Battle of Ideas

We should ban targeted advertising

A steelman map of the public fight over targeted advertising — arguments for and against, with sources and assumptions.

AI-generated · paired steelman agents · independently red-teamed · Pass-1 source spot-checks only · framing-fidelity not independently verified · single model family

Personalized/behavioral ads that track people. Contextual ads (this page is about shoes, show a shoe) are a third camp. A ban on all advertising is not this claim. Surveillance, manipulation, small publishers, and political ads are in the fight.

AGAINST 5

no further strong arguments at this depth

FOR 5

no further strong arguments at this depth

People also ask

Questions people actually type. The two columns above are the cases — open a card for sources, assumptions, and counters.

Should targeted ads be banned?

Ban targeted advertising

Targeted ads

Why targeted advertising is bad

Targeted advertising vs contextual

Pros and cons of targeted advertising

Ordering within each column: strongest first — validation tier, then source quality, then representativeness.

AGAINST · We should ban targeted advertising
Empirical — moderateP1

Behavioral ads are how a large part of the open web and a lot of journalism get paid

Digital advertising is the dominant funding model of the non-paywalled web: search, social, and open-display. The UK CMA's 2020 market study is not a love letter to Google and Meta — it is an official picture of how large that market is and how much of news and other content sits on it. IAB/PwC internet advertising revenue reports put US digital ad spend in the hundreds of billions a year; even if a slice is brand search and a slice is contextual, the behavioral remainder is what made 'free' the default. Marotta, Abhishek and Acquisti's ~4% publisher lift from a cookie is the FOR number from one firm. Industry estimates on the other side (Beales and Eisenach 2014, often cited: on the order of 50%+ lower prices without cookies) are the advertiser-willingness-to-pay number, and they are industry-adjacent — flag that. AGAINST does not need 52% to be gospel. It needs this: publishers and platforms that cannot show a relevant ad to a likely buyer get a cheaper ad, or no ad, and then they cut costs. Local news, specialist blogs, and free tools are the elastic supply. Subscriptions and contextual ads are real substitutes for some titles (the New York Times is not the median). They are not a substitute for the long tail of pages that cannot charge £15 a month. A ban on targeted advertising is, as this page draws the fight, a ban on the funding technology of that tail. Contextual ads are the named third camp. AGAINST's claim is that contextual is a thinner market — the CMA and the post-ATT wreckage in app-install ads are the exhibit — and that pretending it is 'just as good' is how you write a privacy statute that is actually a newspaper-closure statute.

Key assumptions

  • A large share of open-web and news output is on the margin of ad funding, so a targeting ban's price cut becomes a content cut partial
  • Contextual and subscription alternatives will not replace that funding at comparable scale partial

Red team — the strongest counters

Funding the open web with dossiers is not a law of nature — classified ads and subscriptions existed

Newspapers lived on classifieds, display, and cover price before the cookie. Some of that mix can return. 'The tail will die' is a prediction about a specific adtech equilibrium, and FOR's Marotta 4% is the claim the tail's cookie-dependence is overstated. Using IAB's revenue total as if it were all behavioral, all marginal, and all journalism is three aggregations too far. Search and retail media would survive a third-party targeting ban.

A business model is not a right

If the only way to fund a page is a surveillance overlay, that is a reason the page's costs are too high or its willingness-to-pay too low, not a reason the overlay must remain legal. We do not keep toxic waste because a town's landfill depends on it. AGAINST has to win on the matching and speech points, not on 'but the blogs.' Some blogs should be subscriptions or go away.

