A passback is an ad tag returned in place of an advertisement. When a network, exchange or ad server cannot fill a request, because no campaign is eligible, no bid clears the floor, the geography is wrong or the category is blocked, it can answer with markup that fires a fresh ad call somewhere else. The impression is handed on rather than dropped. Microsoft's Xandr glossary defines the term as "the redirecting of an impression back to an ad server when no acceptable bid is received from a mediated bid, in order to allow the next highest bidder a chance to win the impression."

The mechanism exists because an empty ad slot earns nothing. Publishers have always operated more inventory than any single buyer wants, and the surplus has to be offered repeatedly until someone takes it.

How a passback executes

The sequence starts with an ordinary ad request. A page loads, the ad server tag calls the first demand source, and that source evaluates the opportunity. If it declines, the response carries not a creative but another tag: a script, an iframe, or in video an XML document pointing at a different endpoint. The browser executes it, a second request leaves the device, and the cycle repeats until something fills or the chain runs out. The last position holds a default creative, usually a house ad or an instruction to collapse the slot.

Google's Publisher Tag documentation describes passback tags as tags that "allow a third-party ad server to ultimately fill an ad request with an ad from your Google Ad Manager network." Implementation is a normal googletag.defineSlot() call with one hard requirement: the tag must render inside an iframe, so it does not inherit page-level settings from other instances of the library. Because the iframe conceals the parent document, the page URL has to be supplied manually through a page_url attribute, and Google states that omitting it can reduce fill rates and CPMs, since Ad Exchange and AdSense lose contextual signals. Click tracking needs a clickUrl value carrying the upstream macro, commonly %%CLICK_URL_UNESC%%. Privacy flags do not travel automatically either.

Video works differently. A VAST response can be a wrapper, an XML file containing no media at all, only a redirect to another VAST endpoint. The IAB Tech Lab's VAST 4.3 specification, released in December 2022, states that wrappers "should be limited to five before resulting in an InLine response," with error code 302 reserved for a breached wrapper limit and 301 for a wrapper URI that times out.

In apps the equivalent is a mediation adapter. Google's mobile ads SDK returns error code 3, defined in AdMob documentation as an outcome where "the ad request was successful, but no ad was returned due to lack of demand," and the mediation layer moves to the next adapter. AdMob's help pages describe waterfall mediation as calling ad sources "one-by-one based on the average eCPM you set, not what the ad source is willing to pay."

Server-to-server exchange traffic does not use passbacks in the same sense. Under OpenRTB, a bidder with nothing to offer returns HTTP 204 with an empty body, or a no-bid reason code from the nbr list of 18 standard values plus exchange-specific codes above 500. The exchange decides what happens next, not the browser.

Origin and evolution

Passbacks are older than programmatic. They belong to the ad network era of the late 1990s and 2000s, when publishers sold premium inventory directly and pushed the remainder through a queue of networks ranked by expected yield. The arrangement was called a waterfall, or a daisy chain. The yield optimisation platform Admeld operated inside exactly that structure: the impression was offered to each network in turn, and a network that declined, or could not meet the configured floor, returned a passback tag handing the impression to the next tier, continuing until something filled or a house ad ran.

The technique is documented in patent filings from the period. Right Media filed application 11/776,150 on July 11, 2007, published on January 15, 2009, describing daisy-chain networks, passback tags and a method for detecting that an ad call originated from a previous default, then excluding the network that defaulted from the auction for the re-offered impression.

Header bidding attacked the sequence itself. Prebid.js was created in early 2015 by Matt Kendall and Paul Yang at AppNexus with Nick Jacob of the publisher Aplus, replacing serial calls with a parallel auction resolved before the ad server was invoked. Version 1.0 arrived on December 13, 2017, unifying banner, video and native ad units. Google answered with Exchange Bidding, announced in open beta on June 8, 2017, moving the same competition server-side and removing the need to maintain, as the company put it, "hundreds or thousands of new line items."

Google then began retiring the browser-side helpers. Release notes show a recommended passback snippet published in the week of November 11, 2019, and the definePassback() and defineOutOfPagePassback() methods deprecated in the week of August 31, 2020, on the grounds that they behave synchronously and can be blocked by some browsers. Mobile followed later: DV360 stopped bidding on multiple waterfall calls to Google Ad Manager and the AdMob Network for app inventory on October 23, 2024, forcing publishers to rebuild ad stacks around real-time bidding.

Passbacks did not disappear. They migrated into hybrid setups, most visibly post-bid, a Prebid configuration where the entire library loads inside an ad server creative after allocation. The reference implementation uses a hidden zero-by-zero iframe and a PREBID_TIMEOUT set to 1,000 milliseconds by default; when no bid returns, the creative falls through to a passback.

Why the mechanism matters commercially

Every hop costs time and money. A passback chain adds a full network round trip per tier, executed while the slot sits blank, and each hop hands the impression to a party the previous one does not control. IAB Spain's first SSP guide, published on April 15, 2026, found that only 41% of programmatic spend reaches genuine measurable inventory, and put the difference between routes bluntly: direct SSP connections deliver 70% to 80% of an advertiser's euros to publishers, while indirect paths through multiple intermediaries deliver 40% to 50%.

Reporting suffers too. Google Ad Manager defines an unfilled impression as "an ad request that did not return a line item," and recommends either removing the slot or trafficking remnant and house inventory against it. An impression passed to a second server is counted once by each, a standard source of discrepancy between platforms.

