How publishers and sellers package their own audience segments, declare them in the bidstream with standardized IAB taxonomy IDs, and sell audience without exposing a single user identity — and where the audience data behind credible SDA segments actually comes from.
Seller Defined Audiences (SDA) is an IAB Tech Lab specification that lets publishers and other sellers package their own first-party audience segments and declare them directly in programmatic bid requests — using standardized segment IDs from the IAB Audience Taxonomy — without exposing any user identity to the buyer. Instead of a data broker following users across the web and selling the resulting profile, the seller asserts “this impression reaches business decision makers” or “this impression reaches in-market travel intenders,” and the buyer's DSP reads that assertion as a structured, taxonomy-coded signal. Released by IAB Tech Lab's Addressability Working Group in early 2022 as part of the Project Rearc effort, SDA is the supply side's structural answer to cookieless audience segmentation: it moves audience packaging from the buy-side data marketplace, which depended on third-party cookies, to the sell side, which never needed them.
Three ideas make the spec work. First, standardization: segments are not free-text labels a publisher invents, but IDs drawn from a shared vocabulary — the IAB Audience Taxonomy, with its demographic, interest and purchase-intent branches — so that “Business Decision Makers” means the same thing from every seller. Second, declaration without identity: the segment IDs travel in the bid request itself, attached to the impression, so no cookie sync, no identity graph and no user-level data leaves the seller's environment. Third, accountability: the existing supply-chain transparency stack — sellers.json and the SupplyChain object — tells the buyer exactly who declared the segment, so trust attaches to a named, auditable seller rather than to an anonymous data pipeline.
That combination sounds simple, and mechanically it is. The hard part of SDA is not the plumbing — it is the audience data a seller needs before it can declare anything credible. This guide walks through the spec end to end, then addresses that data problem directly, because it is where most SDA initiatives actually stall.
For most of programmatic history, audience was a buy-side product: third-party data vendors observed users across sites, packaged the behavior into segments, and sold them to advertisers through DSPs. Every link in that chain assumed cross-site identity. SDA exists because that assumption stopped holding.
Safari and Firefox have blocked third-party cookies by default since 2019–2020, and iOS restricts app-level identifiers. Chrome, by contrast, has kept third-party cookies and plans to continue supporting them with user-level controls — but the blocking browsers alone put roughly 40%+ of traffic beyond the reach of cookie-based audience data today.
Third-party segment vendors were cookie products end to end — several major providers scaled their advertising data businesses down or exited entirely as the signal beneath them eroded. Buyers still plan campaigns against audiences; the supply of tradable audience data shrank underneath them.
GDPR, CCPA and their successors make cross-site tracking consent-dependent, auditable and expensive. Data that never leaves the publisher's own environment — and signals that describe impressions rather than people — carry structurally less regulatory exposure than shipped user profiles.
Publishers always had the most defensible audience knowledge — their own readers on their own pages — but no standardized way to sell it programmatically. Bespoke segment names in ad server keys don't scale across a thousand DSP seats. SDA gives that knowledge a common wire format.
The logic of SDA, then, is a transfer of responsibility: if cross-site observation of users is going away, let audience be asserted by the party that still legitimately knows it — the seller with the direct relationship to the content and its readers — and give buyers standardized vocabulary and supply-chain transparency so the assertion can be trusted and priced.
SDA is not one invention but an assembly of existing IAB Tech Lab standards — a taxonomy, an OpenRTB extension, and the supply-chain transparency layer — wired together into a four-step flow.
The seller assembles evidence of who its audience is: registration and subscription data, survey panels, analytics, contextual signals from the content itself, or licensed content-inferred audience data. This layer is the seller's own — the spec deliberately does not dictate methodology, only how the result is declared.
Each cohort is expressed as one or more standardized segment IDs from the IAB Audience Taxonomy (version 1.1), whose three branches cover demographics, interests, and purchase intent. Standard IDs are what make a segment tradable: every DSP resolves the same ID to the same meaning.
The segments travel in OpenRTB's user.data array (audience signals) or site.content.data (content signals), each block naming its source and carrying a segtax extension that identifies which taxonomy the IDs come from — segtax: 4 designates IAB Audience Taxonomy 1.1. No user ID accompanies the declaration.
Buyers verify who declared the segment via sellers.json and the SupplyChain (schain) object, and can consult the seller's published transparency documentation describing how segments were built. DSPs that read the signals target, price and report against them like any other audience.
Two details of the wire format matter in practice. First, the name field of each data block identifies the domain of the party asserting the segment — the publisher itself, or a data partner working on its behalf — which is what lets a buyer cross-reference the assertion against sellers.json and decide whose declarations to trust. Second, because segtax is explicit, sellers can send multiple signal sets side by side: audience segments coded to Audience Taxonomy 1.1 and content categories coded to the Content Taxonomy can coexist in one request, each unambiguous. IAB Tech Lab's companion Data Transparency Standard supplies the “nutrition label” format sellers use to disclose segmentation methodology — source of data, recency, and how membership is derived — so a buyer can distinguish a survey-calibrated segment from a purely declared one.