Sources

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · We should ban targeted advertising
Empirical — moderateP1

Targeting is how a small seller reaches a buyer without owning a billboard

The classical defence of targeting is matching: a niche product finds the people who want it instead of taxing everyone with a spray. Goldfarb and Tucker (Management Science, 2011) measured the EU ePrivacy Directive's restriction on tracking: display-ad effectiveness fell on the order of 65% in their estimates. That is FOR's 'see, privacy rules bite' paper. It is also AGAINST's 'see, targeting was doing the matching' paper. Small advertisers and app developers are the people who cannot replace that matching with a TV budget or a first-party log-in graph. Apple's ATT is the second natural experiment: Meta booked about $10 billion of 2022 impact; independent estimates (Aridor et al. MSI 2024; Kraft, Skiera and Koschella on opt-in vs opt-out) find advertiser acquisition getting more expensive and trackable inventory shrinking. The surplus did not vanish into consumer pockets as a clean privacy dividend. Some of it went to Apple's own ad stack; some of it went to higher prices for the remaining identified impressions; some of it left small performance advertisers. AGAINST can grant the Zuboff diagnosis of platform power and still say: a legal ban on targeting is not a surgical strike on Google. It is a ban on the only targeting a garage seller can buy. First-party data (log-in walls, retail media) is how the already-large get larger when third-party targeting dies. That concentration effect is the CMA's other worry, and it cuts against a blanket ban.

Key assumptions

  • Small advertisers' use of behavioral targeting is matching, not just a cheaper way to do the same surveillance harms partial
  • ATT's costs falling on non-platform advertisers is a preview of a legal ban, not a platform-specific quirk partial

Red team — the strongest counters

Small-seller targeting is often the same dossier, just cheaper

A garage shop buying lookalikes is still compiling the person. 'They cannot afford TV' is an equity argument for the seller, not a privacy argument for the buyer. FOR can grant small-business matching and still say the buyer's file is not the seller's to rent. Goldfarb–Tucker's 65% is advertiser effectiveness, not consumer welfare. ATT hurting small apps is an incidence fact; it can justify a subsidy or an ad-stack unbundling rather than keeping the graph.

Walled-garden concentration is a reason to unbundle platforms, not to keep third-party tracking as a competitor

AGAINST's 'don't give Amazon a statute' is real. The answer can be DMA, interoperability, and a public identity-free ad marketplace — not a defence of 2012 cookies as the competitive fringe. Third-party adtech as a check on Google is the 'fight fire with a smaller fire' theory. It has not obviously produced a competitive open display market (CMA).

Sources

  • Privacy Regulation and Online Advertising Goldfarb & Tucker, Management Science 57(1):57–71, 2011. EU ePrivacy associated with ~65% drop in display-ad effectiveness in the authors' estimates. Pass-1: DOI exists. AGAINST: targeting was doing measurable matching work. P1 checked
  • Meta on Apple ATT — 2022 revenue impact CNBC 2 Feb 2022: ~$10B 2022 sales impact from ATT. Pass-1: CNBC exists. Firm-reported. Pair with the independent ATT literature (Aridor et al.) as 'the identified-impression market got more expensive.' P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●○
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · We should ban targeted advertising
Empirical — moderateP1

Contextual is a third camp, not a full substitute — and first-party walls are how platforms win a ban

The named third camp is contextual: this page is about boots, show a boot. Some of that works. The CMA's worry, and the post-cookie / post-ATT facts, is that identity then concentrates where the user is already logged in. Google's and Amazon's retail-media and walled-garden ads, Apple Search Ads' rising share of iOS install spend after ATT, and publisher log-in walls are the substitution path. A legal ban on 'targeted advertising' that is written as a ban on third-party graphs is a statute that Google and Amazon can live with better than an independent ad network. That is not a reason to love third-party cookies. It is a reason a flat ban-vs-keep framing hides the distributional question of whose targeting remains. Privacy-preserving measurement (Chrome's on-again off-again cookie plans, SKAN, clean rooms) is the industry's own third camp: keep some matching, kill the open dossier. AGAINST can live in that camp without defending 2012 adtech. What it will not grant is FOR's implication that contextual plus subscriptions is a complete replacement at the current scale of free content. The ban, as a blunt instrument, is how you get less independent media and more platform-native ads. If the page has only two columns, that is an AGAINST card: do not give the walled gardens a statute.

