The social internet was built on a promise: connect with the people you know. That promise has been quietly violated, and the data trail is unambiguous.

In October 2025, Mark Zuckerberg described the trajectory in twelve words that should concern every marketing strategist alive: "Social media started out as people primarily interacting with their friends. And now, at least half of the content is basically people interacting with creators." He said this not as a confession but as an observation — a matter-of-fact description of a platform that had, over fifteen years, systematically replaced its core product. The product was human connection. The replacement was optimised attention.

This is not a cultural commentary. It is a structural shift in how human beings encounter brands, form preferences, and make purchase decisions. And it has profound, measurable consequences for anyone responsible for marketing strategy.

The shift Zuckerberg described did not happen because users chose creators over friends. It happened because recommendation algorithms tested friends against strangers and measured the result. The stranger with better lighting, better timing, and a more compelling hook held attention three seconds longer. The algorithm registered the difference. It served more of what held attention. It served less of what did not. Over millions of iterations, across billions of sessions, the feed transformed — not by editorial decision, but by optimisation pressure.

This article is the seventh in the Catchlight series. Article 1 established visibility as the precondition. Article 2 showed that AI systems carry their own biases about which brands to surface. Article 3 revealed that viewable does not mean viewed. Article 4 dismantled the persuasion model in favour of behavioural brand choice. Article 5 showed that mental availability determines which brands surface. Article 6 demonstrated that the time horizon of investment determines whether effects build or erode.

This article examines a different dimension: what happens when the infrastructure through which brands achieve visibility — the social feed itself — is restructured around a logic that systematically displaces human connection in favour of algorithmically optimised content. And what happens when artificial intelligence enters this system as a content producer, not merely a content distributor.

The implications for brand strategy are not speculative. They are already measurable.

The first substitution: friends for strangers

The academic evidence for what happened to social feeds is now substantial. In 2023, a research team led by Andrew M. Guess at Dartmouth College, in collaboration with researchers at the University of Chicago and New York University, conducted one of the largest controlled experiments ever run on a social media platform. Working directly with Meta, they randomly assigned Facebook users to either an algorithmic feed or a reverse-chronological feed during the 2020 US election period. The study, published in *Science*, measured the composition of what users actually saw.

The finding was stark: under algorithmic conditions, the relative share of content from friends and family decreased by an average of 24 percent compared to chronological conditions. The algorithm did not remove friend content. It deprioritised it. It allocated feed real estate to content that generated higher engagement — and that content was disproportionately from pages, groups, and public figures rather than from personal connections.

The engagement metrics confirmed the mechanism. Users in the algorithmic group liked an average of 6.7 percent of the content exposed to them, compared to 3.1 percent in the chronological group. The algorithm was doing exactly what it was designed to do: maximising engagement. The cost was that the feed ceased to be a social tool and became an attention marketplace.

This was not a glitch. It was an optimisation outcome. Facebook's own internal research, portions of which were disclosed during the Frances Haugen testimony in 2021, confirmed that the platform's ranking system consistently amplified content that triggered strong emotional reactions — particularly anger and outrage — because these emotions generated more comments, more shares, and longer session durations. The meaningful social interactions framework that Facebook introduced in 2018, which was explicitly designed to reweight the algorithm toward friend content, inadvertently amplified the most divisive posts because the metric it optimised — comment volume — correlated with controversy, not connection.

By 2025, the trajectory had accelerated. Meta's own disclosures indicate that more than 40 percent of content in the average Facebook and Instagram feed is now recommended by AI from accounts the user does not follow. The platform has adopted what is functionally TikTok's model: a discovery engine where the social graph is an input to the algorithm, not the organising principle of the feed. The distinction matters enormously. In a social graph model, the user's relationships determine what they see. In a recommendation model, the algorithm's engagement predictions determine what they see. The user's relationships become one signal among many — and not the strongest one.

For brands, this transformation has been catastrophic in ways that are only now becoming fully visible. An analysis of 15,000 social media profiles conducted over three years found that organic social media reach declined by 61.83 percent across the period studied. Posts linking to external websites — the primary mechanism through which brands drive traffic from social platforms — saw reach penalties increase from 74.51 percent to 219.49 percent. The platforms are not merely deprioritising brand content. They are actively penalising any content that attempts to move attention off-platform.

The first substitution is complete. The feed that was built to show you your friends now shows you strangers. And the strangers are winning because they are optimised for the metric that the platform cares about: time spent.

The attention arms race: Why creators cannot win

The creator economy that emerged from this substitution has been celebrated as a democratisation of media. Anyone with a phone can build an audience. The barrier to entry has collapsed. What is less discussed — and what the attention data makes unavoidable — is that the barrier to sustained attention has risen exponentially, and the humans producing content are caught between an algorithm that demands more and an audience whose capacity to give more is physiologically contracting.

