A mid-roll is an advertising break placed inside a piece of video or audio content, after playback has started and before it finishes. It exists because the two alternatives spend attention badly. An advertisement before the content competes with a viewer who has not yet committed to anything, and one after the content reaches only the minority who stay to the end. By the time a mid-roll fires, the audience has already chosen to be there and has something to come back to. That single behavioural fact drives the pricing, the technical plumbing and most of the arguments attached to the format.

How a break is signalled

The instruction that a break exists at a given point is separate from the advertisement that fills it. That separation is the core of the design.

The scheduling layer is usually the Video Multiple Ad Playlist, or VMAP, published by the Interactive Advertising Bureau on 19 July 2012 and never revised since. A VMAP document is a list of AdBreak elements. Each carries a timeOffset, a breakType and an optional breakId. The offset accepts four forms: a timecode in hh:mm:ss.mmm, the literal strings "start" or "end", a percentage of content duration, or a #m sequence pointer to a cue point the player already knows about. Anything that is not "start" or "end" is, by definition, a mid-roll. A break at timeOffset="00:14:20.000" and a break at timeOffset="25%" are both mid-rolls; they simply resolve differently against a 40-minute episode.

The creative layer is the Video Ad Serving Template, or VAST, referenced from each break either as an inline document or as an AdTagURI pointing at an ad server. VAST 3.0, released in 2012, introduced ad pods, meaning a set of sequential linear advertisements delivered into a single break. The allowMultipleAds attribute in VMAP tells the player whether to honour a pod or take one advertisement and move on.

Live and broadcast-derived streams do not use a fixed playlist. They carry SCTE-35 cue messages embedded in the transport stream, which mark the point at which a break may be taken. Video-on-demand and client-driven workflows instead rely on manifest metadata, VMAP, content management system metadata or player-side timelines, according to guidance published by IAB Australia in August 2026. Where SCTE-35 is used, preserving the cues intact through encoding and packaging is a hard prerequisite; a break that is stripped during transcoding cannot be sold.

Insertion happens in one of two places. Client-side ad insertion leaves the player to fetch and render the advertisement, which is simple but visible to blockers. Server-side ad insertion, abbreviated to SSAI, stitches the advertisement into the content manifest before it reaches the device, producing a single continuous stream. Server-guided insertion, or SGAI, splits the difference: the server controls break signalling while the client retains rendering.

How the break is bought

In OpenRTB, the real-time bidding protocol maintained by IAB Tech Lab, the position of a break lives in one field on the Video object. The startdelay integer is set to 0 for pre-roll, to a positive number of seconds for a mid-roll at a known offset, to -1 for a generic mid-roll where the offset is unknown or irrelevant, and to -2 for a generic post-roll. The field has been present since the earliest published versions of the specification and remains the primary signal a demand-side platform reads when deciding what a break is worth.

OpenRTB 2.6, published in spring 2022, added the vocabulary for pods that connected television needed. Sellers can now pass podid to group impression opportunities belonging to the same break, podseq to indicate whether that break is the first, last or middle one in a programme, slotinpod for position within the break, poddur for total fillable seconds, maxseq for the ceiling on creatives, rqddurs for exact permitted durations where dead air is unacceptable, and mincpmpersec for floors that scale linearly with creative length. Before those fields existed, buyers had no standard way to distinguish the first slot of an early break from the fourth slot of a late one, and priced them identically.

On the sell side the break is operated by a publisher ad server and, in streaming, an SSAI vendor. On the buy side it is a demand-side platform submitting bids into slots it may not control; IAB Tech Lab implementation guidance is explicit that final pod construction belongs to the seller, and that a buyer submitting several bids for one pod may see none of them used.

From commercial breaks to cue points

The format is inherited from broadcast, where the interruption is the norm rather than the exception. Online video started from the opposite position, because early players were built around a single advertisement before a short clip.

Facebook confirmed mid-roll advertisements in February 2017, initially requiring videos of at least 90 seconds with the first break no earlier than 20 seconds. In December 2017 it raised the bar to three minutes with the first break at 60 seconds, citing consumer research showing an 18% increase in satisfaction when the first break was delayed, according to the company.

YouTube ran mid-rolls only on videos longer than 10 minutes, a threshold that visibly shaped upload behaviour toward clips of 10 minutes and one second. In July 2020 it lowered the minimum to eight minutes and switched mid-rolls on by default for eligible videos, with an opt-out deadline of 29 July. On 12 May 2025 it changed the placement logic itself, shifting slots toward natural pauses and transitions and away from mid-sentence cuts, and adding automatic slots to videos uploaded before 24 February 2025 that carried only manual breaks. YouTube said an experiment in July 2024 found channels combining automatic and manual slots earned over 5% more advertising revenue than channels using manual slots alone, and warned that interruptive manual placements might earn less after the change. Creators can still nominate preferred break locations, though YouTube's systems determine final placement.