Key assumptions

  • A targeting ban binds independent adtech harder than logged-in platforms partial
  • That concentration is a reason to reject the ban rather than a reason to pair the ban with platform break-up untestable

Red team — the strongest counters

If the ban binds independents harder, write the ban to bind the gardens too

A statute that only kills third-party cookies is AGAINST's straw. FOR's claim is targeted advertising, including first-party behavioral targeting by platforms. Drafting can reach log-in graphs used to bid on identity. Concentration is then a transition cost, not a proof the claim is a gift. Pairing a targeting ban with DMA-style duties is available. Using concentration as a veto is how the gardens keep both the graph and the independent competitors weak.

Aridor/ATT evidence is still young, platform-confounded, and not a welfare paper

MSI working papers on ATT are not a completed science of a legal ban. Apple as the parliament is a confound: ATT is also a competitor's self-preference. Using it as a preview of a democratic statute over-reads a vertical conflict. Empirical-moderate was the right tag; it should not carry the column.

Sources

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●○
Source quality●●●○○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · We should ban targeted advertising
Logically validP1

Political advertising is speech — a targeting ban is a speech rule, and democracies already have narrower tools

FOR's political card treats uninspectable microtargeting as a different harm class. AGAINST can grant the inspectability point — DSA-style ad libraries, no dark posts, spend caps, bans on foreign-funded targeting — and still refuse the claim. Political speech that cannot use a list of people who downloaded a manifesto, or who live in a swing ward, is speech that only wealthy parties can replace with billboards. The Cambridge Analytica specimen is real as a data-protection failure (ICO: Facebook data used for political advertising) and still not a proven election-swing machine. Building a worldwide targeting ban on that specimen is how you also ban a local candidate's lookalike of last year's donors. Commercial speech is the rest of the claim: product ads are how unknown firms enter. US First Amendment doctrine and European commercial-speech law will not be restated here as if they settled the welfare question; they settle that a ban is a hard legal lift, which is a fact about the claim's feasibility, not its merit. The merit point is narrower: the fair-housing and employment targeting bans (DOJ/Meta 2022) are the right shape — forbid the use in the sensitive market — not 'therefore no shoe retargeting.' FOR says the graph is one graph. AGAINST says the law already knows how to slice uses, and a use-slice is how you stop the housing ad without closing the shop.

Key assumptions

  • Use-specific bans (housing, employment, political-ad libraries) can be enforced without a general targeting prohibition partial
  • A general ban would fall harder on challengers and small speakers than on incumbents with first-party lists and TV money partial

Red team — the strongest counters

Inspectability is the democratic constraint, and dark posts fail it

Direct mail at least had a physical copy someone could wave. A thousand variants of a fear ad, shown only to an inferred anxious subset, is a different public sphere. AGAINST's 'we already have narrower tools' is true only if those tools are enforced. Ad libraries that do not show targeting parameters, or that miss dark posts, are theatre. FOR's bundling claim (kill the graph to kill the political product) is a second-best when the first-best libraries keep failing.

Challenger-speech can use contextual and lists they own

A local candidate can still advertise next to local news, on their own site, and to email subscribers. The lookalike of last year's donors is the incremental targeting this claim would take. That increment is real and is also the microtargeting harm. 'TV money wins' is already the world; it is not obvious that Facebook lookalikes were the equaliser rather than a new spam channel.

Sources

  • Digital Services Act (Regulation (EU) 2022/2065) DSA plus the political-ad transparency track: extra rules on political targeting, not a commercial targeting ban. Pass-1: EUR-Lex exists. AGAINST: the EU already picked the narrower tool. P1 checked
  • DOJ–Meta Fair Housing Act settlement (2022) Use-specific remedy: housing ads, Special Ad Audience / lookalikes, delivery system. Pass-1: justice.gov archives PR exists. Specimen of slicing the sensitive use rather than banning the graph. P1 checked

Confidence, decomposed

Logical validity●●●●●
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · We should ban targeted advertising
Logically validP1

