AI is Aggregating the Aggregators

If I had to state the remit of this piece in one sentence, it is this: the shift from social platforms to AI interfaces changes who controls distribution and who captures the value created by online attention.

The common assumption is that social platforms still sit at the centre of the internet's attention economy. I think that assumption is already breaking. The more interesting question isn't which network wins the next format war. It's who becomes the primary interface between users and information.

That distinction matters because platform power has never really been about owning supply. It has been about aggregating demand, then using that position to discipline suppliers, absorb data, and capture the economics around discovery. Ben Thompson's Aggregation Theory gave the clearest name to that logic. What matters now is that the logic has not disappeared. It has moved.

I'm writing from a European vantage, so I read this less as a product story than as a governance story. Once aggregation becomes the control point, the owner of that layer gains influence over contracts, procurement, compliance, labour allocation, and regulation. That is why who controls AI isn't a philosophical question. It's an operational one.

The Locus of Power Is Shifting

For roughly the past fifteen years, social networking aggregation defined digital power. Platforms organised feeds, identities, and attention at scale. They became the route through which audiences discovered media, brands, employers, and one another.

That model now faces a higher-order aggregator.

Generative AI systems are increasingly positioned to sit above the platforms that once mediated everything below them. They don't merely organise social content. They ingest, rank, summarise, and re-present it through a new interface. The user no longer has to move across networks, publishers, and search results in the same way. The interface does the aggregation on the user's behalf.

Attention is moving up the stack

Many operators misread the shift. They still think in terms of channels. I think the battleground is now the interface layer itself.

If an AI system becomes the first place a user asks a question, seeks a recommendation, or requests a summary, the value of the underlying platform changes. The platform still hosts content and social behaviour. But it loses some control over discovery, context, and monetisation.

Social networking aggregation is no longer only about pulling many feeds into one dashboard. It is about who sits between human intent and the underlying pool of content.

That change also alters the policy frame. A network that aggregates user attention is already powerful. An AI layer that aggregates across networks, the web, and synthetic outputs at once has broader reach and fewer visible boundaries. It can mediate news, commerce, labour signals, and reputational risk through a single interface.

The old gatekeepers are becoming suppliers

The strategic implication is simple. Yesterday's aggregators can become today's upstream inputs.

One distinction should be drawn up front. Social platforms keep things an AI interface does not replace: the pull of the feed, passive consumption, entertainment that doesn't start from a question, and the most obvious function of all — the social network itself, the messages, the groups, the relationships. Even as feeds visibly drift toward passive, TV-like entertainment, that share of attention stays where it is. What moves is intent-driven attention — an answer, a recommendation, a decision — which is also the most monetisable share. The platforms risk keeping the time and losing the moments that matter most.

That's why I don't see the current moment as a routine platform transition. I see a transfer of bargaining power. If AI systems become the preferred route to information, social platforms may keep scale while losing primacy. They become part of the supply base for a new class of intermediary.

How Social Aggregation Centralised Power

Social platforms won the last era of the internet by aggregating demand before anyone else could. Through an Aggregation Theory lens, their advantage was never just scale. It was control over the point where users discovered information, evaluated relevance, and decided what to do next.

That control changed the economics of everyone upstream. Creators, publishers, brands, and merchants supplied the content and commercial inventory. The platform organised attention around it and captured the highest-value position in the chain.

Social usage long ago reached population scale and spread across multiple services per user. That mattered because aggregation gets stronger as user behaviour fragments. The more accounts, feeds, and interactions people had to manage, the more valuable a central interface became. Convenience looked like a product feature. In practice, it became a mechanism for consolidating power.

Demand concentration came first

Users rarely choose complexity. They choose the interface that reduces search costs, compresses options, and gives fast social feedback. A unified identity, personalised feed, and low-friction discovery loop pulled demand into a few large platforms, even though the underlying supply remained widely distributed.

Once demand concentrated, dependence followed. Suppliers had to publish where audiences already were, then adapt to ranking systems, policy changes, and monetisation rules they did not set. The platform did not need to produce the content itself. It only needed to intermediate discovery at scale.

Three power shifts followed:

• Distribution control: visibility depended on ranking, recommendation, and moderation systems owned by the platform.
• Behavioural visibility: the platform could observe engagement patterns across formats, communities, and transactions.
• Monetisation control: advertisers and merchants paid to reach attention that had already been aggregated and segmented.

