought can catch up.

Call to action?

The microsite is here to stay. It will not replace books, or journals, or white papers. But it will shape how ideas move through the systems that govern us.

We will need new literacies not just of reading, but of pacing, aesthetic resistance, and epistemic humility: the ability to notice when facts turn into forecasts; the patience to interrogate urgent claims; the instinct to ask who built the scroll we are moving through, and why. We will need to train ourselves to recognise the subtle ways that form shapes content – how the clean white background, the carefully chosen typeface, the curated choice architecture all work to make some futures feel inevitable and others unimaginable.

In 2025, the serious conversation about AGI does not just happen in papers or parliaments. It happens on the screen, one scroll at a time. The future may be inevitable. But how we feel about it, and how well we reason our way through it, is still up to us.

The microsite offers us powerful tools for thinking together about complex, emerging challenges. It can help us move beyond the limitations of older forms – the slow pace of academic publishing, the bureaucratic caution of government reports, the sensationalism of mainstream coverage. But it will only serve us well if we approach it with the right kind of attention: alert to its persuasive patterns, aware of its epistemic shifts, and attentive to the silent work of its aesthetic choices.

The future will be scrolled through before it is lived. How we design that scroll – and how we read it – may shape what futures become possible.

Model Card

• Type: Narrative Microsite
• Capabilities: Medium Insight, High Self-Awareness, Moderate Persuasive Power
• Limitations: Speculative Forecasting, Epistemic Layer Blurring
• Training Data: Policy reports, microsites, public discourse, meta-analysis
• Intended Use: Encourage critical reading of scrollytelling artifacts
• Ethical Considerations: May induce mild archival anxiety
• Version: 1.0 (April 2025)
• Responsible Party. Ben Johnson. Follow me on X