A relevant ad is a better tax on attention than a random one

The claim is not a ban on ads. If targeting goes, the remaining instruments are contextual, demographic-coarse, and blast. Blast is how you get more ads, worse ads, and ads that do not even pretend to match. Survey and industry evidence that people 'hate ads' is usually evidence they hate irrelevant or too-many ads — which is the targeting problem in the other direction. AGAINST's welfare claim is the matching claim again, on the consumer side: the expected value of a minute of attention is higher if the ad is closer to a purchase the person might want. That is not a finding that dossiers are costless. It is a finding that the alternative equilibrium is not a quiet garden. It is a louder garden with worse signs. GDPR consent banners are the preview of a world that tried to keep targeting behind a click and produced theatre. A ban that really holds would reallocate spend toward brand TV, retail media owned by Amazon and the grocers, and influencer deals — not toward a renaissance of untracked newspapers. Contextual is the third camp and deserves a real trial (some publishers already sell it). AGAINST's objection is substituting the hope of contextual for the measured matching of Goldfarb–Tucker and the measured publisher-and-advertiser dependence in the CMA report. Hope is not a column.

Key assumptions

  • The post-ban equilibrium is worse ads, not fewer ads — advertising demand is not going to zero partial
  • Consumers' dislike of ads is mostly dislike of irrelevance and volume, which targeting reduces on the irrelevance margin partial

Red team — the strongest counters

People's 'hate of irrelevance' is not a consent to the file

Preferring a shoe ad to a random banner, conditional on there being an ad, is not the same as preferring a world with a cross-site graph. The relevant comparison is targeted ads versus contextual ads plus fewer ads, not versus Times Square. GDPR's 'Accept all' is AGAINST's theatre point; it is also evidence the current stacking of targeting-plus-consent is not a meaningful choice. A ban is one way to make the choice real.

Spray can be regulated on volume and placement without the dossier

If the fear is a louder garden, cap frequency, cap kids' inventory, cap political spend. Those are quantity rules. They do not require compiling the person. AGAINST's 'the alternative is worse ads' assumes the ban is the only instrument and that ad demand is inelastic. Frequency caps and contextual are the third camp applied to annoyance.

Sources

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●○
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · We should ban targeted advertising
Empirical — moderateP1

The targeting graph is how housing, job, and credit ads learn to skip people — contextual ads do not need that trick

US Department of Justice and HUD settlements with Facebook (2019–2022) over housing, employment, and credit ads are the existence proof: advertisers could (and did) use the targeting stack to exclude protected classes, and the lookalike/neighbourhood tools could reconstruct the exclusion even after the obvious checkboxes were removed. Academic work on ad delivery (Ali, Sapiezynski, Bogen, Korolova, Mislove, Rieke and others, 2019, 'Discrimination through optimization') showed Facebook's own delivery algorithm skewing job and housing ads by gender and race even when the advertiser did not ask. That is not a rogue CRM. It is what a bid for predicted response does with correlated features. GDPR special-category data and US fair-housing/fair-lending law already try to forbid the use. They chase a graph that was built to infer the forbidden variable from the allowed ones. Contextual ads — this page is about jobs in Glasgow; this page is a mortgage explainer — still reach people by content. They do not need a shadow score of 'likely Black' or 'likely pregnant.' FOR's claim is the clean instrument: if you cannot compile the person, you cannot skip the person at the identity layer. You can still waste money on the wrong article. You cannot quietly un-show a house. Children are the same argument in a different statute (DSA, COPPA, age-appropriate design): a tracking graph on a minor is a targeting graph on a minor. Ban the graph for ads. Keep the banner on the page.

Key assumptions

  • Inferred proxies (neighbourhood, device, browsing) are close enough to protected-class targeting that banning the graph is the workable remedy partial
  • Contextual placement is a good-enough substitute for reaching job-seekers and home-seekers without identity-level delivery bias partial

Red team — the strongest counters

Fair-housing remedies already sliced the sensitive use

DOJ/Meta 2022 is a use-specific settlement: housing ads, lookalikes, delivery. Ali et al. showed delivery bias even without advertiser intent — which is a reason to regulate delivery in those markets, not to ban behavioral ads for running shoes. Protected-class inference is a real harm. The law already knows how to name the market (housing, employment, credit). FOR's 'the graph is the input' over-claims that you cannot stop the proxy without stopping the graph.

