Jevons Paradox

• Google's AI Overviews Quality Degradation and Remediation Cycle — Google's AI Overviews feature, rolled out at scale through 2025, has produced documented instances of low-quality, hallucinated, or slop-aggregated summaries surfacing as authoritative answers. Google has been in an active remediation cycle through Q1–Q2 2026, deploying quality classifiers and human review pipelines on top of its generative layer. The significance for competitive positioning: Google's core search moat — trusted information retrieval — is being structurally tested by the very AI it deployed to defend against OpenAI's encroachment. This is a live, real-time demonstration of Jevons dynamics in the information retrieval market. Timeline: Ongoing remediation; next major quality milestone expected at Google I/O follow-through, Q3 2026.

• Perplexity AI's "Pages" and Citation Infrastructure Scaling — Perplexity AI has continued scaling its citation-anchored answer engine, positioning its provenance-linking architecture as a direct counter to the slop problem. As of Q2 2026, Perplexity reports tens of millions of monthly active users and has been in active discussions with publishers (including News Corp and others) around licensing frameworks. The moat thesis here is that citation infrastructure — not raw generation — becomes the defensible layer when content abundance renders unattributed generation toxic. Timeline: Licensing framework outcomes expected H2 2026.

• OpenAI's "Memory" and Personalized Filtering Expansion in ChatGPT — OpenAI has been expanding persistent memory and user-preference modeling in ChatGPT through Q1–Q2 2026, effectively building a personalized noise-filtering layer on top of raw generation. This is structurally significant: the entity generating the content is also building the curation apparatus, creating a potential vertical integration of the signal/noise stack. This directly threatens third-party curation middleware players. Timeline: Memory features in broad rollout as of Q2 2026; enterprise memory API expansion anticipated Q3–Q4 2026.

• The Rise of "Vibe Curation" and AI-Assisted Editorial Tools at Scale — Substack, Beehiiv, and Ghost have each introduced or are piloting AI-assisted editorial tools that help human writers filter research, structure arguments, and surface relevant sources — positioning the human writer as the curation layer, not the generation layer. Substack in particular has leaned into its "human voice" brand positioning as a direct counter-narrative to AI slop, with algorithmic recommendation improvements designed to surface high-trust authors. This represents an emergent business model: charging for the human filter, not the content itself. Timeline: Feature rollouts ongoing through 2026.

• Enterprise Knowledge Management: Glean, Notion AI, and Microsoft Copilot's "Noise Problem" — Enterprise AI search players — Glean (valued at ~$4.6B as of its 2024 funding round), Notion AI, and Microsoft Copilot — are all confronting the same internal Jevons dynamic: as AI-generated internal documents, meeting summaries, and reports proliferate, enterprise knowledge bases are becoming noisier. Glean has been building quality-scoring and recency-weighting systems to address this. Microsoft's Copilot roadmap includes "content freshness" and "authority scoring" signals. The enterprise knowledge market is bifurcating between raw retrieval and curated retrieval. Timeline: Glean's next funding or IPO signals expected H1 2027; Microsoft Copilot quality features rolling through 2026.

• The "Slop Collapse" of Undifferentiated Content Platforms [HIGH] — Platforms that monetize content volume rather than content quality — including certain programmatic content farms, low-quality SEO aggregators, and AI-native content mills — face structural revenue collapse as Google's quality classifiers and evolving search algorithms increasingly penalize AI-generated, low-provenance content. Evidence: Google's March 2024 "Helpful Content" updates and subsequent 2025–2026 classifier iterations have already deindexed significant volumes of AI-generated content. Who gets disrupted: Dotdash Meredith's SEO-dependent properties, programmatic content networks, and any publisher over-indexed on volume-driven traffic. Who benefits: High-trust, human-edited publishers; citation-infrastructure players like Perplexity; and quality-scoring middleware companies.

  • KPIs to track: (1) Organic search traffic trends for top-20 programmatic content publishers (available via Semrush/Similarweb monthly data); (2) Google's "Helpful Content" classifier update frequency; (3) CPM trends for verified vs. unverified content inventory.

• Curation Infrastructure Emerging as a Standalone Investment Category [HIGH] — The filtration layer — tools and platforms that score, verify, attribute, and rank AI-generated content — is beginning to attract dedicated venture capital as a distinct category, separate from generation infrastructure. Evidence: Companies including Originality.ai, Pangram Labs, and academic provenance tools are gaining traction. The structural logic is that when supply becomes infinite, the scarce resource is trusted filtering. Who gets disrupted: Generalist AI content platforms without provenance architecture. Who benefits: Specialized curation-layer companies; publishers with institutional editorial brands; and human-in-the-loop verification services.