Aggregation centralises demand first, then turns that concentration into bargaining power over supply.

Suppliers became substitutable inside the feed

Individual suppliers still differed in quality, brand, and audience fit. Inside the aggregator's interface, those differences mattered less than many operators assumed. A post, product, or article entered a ranking system as one candidate among many. Abundance at the supply layer increased the platform's power to sort, prioritise, and price access.

That is why social aggregation did more than organise information. It weakened the negotiating position of the people and firms producing it. If discovery happens inside a feed controlled by someone else, distinctiveness does not disappear, but it competes under rules designed to preserve the aggregator's advantage.

The same pattern now appears in a more advanced form. As I argued in my analysis of how direct-answer interfaces changed the economics of traffic, once the interface satisfies intent directly, the supplier loses both the visit and the chance to shape context. In strategic terms, that is the endpoint of aggregation. The winning layer is the one that captures the query, resolves it, and leaves upstream providers interchangeable enough to bargain down.

Practical rule: the layer that controls discovery can recast differentiated suppliers as replaceable inputs.

The Architectures of Aggregation

Aggregation is an architecture of control before it is a product category. The important question is not whether a system combines social inputs. It is where that combination happens, what gets standardised in the process, and who gains the right to mediate attention once many sources are reduced to one interface.

Under an aggregation lens, four architectures recur because each concentrates a different bottleneck:

Four models of social networking aggregation

• Feed aggregation — unify posts from multiple networks into one operational view. Example: a brand monitoring wall combining several social feeds.
• Meta-platform aggregation — keep users inside one identity and service environment. Example: a large platform linking messaging, social identity, and content distribution.
• Social login and graph aggregation — extend platform identity across third-party services. Example: a single sign-on layer that carries profile and relationship data outward.
• Cross-platform indexing — rank and retrieve social content through a broader discovery layer. Example: a search-style interface surfacing content from multiple networks.

These models may look operationally distinct, but strategically they solve the same problem. They collapse fragmentation into a governed point of access. Once that access point becomes habitual, the aggregator gains more than convenience value. It gains the ability to define defaults.

Identity resolution turns aggregation into institutional power

Every aggregation system has to answer a hard question. Do two records from different contexts refer to the same person, organisation, or relationship?

That decision governs far more than database hygiene. It determines whether an operator can merge audiences, infer intent across contexts, suppress duplicates, and present a single version of the user to advertisers, publishers, employers, or public institutions. In practice, identity resolution is the layer where social aggregation stops being a dashboard feature and becomes a political technology.

Earlier research on modelling social network data framed this in ontology terms, with profiles and relationships expressed through formal structures rather than loose text labels. The strategic implication is easy to miss. If identity is formalised inside the aggregator's model, the model starts acting as the reference point for reality. Whoever defines the schema gets disproportionate influence over what counts as the same person, the same community, or the same signal.

That is why aggregation often strengthens the intermediary even when the underlying content remains distributed.

Infrastructure determines who becomes the trusted interface

Identity alone does not secure power. The aggregator also needs to become operationally reliable enough that organisations act through it rather than merely observe from it.

That depends on mundane technical choices with large strategic effects:

• Connection registries determine where users, devices, and sessions are currently attached.
• Pub/sub backbones determine how quickly events propagate across the system.
• Client reconnection logic determines whether the service remains dependable under ordinary mobile instability.
• Normalisation layers determine how incompatible formats become one comparable stream.

Once those layers work well enough, the aggregator becomes the practical control room. Teams stop checking underlying sources directly because the aggregated layer is faster, tidier, and easier to act on. That behavioural shift matters more than the interface itself. It trains institutions to trust the summary over the source.

A second-order effect follows. The more decision-making moves to the aggregated layer, the more value accrues to the actor that can charge for access, ranking, priority ingestion, or preferred visibility. The same economic logic now appears in AI retrieval systems, which is why the debate is shifting toward who should pay for machine access to source material, as explored in this analysis of the pay-per-crawl model for AI traffic monetisation.