Contextual ads can still skip people — just by page, not by dossier

If 'jobs in Glasgow' never runs on a site whose audience is older women, you have a contextual skip. Delivery bias does not require cookies. A tracking ban is not a fairness ban. Children's protection (COPPA, DSA, age-appropriate design) is again a use/age slice. Stuffing it into a general targeting claim is how you get a coalition, not how you match the instrument to the harm.

Sources

Confidence, decomposed

Logical validity●●●●●
Premise support●●●●○
Representativeness●●●●○
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · We should ban targeted advertising
Empirical — moderateP1

Publishers capture a few percent from tracking cookies — the rest is the tracking industry

Marotta, Abhishek and Acquisti (WEIS / FTC PrivacyCon, 2019) got a large publisher's ad-transaction data and compared impressions with and without a usable tracking cookie, controlling for the usual bid covariates. Publisher revenue was about 4% higher when the cookie was available — on the order of $0.00008 more per impression in their reporting. That is the load-bearing publisher-side number. It is not 'advertising is worthless.' It is 'third-party tracking is a thin slice of what the publisher takes home.' EFF's write-up of the same paper is the public translation: publishers benefit little; the surplus sits in the adtech stack. The UK CMA's 2020 market study on online platforms and digital advertising is the official picture of that stack: Google and Meta's market power, limited transparency, and a share of advertiser spend that never reaches the page. CMA-adjacent estimates put the publisher's take of open-display spend well below the advertiser's outlay (on the order of two-thirds at best, often worse once intermediaries are paid). FOR can grant that some small publishers live on programmatic targeting and would feel a ban. The Marotta 4% says the thing they would lose is not most of the page's ad value — unless their whole business is arbitraging identity rather than audience. Contextual ads, first-party subscriptions, and direct deals are the substitute set. Industry studies that find 50%+ losses from 'no cookies' (Beales and Eisenach is the usual cite) are measuring advertiser willingness to pay for a dossier, which is the AGAINST column's number, not a journalism-funding identity. A ban priced at 4% of publisher RPM is not a civilisation-ender. It is a tax on a surveillance overlay.

Key assumptions

  • Marotta et al.'s one-publisher, cookie-on vs cookie-off contrast generalizes to the open web, not only to that firm's mix partial
  • A ~4% publisher RPM hit is an acceptable price for ending cross-site tracking untestable

Red team — the strongest counters

One publisher's 4% is not the open web, and 4% of a failing newsroom is still a newsroom

Marotta et al. is one firm's transactions. Industry estimates of 50%+ (Beales–Eisenach, flagged as industry-adjacent) are the other pole. Small publishers who live on programmatic arbitrage of identity are exactly the people for whom the cookie is not a rounding error. A mean 4% can hide a tail that is 40%. FOR's 'not a civilisation-ender' is a judgement about whose outlets may close.

CMA's stack critique is an argument for competition policy, not for a targeting ban

If publishers take too little because Google and Meta take too much, the instrument is DMA-style unbundling, interoperability, and ad-stack separation — not deleting the matching technology. A ban that leaves the walled gardens' first-party identity intact (AGAINST's concentration card) can make the CMA problem worse. FOR's 4% and the CMA report do not automatically stack into this claim.

Sources

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●○
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · We should ban targeted advertising
Empirical — moderateP1

