Backfill is the demand a publisher falls back on when nothing better has bought an impression. A page loads, the ad server checks whether a directly sold campaign or a high-value programmatic bid is eligible, and when none is, the request moves to a source that will nearly always return a creative: an exchange, an ad network, a mediation partner, or the publisher's own promotional slot. The logic is arithmetic rather than strategic. An unfilled slot earns nothing at all, so almost any price beats the alternative. Google states the position without decoration in its Ad Manager documentation, describing Ad Exchange line items as a way to monetise inventory "sometimes called 'backfill' inventory" that a publisher finds less valuable or is "simply unable to sell".

How the ad server decides

Every impression starts as an ad request. Inside Google Ad Manager, whose publisher ad server a federal court found held 91% of its market, that request is matched against line items, and each line item type carries a numeric priority in which lower numbers win. Sponsorship sits at priority 4. Standard splits across 6, 8 and 10. Network, Bulk and Price Priority all sit at 12, alongside the AdSense and Ad Exchange line item types. House sits at 16, and Google's documentation states that it serves only when remnant line items, Ad Exchange or Open Bidding demand are unavailable.

Backfill is not itself a line item type. It is the position those numbers create. Guaranteed transactions clear first, because the publisher has promised delivery and owes make-goods if it misses. Non-guaranteed demand competes underneath on price, ranked by CPM or by an assigned value CPM within the same priority band. A Price Priority line item entered with no rate is treated as House, which silently demotes a paying deal to the floor of the stack.

When nothing serves at all, the ad server logs an unfilled impression, which Google defines as "an ad request that did not return a line item" because none was eligible. Publishers can then leave the slot blank, collapse it, or hand the request to another system.

Dynamic allocation and the moving rung

The oldest form of backfill was a fixed rung: an ad network sat at a known position and received whatever the tiers above it declined. That arrangement broke when exchanges began pricing each impression individually.

DoubleClick opened its Ad Exchange on 18 September 2009, promising "real-time dynamic allocation to maximise yield" so that publishers could "automatically generate the highest return for every impression, using real-time data and bids". Dynamic allocation compares a live exchange bid against the value of whatever else could serve, rather than letting the exchange wait in a queue. Under that system the exchange stops being a fallback and becomes a competitor, capable of outbidding a booked remnant campaign for a single impression while losing the next one.

The mechanism retains an odd corner. Google notes that when AdSense and Ad Exchange line items compete against each other, the winner is chosen at random and is not affected by historic CPM values under dynamic allocation.

Pricing changed the arithmetic again. Google began rolling out a unified first-price auction in Ad Manager in the week of 5 September 2019, replacing the second-price mechanics that had allowed its exchange to evaluate a bid after seeing competing prices.

The waterfall, the passback and the daisy chain

Before real-time bidding, backfill ran sequentially. Publishers ranked ad networks by historic yield and offered each impression to them in order, a structure known as the waterfall or the daisy chain. A network that declined, or failed to clear a price floor, returned a passback tag that redirected the browser to the next tier. Each hop added latency and leaked value, and the ranking rested on last month's average CPM rather than the worth of the impression in hand.

Yield optimisers grew out of that inefficiency. Admeld, founded in 2007, replaced a publisher's stack of network tags with a single interface and reordered demand predictively by polling network APIs. Real-time bidding accounted for roughly 20% of Admeld's impression volume in 2010 and, for some publishers, as much as 60% of revenue by mid-2011. Google announced its acquisition on 13 June 2011, closed it on 6 December 2011 for about $400 million, and shut the product down by late 2013.

Backfill in apps and licensed feeds

Mobile applications kept the waterfall long after the open web abandoned it, because SDK integrations are harder to replace than page tags. AdMob mediation still exposes both models, and a July 2025 update let publishers move line items between separate Bidding and Waterfall ad source tabs, added bidding eCPM floors at mediation group level, and brought InMobi, ironSource and Unity into in-app bidding, with Moloco and LINE following. One case study cited a 36% revenue increase and a 9% eCPM improvement after the reordering.

Buy-side platforms have begun refusing to participate in sequential fallback at all. Display & Video 360 stopped bidding on multiple waterfall calls for mobile app inventory on 23 October 2024, a month after announcing the change, on the grounds that the same impression was arriving several times across different tiers.

Licensed content platforms operate backfill as a commercial default. Apple sells unsold inventory in Apple News on behalf of publishers, paying them 70% of revenue from backfill ads, switching the setting on automatically 48 hours after a channel is linked, and restricting it to channels rated General.

Why it matters to buyers

Backfill is where a large share of cheap open-web and in-app impressions originates, which makes it a quality problem rather than a pricing curiosity. Inventory reaching the bottom of a stack is, by construction, inventory that better-informed buyers declined. Advertisers running broad open exchange campaigns without inventory controls buy disproportionately from that layer, which is also where made-for-advertising sites and low-value placements concentrate.

Volume compounds the issue. PubMatic processed 56 trillion impressions in the third quarter of 2022, up 33% year on year and equivalent to roughly 7,000 ads for every person on the planet, and much of that traffic never transacts. Publishers have started throttling their own bid requests rather than sending everything to everyone, with Chegg cutting off a slot after 20 or more requests returning zero bids.