The architecture points beyond social platforms

This is the structural bridge to the next phase of aggregation. Social platforms won power by centralising interaction and discovery inside feeds. New AI systems are positioned to sit one layer above that stack by aggregating not just posts or profiles, but the outputs of many platforms at once.

Seen through Aggregation Theory, the shift is stark. The winning layer is no longer necessarily the network that hosts the conversation. It may be the system that resolves intent, synthesises answers, and routes attention without sending the user back to the original social environment.

In that world, social networking aggregation looks less like the endpoint and more like the training ground for a larger reordering of internet power.

The Political Economy of an Aggregated World

Aggregation does not just simplify the user experience. It reallocates power.

Under Aggregation Theory, the winning layer is the one that controls demand while pushing supply into competition. Social aggregation did exactly that. It pulled fragmented conversations, identities, and signals into a single decision surface, then made that surface the place where attention, ranking, and visibility were allocated. Once that happened, the interface stopped being a neutral utility. It became a gatekeeper over who gets seen, cited, trusted, and paid.

The B+ Trap in aggregation

My own framework, the B+ Trap, helps explain why aggregated systems gain institutional authority so quickly. Their outputs are usually good enough to support action. They are clean, comparable, and easy to circulate across teams. That surface quality creates a dangerous incentive. Executives, regulators, and analysts start treating legibility as completeness.

A key weakness is not random error. It is patterned omission. Aggregated systems tend to overrepresent sources that are easy to ingest, classify, and rank, while underrepresenting activity that sits in low-resource languages, fragmented communities, private groups, or poorly structured networks. The result is a dataset that looks decision-ready while filtering out the edges where political risk, reputational shifts, and market change often begin.

An orderly view can still be strategically wrong.

Incomplete visibility becomes a governance problem

Once institutions operationalise an aggregated view, the costs spread beyond analytics. A firm may misread demand because informal networks are missing. A regulator may target the most measurable actors rather than the most influential ones. A compliance team may inherit provenance gaps precisely where verification matters most.

Three effects matter most:

• Market power concentrates: visibility flows to actors already legible to the aggregation layer, which reinforces incumbency.
• Risk assessment degrades: weak source context makes it harder to distinguish primary evidence from recirculated noise.
• Policy follows the dataset: governance starts to track what systems can capture, not what publics do.

This is the political economy behind current fights over access, ingestion, and compensation. If an intermediary captures user demand while stripping away source context, upstream producers lose distribution power first and pricing power second. That is why debates over pay-per-crawl models for AI traffic monetization matter far beyond publishing. They are early negotiations over who gets paid when the aggregation layer becomes the primary point of access to information.

The same logic now affects brands and institutions that want to remain visible when users stop clicking through to original sources.

Enter the Meta-Aggregator How AI Reshuffles Power

The next power shift is not from one social platform to another. It is from platforms that organise attention to systems that resolve intent.

Generative AI now operates at consumer scale. That matters less because it adds another content format, and more because it inserts a new intermediary between users and the feeds, creators, and publishers that previously captured demand directly. Through an aggregation-theory lens, this is the key change. The highest-value position on the internet belongs to the service that owns user demand and makes suppliers interchangeable.

AI becomes the interface, not just the feature

Social aggregation centralised power by pulling many voices into one feed. AI answer systems go a step further. They aggregate the aggregators.

The mechanism is straightforward. These systems ingest material from multiple upstream sources, compress source context, synthesise a response, and capture the interaction at the answer layer. Users no longer need to visit the original environment to get what they came for. In strategic terms, that shifts power away from hosting and ranking content, and toward interpreting it.

That distinction matters because interpretation is stickier than distribution. A feed still asks the user to sort, compare, and decide. An answer engine performs that labour on the user's behalf. Once that behaviour becomes habitual, social platforms risk becoming upstream inventory for a downstream interface they do not control.

The value shifts from hosting to answering

This is why AI should be understood as a new aggregation layer, not just a product enhancement. The user relationship moves to the system that converts scattered inputs into a usable conclusion.

The commercial effects follow quickly. Discovery happens inside the answer. Attribution weakens because source selection is folded into model output. Brand visibility depends less on placement in a feed and more on whether a system retrieves, trusts, and cites you at all. That is why operators are starting to treat visibility inside answer engines as a distribution problem, not a narrow optimisation tactic.