Note what is absent from the flow: cookie syncing, identity graphs, hashed emails, fingerprinting. SDA's privacy posture comes from architecture, not policy — the audience assertion is a property of the impression, computed inside the seller's environment, and the buyer never receives material that identifies a person.
SDA is a chain: it only pays off when every link — from the party that knows the audience to the party spending against it — handles the signal.
The originators. They own the audience relationship, build or license the underlying data, choose the taxonomy IDs, and declare the segments. For publishers, SDA converts audience knowledge into directly monetizable inventory attributes — and keeps the data premium on their side of the trade.
The carriers. Supply-side platforms pass the seller's user.data signals through in bid requests, expose SDA-based packages as Deal IDs, and increasingly offer tooling for sellers to manage segment declarations across their inventory.
The aggregators. Curators assemble SDA-signaled inventory from many sellers into cross-publisher audience packages — scale a single publisher cannot offer — and sell them as deals. The workflow is the subject of our ad inventory curation guide.
The demand. DSPs parse the segtax-coded segments, surface them as targetable audiences, and let buyers bid on them. Buy-side adoption is SDA's historical bottleneck — which is why transparency, standard taxonomy IDs and verifiable supply chains matter more than any other feature of the spec.
Here is the problem the spec cannot solve on its own: a segment ID is an assertion, and assertions need evidence. A publisher that declares “Business Decision Makers” on every impression across its whole network — because the label prices well — will be found out the first time a buyer's outcome data disagrees, and the discount that follows attaches to the seller's name permanently. Conversely, a publisher with genuinely valuable audiences but no data infrastructure has nothing to declare at all. Logged-in registration data covers only a fraction of most publishers' traffic; survey panels are expensive and sparse; and long-tail sections of even large publisher networks have no first-party audience evidence whatsoever.
This is where content-inferred audience data enters the SDA workflow. The audience of a page is predictable from what the page is about: a section of mortgage-calculator guides is read by mortgage intenders, a cloud-infrastructure blog by senior technical staff, a sleep-training column by new parents. Our contextual audience segmentation engine computes exactly this evidence — demographics, interest groups, purchase-intent segments, life stages and B2B firmographics inferred from content, each with a banded low/medium/high confidence — at two resolutions that match how SDA is deployed: a precomputed domain-level dataset covering 102M domains for network-wide segment planning, and a per-URL real-time API for page-level declaration granularity.
Critically for SDA, the output is already speaking the right language. Every attribute comes from a controlled, versioned vocabulary (v1.0) aligned with IAB Audience Taxonomy 1.1 — the same taxonomy the SDA spec references for segment IDs: 8 age brackets, a 5-point gender skew, 6 income bands, 7 education levels, 14 life stages, 29 interest groups with 285 sub-interests (INT.* codes), and 34 purchase-intent groups with 283 segments (PI.* codes). Mapping a content-inferred profile to a declarable Audience Taxonomy segment is therefore a vocabulary translation, not a research project; the full structure is browsable on the audience segmentation taxonomy page. And because the inference describes pages rather than people, it extends SDA coverage to the traffic first-party data never reaches: anonymous visitors, unauthenticated sections, and newly added network domains — while confidence bands give the publisher an honest, documentable basis for what it declares. Declare only high-confidence attributes on guaranteed deals; accept medium-and-above for broader packages; never declare what the evidence does not support.
A publisher network wants to package its business, finance and enterprise-technology inventory as a Business Decision Makers SDA deal. Step one is evidence: profile the candidate inventory. Below, the content-inferred profile for one section of one network domain — raw coded output left, rendered segment labels right.
{
"input": "https://example-biz-publisher.com/
enterprise-software/",
"audience_profile": {
"demographics": {
"age_brackets": [
{"code": "35_44", "confidence": "high"},
{"code": "45_54", "confidence": "medium"}
],
"income_band": {"code": "upper_middle_income",
"confidence": "medium"}
},
"interests": [
{"code": "INT.business_finance.business",
"confidence": "high"},
{"code": "INT.personal_finance",
"confidence": "low"}
],
"purchase_intent": [
{"code": "PI.software.computer_software",
"confidence": "high"}
],
"b2b": {
"is_b2b": true, "confidence": "high",
"target_roles": ["it_manager", "director"],
"company_size": "201_500"
},
"personas": [
{"persona": "IT Decision Maker",
"source": "deterministic_mapping"}
],
"vocabulary_version": "v1.0"
}
}
The section qualifies: B2B content at high confidence, senior roles, enterprise-software intent. The publisher repeats this across the network — using the domain-level dataset for breadth and the per-URL API for page-level precision — and admits into the package only inventory whose profile clears the threshold: is_b2b high, plus a business interest or enterprise purchase-intent signal at high confidence.
