Build advertising audiences from page content instead of third-party cookies. A model reads what a page is about and infers its likely readers — demographics, interests, intent, life stage and B2B signals — with banded confidence. Works in every browser, zero tracking dependency.
Traditional audience segmentation profiles people via third-party cookies. Cookieless audience segmentation inverts the model: it profiles a page or domain and describes the audience that content predictably attracts — no tracking required.
A third-party cookie follows an individual across thousands of sites. Observed behavior is rolled up into a user profile — “in-market SUV shopper, 35–44, upper-middle income” — and synced into ad platforms as a targetable segment.
A model reads what a page is about and infers who is likely reading it — age, gender skew, income, interests, purchase intent — with a confidence band on every attribute. No user is ever observed anywhere in the process.
Magazine publishers sold ads on audience composition for a century without tracking a single reader: a sailing title reached affluent readers because of what it was about. Large language models now make this computable per URL, at web scale, with structured output and confidence values — exactly what our audience segmentation API produces.
Age brackets, gender skew, income, education, life stage, household composition and more
285 sub-interests plus 283 in-market purchase intent segments
Target role, seniority, company size and industry using LinkedIn-standard firmographic bands
1,667 deterministic personas mapped from IAB content categories — auditable and stable
“Food & Drink” is a category. “Skews female, 25–44, family life stage, in-market for meal kits and kitchen appliances” is a segment a planner can buy against. The output is a targetable audience definition, not a contextual tag.
Every attribute is drawn from vocabularies aligned with IAB Audience Taxonomy 1.1. The same segment codes mean the same thing across every page, report and integration — making segments tradable, not merely descriptive.
The audience data supply chain adtech relied on for fifteen years has lost most of its raw material — not at some future deadline, but already. A broader survey is in our guide to cookieless advertising solutions.
Safari and Firefox have blocked third-party cookies by default since 2019–2020. iOS restricts app identifiers through App Tracking Transparency. Roughly 40%+ of traffic is already beyond the reach of cookie-based segments.
Chrome has not removed third-party cookies, offering user-level controls instead. But consent requirements, signal loss on other browsers, and buyer scrutiny of data provenance all still apply. Cookies on one browser is a partial reprieve, not a strategy.
GDPR, CCPA/CPRA and a growing list of state privacy laws make cross-site profiling consent-dependent, auditable and legally risky. Every consent banner is a leak in segment coverage. Segments derived from content carry none of this exposure.
Cookie-era segment marketplaces were cookie products end to end, and several of the largest wound down entirely. The taxonomies buyers planned against outlived the infrastructure that populated them — leaving a supply gap for granular audience definitions.
Knowing a page is “Automotive” does not tell a buyer whether it reaches first-time car buyers or collectors of vintage engines. The differences between the two approaches, and why audience-level contextual closes the gap, are covered in contextual vs. behavioral targeting.
The pipeline runs from raw page content to a structured, coded audience profile in four stages. It operates in two modes: a real-time API for per-URL granularity, and a precomputed 102M-domain dataset for planning and enrichment at bulk scale.
The system retrieves the page and extracts meaningful text — headlines, body copy, product details, navigation context. This is the entire evidence base. No cookies are read, no pixels fire, no user is present.
The content is classified into standard IAB content categories — the shared language of contextual adtech. This anchors everything downstream. How the taxonomies interlock is explained in our IAB Audience Taxonomy guide.
A language model infers attributes a category label cannot carry: demographics, interests, purchase intent, life stage and B2B signals. Every value carries low / medium / high confidence. A mortgage calculator earns high-confidence intent; a news page earns nothing rather than a guess.
A curated map translates IAB categories into personas from a 1,667-persona taxonomy — deterministically. The same category always yields the same personas, making the layer auditable, stable, and cheap enough to precompute across all 102M domains.
Model inference provides resolution — demographic and intent detail no lookup table can express. Deterministic mapping provides stability — segment definitions that do not drift, so a packaged deal stays consistent for its whole flight. Together, buyers get granular descriptions with a documented derivation trail.