Research on the cognitive effects of algorithmic content consumption, including a meta-analysis of nearly 100,000 participants, has found that heavy consumption of short-form video content is associated with measurable declines in sustained attention, inhibitory control, and working memory. The mechanism is neurological: short-form content delivers frequent dopamine rewards that reinforce compulsive scrolling behaviour while simultaneously training the brain away from the sustained focus required for deeper engagement. Gloria Mark, professor of informatics at the University of California, Irvine, has documented a collapse in average attention span on screens from approximately 2.5 minutes in 2004 to 47 seconds by 2023. The trajectory has not reversed. It has accelerated.

The platforms have responded not by addressing the attention crisis but by adapting their content formats to it. The average length of short-form video content has increased from 13 seconds to 44 seconds over the past three years — not because audiences want longer content, but because platforms are attempting to increase session duration while the underlying attention capacity of users continues to fragment. The result is a paradox: content is getting longer while the capacity to attend to any single piece of it is getting shorter. The algorithm resolves this paradox by testing millions of variations per second, identifying the precise combination of thumbnail, opening frame, pacing, and hook that holds a particular user for the maximum duration. No human creator can run this optimisation loop against themselves.

This creates an arms race that human creators cannot sustainably win. The algorithm rewards content that captures attention within the first one to two seconds and sustains it through carefully calibrated pacing, visual novelty, and emotional triggers. Meeting these requirements consistently demands a production cadence that is incompatible with human creative capacity. A creator who posted weekly two years ago must now post daily to maintain algorithmic visibility — and each post must clear a higher engagement threshold than the last because the baseline is being set by millions of competing creators, all optimising for the same signals. The creator is not competing against other creators. The creator is competing against the algorithm's expectation of what perfect content looks like, an expectation that updates continuously and never resets.

The human cost is documented. Creator burnout is not a psychological curiosity. It is an economic inevitability produced by a system that treats human output as a commodity and measures it against a production standard that scales without constraint. The algorithm does not know or care that the person behind the content slept four hours, missed a family dinner, or is producing their three-hundredth video on the same topic. It measures engagement. When engagement falls, visibility falls. When visibility falls, income falls. The incentive structure is a ratchet: it only tightens.

The data from the 2025 "How Humans Decide" study conducted by WPP Media and the Said Business School at Oxford provides a critical marketing dimension to this dynamic. Despite the enormous reach of creator content, influencer recommendations have only an 8 percent probability of actively influencing a purchase decision. Reviews, by contrast, carry a 36 percent probability. Word of mouth — genuine, unmediated personal recommendation — remains among the top drivers of brand priming. The gap reveals something important: algorithmically optimised content is extraordinarily effective at capturing attention but comparatively weak at converting that attention into the trust structures that drive purchase behaviour.

Algorithmically optimised creator content captures attention at scale. It converts that attention into trust and purchase influence at a fraction of the rate of genuine human recommendation.

The attention metric and the influence metric have diverged. A creator can accumulate millions of views while generating almost no durable impact on the purchase behaviour of the audience. The algorithm counts the view. The brand counts on the conversion. The two are measuring different things, and the gap between them is the structural vulnerability that the second substitution will exploit.

The second substitution: Strangers for machines

If the first substitution replaced friends with algorithmically optimised strangers, the second substitution — already underway — replaces those strangers with algorithmically generated content that no human produced.

By November 2024, the volume of AI-generated articles published on the internet surpassed the volume of human-written articles for the first time. In the twelve months following the launch of ChatGPT, AI-generated articles accounted for 39 percent of new content published online. Meta disclosed in its Q3 2025 earnings call that its AI image generation tools had produced over 20 billion images within its platforms. Zuckerberg explicitly described the emerging "third era" of social media: the first era was content from friends; the second added creator content; the third adds AI-generated content as a primary feed component.

The economics of this substitution are as decisive as those of the first. A human creator needs time, motivation, equipment, lighting, and — crucially — lived experience to produce content. An AI system needs a prompt and electricity. When the marginal cost of producing engaging content approaches zero, the feed fills with content that is indistinguishable from human creation but produced at a volume and consistency that no human can match.

Research from MIT Sloan, led by Yunhao Zhang and Renee Richardson Gosline, provides the critical perceptual finding: when participants did not know whether content was produced by a human or by AI, they preferred the AI-generated content. It was only when participants were told the source that they exhibited what the researchers termed "human favouritism" — rating the human content higher. The preference for human content is real but conditional on disclosure. In an algorithmic feed where source attribution is minimal and content volume is overwhelming, the condition for human favouritism is rarely met.