Limitations and disputes

The sharpest criticism is that whoever controls the fallback controls the auction. Judge Leonie Brinkema found on 17 April 2025 that Google had monopolised the publisher ad server and open-web ad exchange markets, and her opinion identified First Look, Last Look and Unified Pricing Rules as practices that disadvantaged rivals. Each of those turned on sequencing: which exchange saw an impression first, which saw it last, and which prices a publisher was allowed to set for whom. Remedies remain unresolved. At closing arguments on 21 November 2025 the judge questioned whether divesting AdX could be enforced during an appeal, while the Department of Justice argued that behavioural remedies would generate permanent litigation over technical detail.

Cost is the second objection. Every backfill call consumes infrastructure whether or not it fills. PubMatic began enforcing daily impression and bid request caps in mid-June 2026 under a supply policy effective 16 April 2026, charging an excess inventory fee of $0.001 CPM rounded up to the nearest $10 per month, with domain blocking for persistent overages. Publishers reported having no visibility into their allocated caps before the notices arrived.

A third dispute concerns whether filling a slot is always worth it. Low-CPM backfill trains buyers to expect cheap access, depresses floor prices, and can degrade the reading experience for revenue measured in fractions of a cent.

Disambiguation

No fill describes the outcome, not the remedy. A no fill is an ad request that returned nothing; backfill is the demand called on to prevent it.

A passback is the mechanism that moves a declined impression onward, typically a redirect tag. Backfill is what receives it.

Remnant describes unsold inventory as a commercial category. Backfill describes the serving position used to monetise it, and remnant campaigns booked at a fixed CPM can themselves be outbid by exchange backfill under dynamic allocation.

House ads are a publisher's own promotions, sitting below paid backfill at priority 16 in Ad Manager and earning nothing directly.

Data backfill, common in analytics and reporting contexts, means retroactively populating historical records. It shares only the word.

Recent developments

Fallback logic is being pushed further down the stack, into the space after every buyer has declined. AdSense began rolling out a feature on 3 July 2025 that populates empty in-page slots with contextual suggestions, showing ads once a user interacts with them, activated by default within 30 days and controlled through the Ad serving section of blocking controls. Google's justification was blunt: "You don't make any revenue from an unfilled ad space."

Open-source tooling is moving the other way, toward parallel competition rather than ordered fallback. At a webinar on 23 October 2025, Prebid positioned its Mobile SDK as an additional demand source rather than a mediation replacement, with German weather publisher WetterOnline reporting up to a 10% CPM uplift after upgrading to version 3.0, and a 4.0 roadmap aimed at publisher control over closed mediation systems.

Timeline

  • 2007: Admeld is founded to consolidate publisher ad network tags and reorder demand predictively
  • 18 September 2009: DoubleClick Ad Exchange opens, offering real-time dynamic allocation
  • 2010: Real-time bidding accounts for roughly 20% of Admeld's impression volume
  • 13 June 2011: Google announces its acquisition of Admeld
  • 6 December 2011: The Admeld deal closes for approximately $400 million
  • Late 2013: Admeld is shut down
  • Week of 5 September 2019: Google begins rolling out a unified first-price auction in Ad Manager
  • Third quarter of 2022: PubMatic processes 56 trillion impressions, up 33% year on year
  • 23 October 2024: Display & Video 360 stops bidding on multiple waterfall calls for mobile app inventory
  • 17 April 2025: A federal court finds Google monopolised the publisher ad server and ad exchange markets, citing First Look, Last Look and Unified Pricing Rules
  • 3 July 2025: AdSense begins filling empty in-page ad slots with contextual suggestions
  • 7 July 2025: AdMob separates bidding and waterfall ad sources and adds bidding eCPM floors
  • 23 October 2025: Prebid outlines a Mobile SDK 4.0 roadmap targeting control over closed mediation
  • 21 November 2025: Closing arguments conclude in the remedies phase of the US ad tech case
  • 16 April 2026: PubMatic's supply policy introducing impression and bid request caps takes effect

Summary

Who. Publishers and their ad operations teams configure backfill through line item priorities, mediation groups and fallback tags. Exchanges, ad networks, mediation platforms and content licensors such as Apple supply the demand. Advertisers buying broad open-market inventory are the main purchasers of it, usually without labelling it as such.

What. Backfill is the demand called on to fill an impression that higher priority or higher paying sources declined. It is a position in a decision sequence rather than a product, expressed in ad servers as low-priority line items, in apps as waterfall tiers or mediation groups, and in licensed environments as a platform-sold default.

When. The practice predates real-time bidding and ran as a sequential waterfall through the 2000s. Dynamic allocation from September 2009 replaced fixed rungs with live price comparison, header bidding and in-app bidding eroded the waterfall through the 2010s, and buy-side platforms began refusing duplicate waterfall calls from October 2024.

Where. In publisher ad servers, in app mediation SDKs, and in platform-operated inventory such as Apple News, where the host sells what the publisher could not.

Why. Unfilled inventory earns nothing, so publishers accept low prices rather than none. The consequences reach further than yield: the fallback layer concentrates low-value and made-for-advertising inventory, generates bid request volume that sell-side platforms are now charging for, and gave the sequencing advantages that a federal court found unlawful in April 2025.