Every attribute is a code from a fixed list — age_bracket: 25_34 INT.travel PI.travel.hotels_and_resorts — never free text. This lets segments aggregate across millions of pages and remain comparable between versions. Browse the full taxonomy.
Every profile describes a page or domain along five families of attributes. All values come from the versioned v1.0 vocabularies; everything except personas is model-inferred with a low/medium/high confidence band.
| Family | Fields & vocabulary size | Example values |
|---|---|---|
| Demographics | 8 age brackets, 5-point gender skew, 6 income bands, 7 education levels, 14 life stages, plus household composition, employment status, home ownership and urbanicity. Full detail in website audience demographics. | 25_34 skews_female middle_income family_young_children |
| Interests | 29 interest groups containing 285 sub-interests, coded as INT.* — the enduring affinities of the audience, distinct from what they are currently shopping for. |
INT.food_drink INT.travel INT.tech_computing |
| Purchase intent | 34 intent groups containing 283 in-market segments, coded as PI.* — the commercial categories the content implies its readers are actively researching. Covered in purchase intent data. |
PI.travel.hotels_and_resorts PI.auto_ownership.new_vehicles PI.finance_insurance.mortgage_lenders_and_brokers |
| B2B firmographics | Whether content is B2B, plus target role/job function, seniority, company-size and industry bands using LinkedIn-standard firmographic bands — fields that drop directly into ABM and B2B media workflows. | is_b2b: true role: it_decision_maker size: 1001_5000 |
| Personas | 1,667 personas, assigned by a deterministic IAB-category-to-persona mapping. Each persona cites the category it was mapped from. See audience personas in advertising. | Home Chef Data Scientist First-Time Homebuyer |
What the pipeline returns for a recipe page — raw coded output on the left, human-readable segment labels on the right.
{
"input": "https://example-cooking-site.com/
weeknight-30-minute-dinners",
"audience_profile": {
"demographics": {
"age_brackets": [
{"code": "25_34", "confidence": "high"},
{"code": "35_44", "confidence": "high"}
],
"gender_skew": {"code": "skews_female",
"confidence": "medium"},
"income_band": {"code": "middle_income",
"confidence": "medium"},
"life_stages": [
{"code": "family_young_children",
"confidence": "medium"}
]
},
"interests": [
{"code": "INT.food_drink.cooking",
"confidence": "high"},
{"code": "INT.home_garden",
"confidence": "low"}
],
"purchase_intent": [
{"code": "PI.food_beverage.food_delivery_services",
"confidence": "medium"},
{"code": "PI.home_garden_services.appliance_repair",
"confidence": "low"}
],
"b2b": {"is_b2b": false, "confidence": "high"},
"personas": [
{"persona": "Home Chef",
"mapped_from": "Food & Drink > Cooking",
"source": "deterministic_mapping"},
{"persona": "Busy Parent Meal Planner",
"source": "deterministic_mapping"}
],
"vocabulary_version": "v1.0"
}
}
Age and cooking interest are asserted at high confidence because the content supports them directly. Kitchen-appliance intent is flagged low because a recipe page only weakly implies it. A curator building a guaranteed deal thresholds at high; a prospecting campaign accepts medium and above. One dataset, multiple precision tiers — no visitor observed.
The two approaches answer differently at every layer. Neither is a strict superset: cookie segments capture individual history where they still function; content segments describe the placement everywhere, unconditionally.
| Dimension | Cookie-based segmentation | Content-based segmentation |
|---|---|---|
| Unit profiled | The individual user, tracked across sites | The page or domain, read once |
| Signal source | Browsing history, ID syncs, data-broker joins | The content itself — what the page is about |
| Coverage today | Absent on Safari, Firefox and iOS — ~40%+ of traffic | Every page on every browser |
| Dependency | Browser policy, consent rates, ID-sync loss | None — works wherever content exists |
| Privacy posture | Personal data; consent, DSR and audit obligations | No personal data; privacy-safe by construction |
| Freshness | Profiles decay as behavior ages and IDs churn | Re-derived whenever the content is re-crawled |
| Explainability | Black-box segment names from vendor pipelines | Every value cites derivation: category, mapping, confidence |
| Granularity | Individual-level, where it still works | Page-level via API; domain-level across 102M domains |
| Retargeting | Yes, on browsers that still permit it | No — never identifies individuals, by design |
Content-based segmentation does not do retargeting, and that is precisely why it has no consent surface and no browser dependency. It replaces the prospecting and audience-planning functions of cookie segments — the majority of what audience data was actually used for — at the unit media is bought in. Targeting mechanics are covered in cookieless targeting.
