The Undersupply Theorem Meets the Extraction Equilibrium
Meta Perspective
"If you think about the pieces of advertising, there’s content creation, the creative, there’s the targeting, ... and there’s the measurement and probably the first pieces that we started building were the measurement..."
https://stratechery.com/2025/an-interview-with-meta-ceo-mark-zuckerberg-about-ai-and-the-evolution-of-social-media/
Targeting:
"Over the last 5 to 10 years, we’ve basically gotten to the point where we effectively discourage businesses from trying to limit the targeting. It used to be that a business would come to us and say like, “Okay, I really want to reach women aged 18 to 24 in this place”, and we’re like, “Okay. Look, you can suggest to us…” ...If they really want to limit it, we have that as an option. But basically, we believe at this point that we are just better at finding the people who are going to resonate with your product than you are."
https://stratechery.com/2025/an-interview-with-meta-ceo-mark-zuckerberg-about-ai-and-the-evolution-of-social-media/
Creatives:
"it’ll always be the case that they can come with a suggestion or here’s the creative that they want, especially if they really want to dial it in ... we’re going to get to a point where you’re a business, you come to us, you tell us what your objective is, you connect to your bank account, you don’t need any creative, you don’t need any targeting demographic, you don’t need any measurement, except to be able to read the results that we spit out."
As targeting and creatives get better, the cost of an action (for the publisher) will be lower, conditional on the content (audience), which is the last lever (for increasing profits)
"Number two is basically growing engagement on the consumer surfaces and recommendations. So part one of that is just get better at showing people the content that’s out there, that’s effectively what’s happening with Reels. Then I think what’s going to start happening is that the AI is not just going to be recommending content, but it is effectively going to be either helping people create more content or just creating it themselves."
https://stratechery.com/2025/an-interview-with-meta-ceo-mark-zuckerberg-about-ai-and-the-evolution-of-social-media/
Broader Perspective
AI shifts advertising along three margins: reduces creative production and testing costs; improves ad-person matching; expands content and engagement via cheaper production, better recommendations, and more engaging content that increases platform time and ad revenue. These effects depend on usable signals—IDs, conversions, credible holdouts—and are muted by privacy frictions and measurement error.
Welfare Framework
AI advertising creates welfare through better product-consumer matching, solving information problems in large markets, particularly valuable for niche products and specialized services where traditional discovery fails. Reduced creative production costs enable smaller businesses to compete and test informative messaging. Content expansion operates through three channels: cheaper production via AI tools (writing, image generation, video editing), better recommendation algorithms reducing discovery friction, and more engaging content that sustains creators through increased ad revenue. These benefits must be weighed against cognitive costs and attention risks.
Negative effects can offset or overwhelm benefits. AI persuasion induces inferior purchases that consumers wouldn't make with full information. Sophisticated engagement systems capture attention beyond rational allocation—revealed preferences diverge from considered preferences. Market concentration accelerates as AI advantages flow to platforms and large advertisers with superior data, compute, and technical capabilities. Real resource costs—computing infrastructure, energy, and cognitive burden on billions processing ads—represent pure social cost.
Critically, business stealing complicates welfare accounting. Firm-level incrementality can be positive, while market-level incrementality is zero—ads merely redistribute purchases among competitors. This contest for fixed demand creates deadweight loss: resources spent on advertising arms races with no net value creation. AI may intensify these dynamics by lowering advertising costs (triggering more competition) and improving targeting (making stealing more efficient). The social value of advertising depends on the mix of market expansion versus redistribution—a ratio that varies dramatically by category. Mature categories with established preferences show mostly redistribution; emerging categories or those with information problems show genuine expansion.
A first-order welfare accounting would be:
$ \Delta W \approx \tilde{\rho},G,Q - \tilde{\sigma},H,Q - \Phi(\text{concentration}) - \Psi(\text{attention capture}) + \tilde V_{\text{content}}(S) - (\kappa+\alpha+\tau)E $
Term definitions:
- $\tilde{\rho},G,Q$ (incremental value): True incrementality $\tilde{\rho}$ = fraction of attributed conversions that wouldn't occur without the ad at market level. Firm-level studies find search ads often 0.20-0.50, display/social often 0.05-0.15, but these overstate social value when ads primarily steal business from competitors. Market-level incrementality is highest for new categories, information-intensive products, and niche discovery; near zero for mature categories with fixed demand. $G$ = welfare gain per truly incremental match. $Q$ = total attributed conversions.
- $\tilde{\sigma},H,Q$ (persuasion harm): $\tilde{\sigma}$ = fraction of conversions where consumers choose inferior products due to persuasion rather than information. $H$ = welfare loss per manipulated purchase. Both lack robust empirical estimates. Note: business stealing that doesn't involve persuasion toward inferior products is already captured by lower market-level $\tilde{\rho}$.
- $\Phi$ (concentration): Market structure loss from AI advantages accumulating to dominant firms. Magnitude depends on the counterfactual competitive landscape. Includes dynamic effects where advertising arms races create barriers to entry.