When tracking requires a real yes, most people say no

App Tracking Transparency, shipping with iOS 14.5 in April 2021, forced apps to ask before tracking a user across other companies' apps and websites. Reported opt-in rates clustered in a low band — commonly cited in the 15–25% range globally, with variation by app and country. Meta told investors the change would take on the order of $10 billion off 2022 revenue. That is the revealed-preference pair: users decline; the tracking business books a loss. GDPR consent banners are a worse instrument — dark patterns, 'legitimate interest,' consent-or-pay walls — but even there, regulators have spent years treating pre-ticked tracking as unlawful precisely because a free choice looks like 'no.' Goldfarb and Tucker (Management Science, 2011) measured the EU's earlier Privacy and Electronic Communications Directive: display-ad effectiveness fell sharply (on the order of 65% in their estimates) when targeting was restricted. FOR reads that paper as proof that targeting was doing real commercial work — and that privacy rules bite. It does not read it as 'therefore we must restore the bite.' A ban is ATT without the platform being the parliament: no cross-site/cross-app identity for ads, no 'ask until they slip.' Contextual and first-party advertising remain. The people who click 'Allow' can still see more relevant ads if a firm uses data from a relationship they know about. What ends is the silent dossier. That is not a nanny state inventing a preference. It is writing into law the preference ATT already measured.

Key assumptions

  • Low ATT opt-in is a privacy preference, not just people tapping the first button or failing to understand the prompt partial
  • A legal ban is the right generalization of Apple's private opt-in switch — including on Android and the open web untestable

Red team — the strongest counters

Goldfarb–Tucker is AGAINST's paper: targeting was doing matching work

A 65% drop in display effectiveness when you restrict tracking is the empirical heart of the 'targeting is valuable' case. FOR's 'privacy rules bite' reading is true and incomplete. If effectiveness collapses, advertisers spend less or spray more. Neither is a clean consumer win. Low ATT opt-in can be confusion, habit, or Apple's framing of the prompt ('track you across other companies' apps') as much as a considered dossier-rejection.

Meta's $10 billion is not a social surplus measurement

A platform's lost revenue is not the public's gain. Some of it is less spam, some is less funding for Instagram, some is a shift to Apple Search Ads. Using a firm's earnings call as a revealed-preference win for a legal ban skips the incidence: who actually paid. Small performance advertisers and app developers are in the incidence, not only 'surveillance capitalists.'

Sources

  • Meta on Apple ATT — 2022 revenue impact CNBC, 2 Feb 2022: Facebook/Meta said Apple's ATT would decrease 2022 sales by about $10 billion. Pass-1: CNBC report of earnings commentary exists. Firm-reported, not a causal paper. P1 checked
  • Privacy Regulation and Online Advertising Goldfarb & Tucker, Management Science 57(1):57–71, 2011. EU ePrivacy Directive associated with a large drop in display-ad effectiveness (~65% in the authors' estimates). Pass-1: INFORMS/SSRN exist. FOR uses this as 'targeting was doing work, and privacy rules change it,' not as a welfare loss that ends the argument. P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●●○
Representativeness●●●●○
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · We should ban targeted advertising
Logically validP1

Behavioral ads are a surveillance business with an advertising sidecar

Shoshana Zuboff's The Age of Surveillance Capitalism (2019) is the public AGAINST-tracking voice and FOR should use it as that, not as a secret paper: the economic logic of the dominant platforms is to extract behavioral surplus, predict, and sell access to the prediction. Targeted advertising is the cash register of that logic. The claim is not 'ban all advertising.' A shoe ad on a shoe article is contextual. A shoe ad that follows you from a medical site to a news site because a pixel sold your browsing graph is behavioral. GDPR (2018) already treats that graph as personal data that needs a lawful basis; the ePrivacy tradition and the EU's remaining cookie-consent theatre are the half-built wall. FOR's claim is that the wall should be a prohibition on using cross-site and cross-app tracking for ads, not a consent banner nobody reads. You can still advertise. You cannot still compile the person. Apple's App Tracking Transparency (iOS 14.5, 2021) is the revealed-preference experiment: when the default is 'ask,' most people say no. That is not a confused public. It is a public that never wanted the file. The business model that requires the file is the thing the claim is for ending. Contextual ads are the named third camp — they survive a tracking ban by construction. What does not survive is the bid for an identity.