This means the feed is evolving toward a state where the content that captures and holds attention is increasingly produced by systems that have no creative limitations, no burnout, no bad days, and no cost constraints. Every word calibrated. Every frame tuned. Every pause placed at the exact interval that maximises retention. A human creator competing against that is operating under constraints that the algorithm does not value and will not reward.

The parallel to brand strategy is direct and urgent. If AI-generated content floods the attention marketplace, the cost of attention declines for those who deploy AI and increases for those who do not. Brands that rely on organic creator partnerships will face the same competitive dynamics that individual creators face: an escalating production requirement driven by an algorithmic baseline that AI-generated content sets and humans cannot sustainably match.

Consider the economics from a CMO's perspective. A brand currently invests in creator partnerships because creators provide authentic, human-produced content that audiences trust. But the MIT Sloan finding introduces a complication: audiences do not actually distinguish AI content from human content in the absence of disclosure. The authenticity premium exists only when the audience knows the content is human. In a feed where content sources are ambiguous, the authenticity premium evaporates. What remains is the cost differential: a human creator who charges tens of thousands per post versus an AI system that produces equivalent-performing content for the cost of computation.

This does not mean creator partnerships become worthless. It means their value migrates. The value of a creator will increasingly reside not in their ability to produce content — AI will commoditise that — but in their ability to lend genuine human credibility, personal narrative, and social proof to a brand relationship. The creator becomes valuable for who they are, not for what they produce. The distinction is critical, and brands that fail to make it will find themselves paying human rates for work that a machine can do, while neglecting the genuinely human qualities that no machine can replicate.

The visibility paradox: more content, less connection

The paradox of the algorithmic substitution is that it has created more content, more reach, and more engagement than at any point in the history of media — while simultaneously reducing the quality of the connections that content creates.

This paradox is visible in the marketing effectiveness data. The IPA's campaign analysis consistently shows that approximately 80 percent of campaigns achieve their reach potential — they are seen by the intended audience at the intended frequency. But only 25 percent of campaigns achieve their effectiveness potential — they actually shift brand perceptions, build mental availability, or generate incremental sales. The gap between reach and effectiveness has widened precisely during the period when algorithmic distribution made reach easier to achieve.

The 2025 WPP Media research quantified this gap through the lens of creative quality, documenting a measurable trend toward "increasing salesmanship and decreasing showmanship" in advertising between 2019 and 2024. Brands are producing more content, achieving more impressions, and generating less impact per impression. The algorithm rewards volume and engagement signals. It does not reward the depth of connection, the distinctiveness of positioning, or the memorability of creative execution — the qualities that drive long-term brand building.

The research on brand priming reinforces this analysis. The "How Humans Decide" study found that 84 percent of purchases go to brands that consumers are already positively biased toward before the purchase decision begins. This bias is built through sustained, distinctive visibility over time — the kind of visibility that requires consistency, recognisability, and emotional resonance. It is not built through algorithmically optimised content that captures three seconds of attention before the thumb moves on.

The attention metric and the brand-building metric have become decoupled. A brand can achieve massive reach through algorithmic distribution and generate almost no durable mental availability. The feed delivers the impression. The algorithm counts the engagement. The dashboard reports success. But the consumer's memory — the actual substrate of brand value — registers nothing distinctive enough to encode.

Article 3 in this series explored the distinction between viewable and viewed — the gap between an ad that could technically be seen and an ad that actually captured human attention. The algorithmic substitution introduces a third layer to this gap: an ad that captures attention but fails to create distinctive memory encoding. Mike Follett's work at Lumen Research demonstrated that a 30-second television ad generates approximately 13.8 seconds of actual attention. An Instagram Stories ad generates 1.7 seconds. The algorithmic feed is optimised for the shorter format. But the shorter format generates attention that is shallow, transient, and undifferentiated. It does not carry the salience or emotional weight required to build the category entry points that Jenni Romaniuk's work at the Ehrenberg-Bass Institute has shown drive brand growth.

The paradox deepens when you consider the competitive dynamics. In an algorithmically sorted feed, every brand's content competes against every other brand's content — and against every creator, every meme, every AI-generated piece of visual entertainment — for the same finite attention. The cost of standing out is not measured in media spend. It is measured in distinctiveness: the degree to which a brand's creative assets are recognisable, memorable, and uniquely linked to that brand within the first fraction of a second of exposure. In the 0.01 percent of stimuli that penetrate conscious awareness — the figure from Unravel Neuromarketing's Brand Asset Guide that Article 1 examined — the only brands that register are those with distinctive assets strong enough to survive the algorithmic noise.