- $\Psi$ (attention capture): Welfare cost of time spent beyond the user's optimal allocation. The gap between revealed and considered preferences is largely unmeasured. Includes cognitive load from processing competitive advertising in zero-sum contests.
- $\tilde V_{\text{content}}(S)$: Net value from cheaper production, better recommendations, and engaging content. Recommendation algorithms may increase engagement 20-40% based on platform studies, but the welfare implications are unclear.
- $(\kappa+\alpha+\tau)E$ (resource costs): When advertising is primarily redistributive (business stealing), these become pure deadweight loss. Compute/infrastructure $\kappa$ is increasingly visible in platform AI spending. Cognitive cost $\alpha$ and environmental cost $\tau$ lack consensus measurement. $E$ = total ad spend.
Unmodeled Dynamics
The linear framework above misses important effects:
- Dynamic feedbacks: Easier persuasion reduces quality investment (lowering future $G$); market expansion enables previously unviable businesses; stronger targeting accelerates winner-take-all dynamics through data and capital advantages.
- Interactions: Terms aren't independent—better targeting simultaneously increases both incrementality and manipulation; engaging content raises persuasion vulnerability; personalized feeds deepen filter bubbles that amplify persuasion. These interactions mean marginal changes cascade, making additivity misleading. The static, linear form provides first-order intuition but understates both potential benefits and risks.
Consumer Agency and Natural Bounds
Consumers aren't passive victims. Multiple mechanisms limit harm:
- Purchase protection: Returns, exchanges, and chargebacks cap losses. Higher scrutiny for expensive items. Platform choice reflects return policy preferences.
- Information mechanisms: Reviews, ratings, and word-of-mouth provide independent quality signals, particularly for experience goods.
- Learning and adaptation: Consumers develop manipulation recognition and skepticism, though sophistication varies, and adaptation lags AI capabilities.
- Platform competition: Users switch when platforms become exploitative, to those with better protections, less aggressive advertising, and superior content.
- Reputation constraints: Long-term reputation costs constrain aggressive tactics when lifetime value is high or word-of-mouth matters.
These bounds cap rather than eliminate $\tilde{\sigma}$ and $\Psi$. Weaker for vulnerable populations (elderly, low-income, less educated) and low-stakes purchases where decision cost exceeds error cost.
Market Structure Implications
AI changes industry composition, not just efficiency. Small firms displaced without comparable AI tools at similar unit economics, increasing $\Phi$. Capital flows toward marketing-efficient rather than product-innovative firms. Ad-funded "free" content with manipulative tactics cross-subsidizes at scale, disadvantaging paid content.
Who Captures the Benefits
Structural Advantages:
Platforms systematically benefit:
- Control measurement, enabling overstatement of $\tilde{\rho}$ and understatement of $\tilde{\sigma}$
- Extract rents through market power and auction design
- Network effects raise switching costs
- Pass AI/compute costs to advertisers while retaining data advantages
AI/Compute providers (OpenAI, Anthropic, Google, Meta) capture value through:
- API pricing for creative generation and targeting models
- Oligopolistic market structure in frontier models
- Vertical integration opportunities (own ad products)
- Lock-in via fine-tuning and workflow integration
- Pricing power to extract a significant portion of efficiency gains
Advertisers face compressed margins in competitive markets:
- Benefit from genuine market expansion, reaching previously unreachable customers
- Temporary advantages from superior targeting/creative
- Large advertisers gain relative advantages
- Success depends on equilibrium (see scenarios)
Consumers experience mixed outcomes by sophistication:
- Sophisticated: use blockers, benefit from matching, avoid manipulation
- Vulnerable: bear disproportionate $\tilde{\sigma}H$ and $\Psi$ (excess platform time, addictive engagement, cognitive diversion)
- All: potentially benefit from content and matching, face a higher cognitive load
Scenario Outcomes:
1. Informative targeting, audited lift ($\tilde{\rho}↑$, $\tilde{\sigma}·H↓$)
- Consumers: better matches, higher-quality content
- Platforms: higher spend $E$
- Advertisers: benefits only with cost/creative advantages—competition bids surplus to zero
2. Persuasion escalation ($\tilde{\sigma}·H↑$, little new demand)
- Platforms: capture increased spend
- Advertisers: break even marginally, possible brand damage
- Consumers: net negative from manipulation
3. Niche entry (fixed costs ↓, high $\tilde{\rho}$ in new categories)
- New advertisers: temporary rents before competition
- Consumers: variety and fit benefits
- Platforms: market expansion
- Incumbents: lose share
4. Brand-capital play (strong ad-stock, long horizons)
- Brand incumbents: future margins from investment
- Platforms: sustained high spend
- Consumers: depends on quality vs persuasion
- SMEs: squeezed by capital requirements
5. Concentrated markets ($Φ↑$)
- Dominant platforms/advertisers: retain surplus through moats
- Consumers: often worse if $\Psi$ rises
- Smaller advertisers: higher costs, reduced tool access
Platforms and AI providers benefit across scenarios; advertiser and consumer outcomes vary dramatically by equilibrium.