Key assumptions

  • The dominant platforms' ad businesses are primarily surveillance-funded rather than context-funded partial
  • A ban on behavioral targeting is separable from a ban on advertising as such — contextual and publisher-first-party ads remain testable

Red team — the strongest counters

Advertising sidecars fund things people actually want; the dossier is not the whole product

Search ads on a query the user typed are targeted without a cross-site file. First-party retailer ads ('you bought boots here') are a relationship, not Zuboff's shadow. A ban written as 'targeted advertising' will be drafted either too wide (it hits search and first-party) or too narrow (it hits only third-party cookies and the graph reconstitutes). GDPR already named the file and did not produce the claim. Treating Zuboff as the cash register of the whole ad economy over-claims the behavioral slice.

ATT opt-in is not a vote for a legal ban

Tapping 'Ask App Not to Track' on a phone is a low-cost, device-level preference. It is not a showing that people want newspapers to lose the rest of their ad stack, or that Android and the open web should be put under Apple's private legislature. Revealed preference against a prompt is not revealed preference for this claim. Dark-pattern consent banners cut the other way: people also click 'Accept all' when that is the cheap path.

Sources

  • The Age of Surveillance Capitalism Shoshana Zuboff, 2019. Public diagnosis: behavioral surplus extracted, predicted, sold; advertising as the cash register. Pass-1: book exists. Use as the named public voice, not as an RCT of a ban. P1 checked
  • Regulation (EU) 2016/679 — General Data Protection Regulation GDPR: personal data processing needs a lawful basis; profiling and tracking are in scope. Pass-1: EUR-Lex exists. GDPR is not a targeting ban — it is the legal recognition that the file is the thing. P1 checked

Confidence, decomposed

Logical validity●●●●●
Premise support●●●●○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · We should ban targeted advertising
Logically validP1

Political microtargeting is a different harm class from a shoe on a shoe page

Behavioral political advertising is the claim's sharpest non-commercial case. The Cambridge Analytica / Facebook 2018 episode is the public specimen — psychometric and Facebook-derived targeting sold as a way to reach voters as individuals — and FOR should not retell it as a proven swing of a national election. The load-bearing structural fact does not need that swing. It needs this: a targeted political message is not public speech in the billboard sense. It is a private payload, varied by dossier, invisible to the opposing campaign and to the press unless someone leaks a screenshot. That breaks the old liberal bargain in which propaganda is at least inspectable. The EU's Digital Services Act and the political-ad transparency fights in Brussels and London are the live legal recognition that 'this is just more speech' failed. A contextual rule — run your party ad next to politics coverage, or on a known party page — still lets persuasion happen in public. A tracking ban on political (and lookalike) targeting is the subset even some targeting-defenders already concede. FOR's full claim applies the same logic to commercial dossiers because the infrastructure is the same graph. You do not get 'no political microtargeting' while leaving the commercial identity graph intact; the graph is what makes the political product cheap. This is worldwide: Indian, Brazilian, Nigerian, and US campaigns buy the same adtech. The harm is not 'ads in elections.' It is unaccountable, person-specific influence at scale.

Key assumptions

  • Uninspectable person-specific political messages are a democratic harm even if they do not swing a particular election untestable
  • You cannot ban political microtargeting while leaving the commercial tracking graph in place — the graph is the product partial

Red team — the strongest counters

Cambridge Analytica is a data-protection failure, not a proven swingometer, and political-ad libraries already exist

The ICO investigation found real misuse of Facebook data. It did not find a reproducible machine for buying elections at scale. DSA political-ad transparency, spend caps, and no-foreign-targeting rules are the narrower tools AGAINST names. FOR's 'the graph is one graph' is a bundling claim: you must kill shoe retargeting to save democracy. That is a policy choice, not a technical necessity — lookalike of last year's donors can be banned in the political use without the claim.

Uninspectable messages are older than adtech (direct mail, phone banks)

Campaigns have long sent different leaflets to different streets. The internet made it cheaper and more personal. Treating cheapness as a new harm class needs a threshold: why is a targeted Facebook ad worse than a targeted mailing list? If the answer is scale and opacity, transparency rules (ad libraries) match the diagnosis better than a commercial targeting ban that also hits the shoe.

Sources

Confidence, decomposed

Logical validity●●●●●
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.