The same profiles — domain-level for scale, URL-level for precision — plug into every stage of the media workflow, on both buy and sell side.
Slice the 102M-domain dataset by any attribute combination and hand planners a ranked domain list instead of a taxonomy document. The precomputed dataset carries this entire use case.
Curators assemble inventory packages — “pages reaching new parents, high confidence only” — and expose them as Deal IDs in any DSP, with per-URL derivation. See ad inventory curation.
Publishers segment their own inventory and declare audiences in the bidstream — the IAB Tech Lab’s SDA pattern — monetizing audience data without leaking user data. See seller-defined audiences.
Join audience attributes onto any table keyed by domain or URL: enrich a CRM with firmographic profiles, score supply-path logs by audience quality, or add demographic columns to a warehouse.
Target line items against audience attributes per placement in real time — page-level for precision, domain-level for reach — on 100% of traffic, every browser. Setup is in cookieless targeting.
Firmographics use LinkedIn-standard bands, so B2B marketers can find the open-web reading list of their ICP — role, seniority, company size, industry — and run programmatic ABM without any identity graph.
It is the practice of building advertising audience segments from the content of web pages and domains instead of from third-party cookies or user tracking. A model reads what a page is about and infers the audience it predictably attracts — demographics, interests, purchase intent, life stage and B2B firmographics — each with a confidence band. Because no individual is observed, the segments work in every browser, including Safari, Firefox and iOS where third-party cookies are already blocked.
By profiling the page instead of the person. Content predicts its own audience: a weeknight-recipes page is read predominantly by busy home cooks with families; a cloud-infrastructure blog is read by technical decision makers. A language model infers those attributes from the text with banded confidence, and a deterministic mapping converts the page's IAB categories into personas. The output is a structured audience profile per URL or domain — no cookies, device IDs or user data anywhere in the pipeline.
No. Chrome has kept third-party cookies and announced it will continue supporting them with user-level controls rather than removing them. The cookieless shift is driven by the browsers that already block them — Safari and Firefox have done so by default since 2019–2020, and iOS restricts app identifiers — which together account for roughly 40%+ of traffic, plus privacy regulation that makes cross-site tracking consent-dependent and costly.
It describes the aggregate audience of a page, which is the unit media is actually bought in, and it is explicit about certainty: every model-inferred attribute carries a low, medium or high confidence band, and attributes the content does not support are omitted rather than guessed. Users set their own thresholds per use case — high-confidence only for guaranteed audience deals, medium and above for prospecting reach. Personas come from a deterministic category-to-persona mapping that is fully auditable.
Classic contextual targeting stops at the topic: the page is about “Automotive.” Cookieless audience segmentation goes further and describes the people the topic attracts: age brackets, gender skew, income band, life stage, in-market segments and job function, drawn from controlled vocabularies aligned with IAB Audience Taxonomy 1.1. Both are content-derived and privacy-safe; the difference is that one outputs a category and the other outputs a targetable audience definition. See our contextual vs. behavioral targeting guide.
The segmentation itself processes no personal data — its inputs are public page content, and its outputs describe pages, not people — so building and using the segments does not depend on tracking consent. That removes the coverage loss from consent banners and the exposure of cross-site profiles. Publishers and advertisers should still assess their overall ad stack with their own counsel, but the audience data layer itself is privacy-safe by construction.
This pillar is part of a ten-guide hub. Each guide goes deep on one layer of the stack.
Open the interactive demo, paste any page or domain, and watch the full field set — demographics, interests, intent, firmographics and personas — come back with confidence bands.