  • KPIs to track: (1) Venture funding volume into "AI content verification" and "provenance" categories (track via PitchBook quarterly); (2) Enterprise contract wins for detection/scoring tools; (3) Adoption rate of C2PA (Coalition for Content Provenance and Authenticity) content credentials across major platforms.

• Attention Economics Inversion: Time-Spent Metrics Decoupling from Content Volume [MEDIUM] — Early data from social and media analytics firms suggests that user time-spent is not scaling proportionally with content volume — a signal that human attention has hit a structural ceiling that Jevons cannot overcome. If confirmed at scale, this would mean the paradox has a hard upper bound in the attention economy, fundamentally altering assumptions about content platform TAM. Who gets disrupted: Advertising-dependent platforms assuming continued engagement growth. Who benefits: Premium, low-volume, high-trust content models; subscription-based editorial businesses.

  • KPIs to track: (1) Time-spent-per-session trends across Meta, YouTube, and TikTok (quarterly earnings disclosures); (2) Subscription conversion rates at premium publishers (NYT, The Atlantic, Substack top-tier); (3) Advertising CPM premium for "verified human content" inventory.

• Regulatory and Standards Pressure on AI Content Provenance [MEDIUM] — The EU AI Act's transparency requirements for AI-generated content, combined with the C2PA standard gaining adoption at Adobe, Microsoft, Google, and OpenAI, is creating a regulatory tailwind for provenance infrastructure. If C2PA adoption reaches critical mass by H1 2027, it could structurally disadvantage unattributed AI content and create a compliance moat for platforms with embedded provenance tooling. Who gets disrupted: Platforms without provenance architecture; content aggregators relying on unattributed AI generation. Who benefits: Adobe (Firefly + Content Credentials), Microsoft, and provenance-native startups.

  • KPIs to track: (1) C2PA member adoption rate and platform integration announcements; (2) EU AI Act enforcement action timeline (first major enforcement expected 2026–2027); (3) Advertiser policy updates requiring content provenance disclosure.

Strengthening Moats

• The New York Times / Premium Editorial Brands: The NYT's aggressive legal posture against OpenAI (ongoing litigation as of mid-2026) combined with its subscription model and brand trust creates a structural moat in an environment where undifferentiated content is abundant. The scarcity of trusted, attributed, human-edited journalism strengthens the economic case for subscription over ad-supported volume models. The NYT's Games and Cooking verticals demonstrate that engagement depth — not volume — is the defensible metric.

• Adobe: Adobe's integration of Content Credentials (C2PA standard) into Firefly and its Creative Cloud ecosystem positions it as the provenance infrastructure layer for professional content creation. As regulatory and platform pressure on AI content attribution increases, Adobe's early mover position in embedding provenance at the creation layer strengthens its moat against both generative AI disruptors and content verification competitors.

• Glean (Private): In the enterprise segment, Glean's focus on quality-scored retrieval — not just retrieval — over internal knowledge bases is building a moat that becomes more defensible as AI-generated internal content proliferates. The more noise enterprises generate internally, the more valuable Glean's signal-extraction architecture becomes. Classic Jevons moat: the paradox creates the market.

Eroding Moats

• Google Search: This is the highest-stakes moat erosion story in the domain. Google's PageRank-era moat was built on the assumption that web content was predominantly human-generated and that link signals were reliable quality proxies. AI-generated content at scale breaks both assumptions simultaneously. Google is spending significant engineering resources on quality classifiers, but the structural problem — that its own AI products (Gemini, AI Overviews) contribute to the noise they are trying to filter — creates an internal contradiction that is difficult to resolve cleanly. Investment teams with exposure to Alphabet should be tracking search revenue per query trends as a leading indicator.

• Programmatic Advertising Networks: The CPM premium for verified, high-quality content inventory is beginning to diverge from commodity AI-generated inventory. Networks that cannot credibly differentiate content quality — including mid-tier programmatic players — face structural margin compression as brand-safety concerns and provenance requirements from major advertisers tighten.