The deeper point is structural. Social networks won by aggregating users and suppliers in one place. AI systems can win by aggregating the social networks, the open web, and proprietary datasets into a single response surface. If that model holds, the dominant interface no longer needs to own the content environment. It only needs to own the moment of resolution.

Concentration risk rises again

That raises the stakes on concentration. A small group of firms already controls a large share of the inputs that matter most for AI: compute, capital, training data, and specialised talent. If those same firms become the default answer layer, they gain power at two levels at once. They shape both what information is available for inference and how that information is interpreted for end users.

This is a sharper form of aggregation than the social era produced. Social platforms mediated visibility. Meta-aggregators mediate judgment. Once a system decides which sources count, how they are combined, and what final answer is shown, it does more than route attention. It sets the terms under which knowledge is encountered.

Navigating the New Aggregation Layer

The next contest for digital power will be decided less by who hosts content than by who governs the answer layer. For operators, that shifts AI aggregation out of the marketing budget and into procurement, compliance, and policy.

The practical problem starts with dependency. Many organisations now rely on third-party systems to retrieve, rank, summarise, and present information they did not create and cannot inspect in full. Under an aggregation model, that dependence is not a feature add-on. It is a transfer of decision rights over visibility, attribution, and interpretation.

Procurement leaders need tighter allocation of liability

Contracting has not caught up with that shift. Many agreements still treat aggregated output as a low-risk information service, even when it informs customer communications, internal operations, or regulated decisions. That leaves buyers carrying risks they did not price properly, especially around provenance, copyright, defamation, synthetic content, and factual misrepresentation.

A stronger procurement posture starts with clear allocation on three points:

• Source provenance: contracts should specify what the provider can verify, what remains unverifiable, and how synthetic or transformed material is labelled.
• Liability boundaries: exposure tied to copyright, defamation, and false or misleading output should be assigned explicitly rather than pushed downstream by default.
• Audit rights: buyers need a practical way to test claims about data handling, inclusion criteria, and output quality where decisions rely on the system.

For smaller firms, this matters even more. They often adopt aggregation layers to extend capacity, but weak contract language can reverse that advantage by shifting legal and operational uncertainty onto the buyer.

Compliance teams should treat aggregation as data stewardship

Governance also changes at the point where systems turn many inputs into one actionable output. Social aggregation once raised questions about lawful collection and platform moderation. AI aggregation adds a harder question. Is the representation reliable enough to support a decision?

That standard is higher than basic data hygiene. A compliant process needs to account for missing context, source dilution, model error, and the possibility that synthetic material shaped the output without clear disclosure. If an organisation cannot explain where a claim originated, how it was transformed, and why it appeared credible, it will struggle to defend any decision built on top of it.

Treat aggregated AI output as a contested input, not as a neutral fact pattern.

Aggregation Theory offers a useful operating lens. In social media, the platform captured value by concentrating demand and controlling discovery. In the AI layer, the aggregator captures more. It can compress multiple sources into a single answer and remove the user's need to visit the underlying suppliers at all. Compliance teams should read that as a shift from distribution risk to epistemic risk.

Policymakers should regulate the control point

Policy analysis has to follow the new bottleneck. Regulating only the visible content platform misses the actor that increasingly determines what users see, what sources are weighted, and which claims appear settled.

The sharper question is who controls the means of prediction and the means of access. That framing aligns with this work on democratic control of AI's predictive infrastructure, which argues that optimisation systems sit inside societies with competing goals and therefore require public oversight of algorithms, data, and compute.

That has direct consequences for competition policy. If a small group of firms becomes the default answer layer, market power no longer rests only on audience aggregation or hosting scale. It rests on the capacity to intermediate judgment across the web, including across the social networks that previously dominated attention. That is a more durable position because it sits one layer above the feed.

If this section's argument holds, strategy changes too. The task is no longer just to publish, distribute, and optimise for reach. It is to remain legible to systems that decide what counts as a credible answer.

If you want analysis like this each week, grounded in markets, contracts, regulation, and labour rather than AI theatre, subscribe to ELECTE's Newsletter.

Sources

• The political economy of AI, towards democratic control of the means of prediction

Fabio Lauria

CEO & Founder, ELECTE

Every week, we explore AI without the hype — using data, analysis and an independent perspective.