// Bid request fragment (segment IDs illustrative) "user": { "data": [{ "name": "example-biz-publisher.com", "segment": [ {"id": "553"}, // interest branch {"id": "1216"} // purchase-intent branch ], "ext": {"segtax": 4} // Audience Taxonomy 1.1 }] }, "source": { "ext": {"schain": { /* verifiable seller chain */ }} }
The qualifying inventory is mapped to Audience Taxonomy 1.1 segment IDs in the interest and purchase-intent branches, declared in user.data with segtax: 4, and exposed to buyers as a Deal ID — “Business Decision Makers, network-wide.” The name field ties the declaration to the seller's sellers.json identity; the SupplyChain object makes the path auditable.
What the publisher can now show a skeptical buyer is the decisive part: a per-URL derivation trail — this page, this content, these inferred attributes at these confidence levels, mapped to these taxonomy IDs — instead of “trust our segment name.” That evidentiary story, documented in the transparency label, is the difference between an SDA deal that commands a premium and one that gets priced as run-of-network.
Three ways audience reaches a programmatic buyer, compared. They are not mutually exclusive — sophisticated sellers run SDA on top of content-derived evidence, which is effectively the third column powering the first.
| Dimension | Seller-defined audiences | Buyer-side (third-party) audience data | Contextual / content-based deals |
|---|---|---|---|
| Who packages the audience | The publisher or seller | Data vendors and DMPs on the buy side | Either side, from page/domain signals |
| Underlying data | Seller's first-party and content-inferred data | Cross-site behavioral observation | What the content is about and who it predictably attracts |
| Identity dependence | None exposed — segments describe impressions | Requires cookies or ID graphs to build and match | None — describes pages, not people |
| Cookieless coverage | Full, wherever the seller declares | Degrades with blocked cookies and consent loss | Full — every page on every browser |
| Where it lives in the transaction | user.data + segtax in the bid request; Deal IDs | DSP audience marketplace, matched pre-bid | Deal IDs, allowlists, pre-bid contextual segments |
| Standard vocabulary | IAB Audience Taxonomy 1.1 segment IDs | Vendor-proprietary taxonomies | IAB Content Taxonomy; audience attributes if enriched |
| Transparency | sellers.json, SupplyChain, transparency labels | Historically opaque methodology | Fully explainable — derivation from the content itself |
| Main risk | Over-declaration eroding buyer trust | Shrinking signal, regulatory exposure | Coarse if it stops at topic level rather than audience level |
| Who captures the data value | The seller | The data vendor | Whoever holds the enrichment |
The takeaway from the table is the diagonal relationship: buyer-side audience data is the model in structural decline, contextual inference is the evidence layer that scales without identity, and SDA is the standardized rail that lets sellers monetize that evidence. How buyers activate the same signals from their own side is covered in cookieless targeting.
Seller Defined Audiences (SDA) is an IAB Tech Lab specification, released in 2022, that lets publishers and sellers package their own first-party audience segments and declare them in OpenRTB bid requests using standardized segment IDs from the IAB Audience Taxonomy. Buyers receive a structured audience signal attached to the impression — for example “business decision makers” or “in-market for travel” — without any user identity, cookie sync or ID graph being involved.
Through OpenRTB's data objects: audience segments travel in the user.data array and content signals in site.content.data. Each data block names the asserting party and carries a segtax extension identifying the taxonomy the segment IDs belong to — segtax: 4 designates IAB Audience Taxonomy 1.1. Buyers verify who made the declaration through sellers.json and the SupplyChain object.
No. SDA was designed so that no third-party cookie, device ID or cross-site identifier is needed at any point: segments are computed inside the seller's environment and declared as properties of the impression. That makes SDA fully functional on Safari, Firefox and iOS, where third-party cookies are already blocked — roughly 40%+ of traffic — and independent of whatever Chrome, which has kept third-party cookies, does next.
The IAB Audience Taxonomy, version 1.1, whose three branches cover demographics, interests and purchase intent. Standardized IDs are what make seller-declared segments tradable across every SSP and DSP. Our own audience vocabularies — 29 interest groups with 285 sub-interests, 34 purchase-intent groups with 283 segments, plus demographic and life-stage attributes — are versioned and aligned with Audience Taxonomy 1.1, so content-inferred profiles translate directly into declarable SDA segment IDs. See the IAB Audience Taxonomy guide for the full structure.
They are complementary layers. SDA is the signaling standard — how a seller declares audience on impressions. Curation is a packaging workflow — assembling inventory, often from many sellers, into a Deal ID a buyer can transact. A curated audience deal frequently uses SDA-style signals as its inputs, and cross-publisher curation is how seller-declared audiences achieve the scale a single publisher cannot offer. Our ad inventory curation guide covers the packaging side in depth.
From three sources, usually combined: authenticated first-party data (registrations, subscriptions), research panels, and content-inferred audience data — profiling what each page or domain is about and inferring the audience it predictably attracts, with confidence bands. Content inference matters most in practice because it covers the anonymous and long-tail traffic that first-party data misses, provides a per-URL derivation trail to show buyers, and outputs attributes aligned with the same IAB Audience Taxonomy 1.1 that SDA declarations reference.
Open the interactive demo, paste any page or domain from your network, and see the full content-inferred audience profile — demographics, interests, intent, firmographics and personas with confidence bands — the evidence layer behind a defensible SDA segment.