Emerging Moats

• Human-Verified Curation as a Premium Service Layer: A new defensible position is forming around the explicit certification of human editorial involvement — what might be called the "human-in-the-loop" provenance moat. This did not exist as a distinct commercial category 12 months ago. Substack's "human voice" positioning, the NYT's legal strategy, and the emerging C2PA ecosystem are all early expressions of this. The emerging moat is not about being human instead of AI, but about being verifiably human in addition to AI — a hybrid curation credential that commands a price premium. Companies that build the infrastructure to certify and monetize this credential are forming a genuinely new defensible position.

• Taste and Judgment as Infrastructure: Recommendation systems that encode demonstrated editorial taste — rather than engagement-maximizing algorithms — are beginning to emerge as a distinct infrastructure category. Entities like Artifact (acquired by Yahoo in 2024) and the algorithmic curation layer at Substack represent early experiments. The moat is the accumulated signal of human editorial judgment encoded at scale.

Counter-Thesis: Why Jevons May Not Dominate the Information Economy

The following section steelmans the argument that incumbent information structures will successfully absorb the AI content shock without systemic disruption.

The Jevons Paradox framing assumes that demand for information consumption is elastic enough to absorb infinite supply expansion — but there is a strong structural counter-argument: human attention is genuinely inelastic at the margin. Unlike energy consumption (the original Jevons domain), where efficiency gains unlock new use cases (electrification, transportation, industry), information consumption is ultimately bounded by cognitive bandwidth, waking hours, and the neurological cost of context-switching. The empirical signal supporting this counter-thesis: social media time-spent metrics have not scaled proportionally with content volume over the past three years, suggesting the paradox has a harder ceiling in attention economics than in energy economics.

Furthermore, platform algorithms have historically proven adaptive to content inflation. Google survived the SEO spam era of 2010–2015, the link-farm era, and the content-farm era (Demand Media's collapse post-Panda update is instructive). Each time, algorithmic adaptation — not structural market collapse — was the resolution. There is a credible case that Google's current quality classifier investment represents the same adaptive cycle, and that the search moat survives intact.

Finally, the "slop" problem may be self-limiting through market feedback. Advertisers, whose CPM decisions are the economic signal that funds content production, are already implementing brand-safety measures that defund low-quality AI content. If the economic incentive for slop production is removed faster than the volume compounds, the Jevons dynamic may be contained before it reaches a systemic tipping point. Investment teams should weight this counter-thesis seriously before assuming the curation infrastructure market is as large as the bull case implies.

1. Track C2PA Adoption Velocity as a Leading Indicator of Provenance Infrastructure Value — The Coalition for Content Provenance and Authenticity's adoption rate across major platforms (Adobe, Microsoft, Google, OpenAI are members) is the single most important standards-layer signal in this domain. Investigate the timeline for C2PA becoming a default rather than optional feature in major content creation tools. A signal that would accelerate this: a major advertiser (P&G, Unilever scale) publicly requiring C2PA credentials for inventory purchasing. Monitor quarterly platform integration announcements and the EU AI Act enforcement calendar.

2. Evaluate the Technology Trajectory of Enterprise Knowledge Curation Players — Specifically Glean — As AI-generated internal content proliferates inside enterprises, the quality-scoring retrieval layer becomes structurally more valuable. Investment teams monitoring enterprise software should assess Glean's revenue trajectory, customer retention data, and whether its quality-scoring architecture is genuinely differentiated from Microsoft Copilot's retrieval layer. A key inflection signal: Glean filing for IPO or raising a late-stage round at a materially higher valuation, which would confirm institutional conviction in the enterprise curation moat thesis.

3. Monitor Google Search Revenue Per Query as the Jevons Stress Test for Alphabet — The most direct financial signal of whether the Jevons content explosion is materially damaging Google's core moat is revenue per query (or revenue per search session) — a metric that Alphabet does not report directly but can be approximated from total search revenue divided by disclosed query volume proxies. Investment teams with exposure to Alphabet should be building this approximation quarterly. A declining trend would signal that AI content inflation is degrading query quality and advertiser confidence faster than Google's classifier remediation can compensate.

4. Investigate the "Human Voice Premium" Business Model at Substack and Beehiiv — The hypothesis that human-attributed, subscription-funded content commands a durable price premium over AI-generated alternatives is testable through Substack's and Beehiiv's revenue and subscriber growth data. Monitor paid subscriber growth rates, average revenue per subscriber, and churn metrics at both platforms through H2 2026. A signal that would strengthen the thesis: a major legacy media brand (Condé Nast, Atlantic Media scale) migrating a flagship publication to a Substack-style direct subscription model, explicitly citing AI slop as the competitive tailwind.