This is the visibility paradox: in the most content-saturated environment in human history, the brands that invest most heavily in algorithmic content production may be the ones building the least durable competitive advantage. They are winning the engagement metric while losing the memory metric. And in the long run — as Article 6 demonstrated — it is the memory metric that determines market share.

The strategic imperative: distinctiveness over distribution

The series through-line arrives at a point of convergence. Visibility is the strategy — but not all visibility is equal. Attention is the currency — but not all attention converts to memory. Mental availability determines purchase — but mental availability is built through distinctive, sustained, emotionally resonant brand exposure, not through volume-optimised content production.

The algorithmic substitution has created an environment where distribution is effectively free but distinctiveness is more expensive and more valuable than ever. When every brand can flood the feed with algorithmically optimised content — and soon, with AI-generated content that costs almost nothing to produce — the differentiator is no longer who reaches the most people. It is who creates the associations that survive the scroll.

The research on distinctive brand assets, which Article 1 explored through the lens of the Ehrenberg-Bass Institute's Report 52, becomes more critical in this context, not less. Distinctive assets — visual, auditory, and sensory cues that are uniquely and instantly linked to a brand — function as a filter against algorithmic noise. They are the reason a consumer recognises a brand in the 0.01 percent of stimuli that penetrate conscious awareness. In a feed dominated by AI-generated content that is optimised for engagement but not for any specific brand's memory structures, distinctive assets are the only mechanism through which a brand can convert attention into recall.

The strategic response is not to produce more content. It is to produce more distinctive content. It is not to optimise for the algorithm's engagement signals. It is to optimise for the consumer's memory structures. It is not to compete on volume — a competition that AI will always win — but to compete on recognisability, emotional resonance, and the kind of sustained visibility that builds the category entry points through which brands are recalled at the moment of purchase.

The feed has changed. The algorithm replaced friends with strangers, and it will replace strangers with machines. What has not changed is the mechanism through which brands grow: being easy to think of and easy to find when the purchase moment arrives. The infrastructure of attention has been restructured. The science of mental availability has not.

The brands that understand this distinction will invest in what the algorithm cannot replicate: human creativity deployed in service of distinctive, memorable brand building. They will measure success not by engagement rates — which AI content will always win — but by mental availability metrics, unaided brand recall, and the category entry points through which their brand is accessed in real purchase moments. They will treat the feed as one channel among many, not as the centre of their strategy. And they will recognise that television, outdoor advertising, and other broad-reach media — the channels that build the 84 percent pre-decision bias documented in the WPP research — are more strategically important in an AI-saturated feed environment, not less.

The brands that do not understand this will optimise for a feed that moves faster every quarter, producing more content that is seen by more people and remembered by fewer. They will celebrate reach metrics while their mental availability erodes. They will invest in a content volume race against a competitor — artificial intelligence — that has no cost floor, no creative fatigue, and no strategic intent.

Somewhere in the scroll, between a cooking video from Dubai and an AI-generated face that feels familiar, there is a brand that has been present long enough and distinctive enough to be recalled when it matters. That brand is not the one that posted most often. It is the one that was impossible to forget.

References

• Guess, A. M., Malhotra, N., Pan, J., Barberá, P., Allcott, H., Brown, T., ... & Tucker, J. A. (2023). "How do social media feed algorithms affect attitudes and behavior in an election campaign?" *Science*, 381(6656), 398-404.
• Zuckerberg, M. (2025). Meta Q3 2025 Earnings Call Transcript. Meta Platforms, Inc.
• WPP Media & Said Business School, University of Oxford. (2025). *How Humans Decide: Consumer choice, brand availability, and media effectiveness.* WPP Media.
• Zhang, Y., & Gosline, R. R. (2023). "Human Favoritism, Not AI Aversion: People Prefer Human-Generated Content When Source Is Disclosed." MIT Sloan Working Paper.
• Haugen, F. (2021). Testimony before the US Senate Committee on Commerce, Science, and Transportation. "Protecting Kids Online."
• Wittenberg, C., Epstein, Z., Berinsky, A. J., & Rand, D. G. (2024). "Labeling AI-Generated Content: Promises, Perils, and Future Directions." MIT.
• Aggarwal, P., & Murahari, V. (2024). "GEO: Generative Engine Optimization." Princeton University.
• Romaniuk, J., & Sharp, B. (2016). *How Brands Grow Part 2.* Oxford University Press.
• Ehrenberg-Bass Institute. (2018). *Report 52: Building Distinctive Brand Assets.* University of South Australia.
• Binet, L., & Field, P. (2013). *The Long and the Short of It.* IPA.
• Mark, G. (2023). *Attention Span: A Groundbreaking Way to Restore Balance, Happiness, and Productivity.* Hanover Square Press.

Unsubscribe