Limitations and disputes

Latency is the oldest complaint and the least disputed. Sequential calls execute one after another, so the worst case is the sum of every timeout in the chain, and video compounds it: five wrapper redirects each carry their own connection delay before a single frame plays.

Price blindness is the structural criticism. A waterfall ranks sources by historical average eCPM, not by what any would pay for the impression in front of them, so a network placed third may have been willing to outbid the network placed first. Header bidding was built on that observation.

Supply-chain integrity is the newer objection. Passbacks obscure origin: the iframe requirement strips the page URL unless a publisher restores it, and each additional hop adds a seller that must be declared in ads.txt and sellers.json and represented in the SupplyChain object for buyers to validate the route. Chains assembled by resellers are where those declarations break.

Cost has now been attached to the volume passbacks generate. PubMatic began charging publishers $0.001 CPM on inventory exceeding undisclosed daily request caps under a policy effective April 16, 2026, with enforcement emails arriving in June. Practitioners objected that the company does not distinguish low-quality supply from legitimate volume. The underlying economics are not contested: requests that never convert into monetised impressions still consume infrastructure.

Disambiguation

No fill is the outcome, not the mechanism: a request that returned no advertisement. A passback is one possible response to that state; the alternatives are a default creative, a collapsed slot or nothing at all.

Backfill is a commercial category, naming the demand, typically network or exchange inventory, used to monetise what direct sales left unsold. Passbacks are frequently how backfill is reached, but a header bidding auction reaches it without one.

A default creative terminates a chain. Xandr's glossary describes it as a creative that runs when no others are available, either a house ad or a redirect to a third-party tag. Where a passback continues the sequence, a default ends it.

Post-bid is an auction, not a redirect. It runs demand partners inside an ad server creative after allocation, and uses a passback only as its fallback when no bid returns.

Recent developments

The industry is now reducing unfilled requests rather than routing them onward. Google added a "fill empty in-page ads" setting to AdSense on July 3, 2025, placing contextual content suggestions in slots that returned no advertisement and switching it on automatically over 30 days. AdMob's mediation update on July 7, 2025 brought its bidding partner count to seven, continuing the shift of app demand out of ranked waterfalls.

Filtering has become the preferred alternative to passing impressions along. OpenX integrated Amazon machine learning to suppress low-value impressions on December 22, 2025, and Mediavine and SWYM extended pre-bid supply shaping across more than 18,000 publisher sites on June 16, 2026. In apps, Amazon's APS bridge with InMobi, introduced on May 28, 2026, opened 28 additional bidders without new SDK integrations.

What survives is the tag itself. Passbacks still connect ad servers across publisher boundaries, carry video demand between platforms that do not share an auction, and sit at the end of post-bid creatives. The waterfall that made them ubiquitous is largely gone. The redirect it depended on is not.

Timeline

  • Late 1990s: Ad networks and publisher ad servers establish sequential fallback tags for unsold inventory
  • July 11, 2007: Right Media files patent application 11/776,150 describing daisy chains, passback tags and exclusion of defaulting suppliers from re-auctions
  • January 15, 2009: The application publishes
  • Early 2015: Prebid.js is created at AppNexus, replacing sequential calls with a parallel header auction
  • June 8, 2017: Google announces the open beta of Exchange Bidding, moving competition server-side
  • December 13, 2017: Prebid.js 1.0 unifies banner, video and native ad units
  • Week of November 11, 2019: Google publishes a recommended snippet for building passback tags with the Publisher Tag library
  • Week of August 31, 2020: definePassback() and defineOutOfPagePassback() are deprecated as synchronous methods vulnerable to browser blocking
  • December 2022: VAST 4.3 sets five as the recommended ceiling on wrapper redirects, with error code 302 for breaches
  • October 23, 2024: DV360 stops bidding on multiple waterfall calls for mobile app inventory
  • July 3, 2025: AdSense begins filling empty in-page ad slots with contextual suggestions
  • April 15, 2026: IAB Spain publishes an SSP guide measuring working media at 41% of programmatic spend
  • April 16, 2026: PubMatic's excess inventory fee of $0.001 CPM takes effect
  • May 19, 2026: Google's passback tag documentation is last updated, retaining the iframe requirement and deprecation notice

Summary

Who. Publishers and their ad operations teams configure passback tags; ad networks, exchanges, SSPs and third-party ad servers return them when they cannot fill. Buyers inherit the consequences through longer supply paths. Google, through the Publisher Tag library, Ad Manager and AdMob, defines most of the implementation detail in use, and the IAB Tech Lab governs the video equivalent through VAST wrappers.

What. A passback is an ad tag served instead of a creative, redirecting an unfilled impression to another demand source or back to the publisher's own ad server. It converts a single ad slot into a sequence of offers ending in a fill, a default creative or an empty space.

When. The technique dates from the ad network era of the late 1990s and 2000s, was documented in a Right Media patent application filed in July 2007, and lost its central role after header bidding arrived in 2015 and Google moved app demand to real-time bidding in October 2024.

Where. Passbacks operate in the browser through iframed tags, in video players through VAST wrapper redirects, in mobile apps through mediation adapters, and inside ad server creatives in post-bid setups. They do not operate inside server-side OpenRTB auctions, where a bidder simply returns HTTP 204.

Why. Unfilled inventory generates no revenue, and no single demand source fills everything. Passbacks buy a second chance at each impression, at the cost of latency, reporting discrepancies, weakened contextual signals and a supply chain that is harder to verify - which is why the industry has spent a decade replacing them with parallel auctions and pre-bid filtering rather than longer chains.