The audio case

Podcasting is where mid-roll dominance is clearest. IAB and PwC data put mid-roll at 74% of US podcast advertising revenue in 2019 and 76% in 2020, settling at 64% in 2021 and 63% in 2022, with pre-roll taking most of the remainder and post-roll never exceeding 6%. Dynamic ad insertion, which replaces a spot at download time rather than baking it into the file, accounted for more than 90% of podcast advertising revenue by the 2023 study, up from 84% in 2021.

Scale follows. AdsWizz processes 16 billion dynamically inserted audio impressions a month, a figure that surfaced when Apple introduced HTTP Live Streaming video podcasts with dynamic insertion in February 2026, extending the same break structure to video episodes while preserving host-read delivery. Host-read spots, most of them mid-roll, represented 46% of quarterly US podcast advertising spending in the third quarter of 2025, up from 43.8% a year earlier. Published rate cards vary widely, but mid-roll host-read inventory is commonly quoted between $25 and $40 CPM against $15 to $30 for pre-roll and $10 to $20 for post-roll, according to podcast agency benchmarks circulating in 2026.

Why the position matters commercially

Duration is the constraint that decides whether the format is available at all. Full-length episodic content supports longer pods and mid-roll placement, while short-form segments generate higher volumes of shorter sessions with different frequency and completion dynamics. Traditional mid-roll formats proved incompatible with videos of 60 seconds or less, which is why short-form monetisation was rebuilt around revenue pools rather than break inventory. The same logic runs the other way in micro-drama, where each episode boundary becomes a break that can be monetised by an advertisement or a micro-payment.

Limitations and disputes

The format's premium rests on an assumption that is contested. Sounds Profitable's 2024 Ad Nauseam study found 46% of listeners said they always or often skip podcast advertisements, while noting that respondents overstate skipping when asked hypothetically; 28% said they listen to all of them. Bumper, analysing playback retention histograms across hundreds of breaks, reported in February 2026 that roughly 90% of the audience on most shows listens through, and that avoidance rises with break length. Both positions are drawn from real data measuring different things: stated intent against observed drop-off.

Measurement is the second dispute. IAB Australia's August 2026 guidance made the point bluntly, noting that insertion standards govern how advertisements are signalled and stitched, not how impressions are counted. SSAI removes the client-side beacon that verification vendors historically relied on, which is why the gap persists.

Ad load is the third. Streaming platforms are now varying it per viewer rather than per programme. Netflix disclosed at its 2026 upfront that it is testing personalised ad loads and frequency caps that adjust dynamically based on viewing behaviour, reported in coverage of geographic targeting expansion in July 2026. Personalisation of load makes the historic assumption that a break is a fixed inventory unit harder to sustain.

Disambiguation

Mid-roll and ad pod. A mid-roll is a position in the timeline. A pod is the set of advertisements inside a break. A mid-roll break may contain one advertisement or six; a pre-roll pod is a pod that is not a mid-roll.

Mid-roll and dynamic ad insertion. DAI is the mechanism that decides what fills a break at request time. Mid-rolls can be baked into a file permanently, and pre-rolls can be dynamically inserted. The terms describe different axes.

Mid-roll and pause or overlay formats. Pause, screensaver, overlay, squeezeback and in-scene placements pair the OpenRTB Video object with NonLinear VAST delivery and do not stop the content. They sit outside the pre-roll and mid-roll break structure entirely.

Mid-roll and mid-stream. In live broadcasting, mid-stream refers to any change made during a broadcast, including switching visibility settings, not specifically to advertising.

Recent developments

Pod construction is being automated. Magnite deployed machine learning inside SpringServe on 24 October 2025 to optimise pod construction for connected television publishers, one of several video ad servers competing on podded ad management and server-side auction capability.

Break inventory is also widening beyond the linear interruption. Netflix said at its 2026 upfront that it is extending dynamic insertion to pause advertisements and live content, buyable through partner demand-side platforms from summer 2026 in the United States and Canada. In podcasting, YouTube's two publishing paths continue to diverge on this point: native uploads unlock full mid-roll tooling, while RSS-generated episodes qualify for automated placement based on audio duration alone.

Timeline

  • 2012, 19 July: IAB publishes VMAP 1.0, defining AdBreak, timeOffset and breakType. It remains the only version.
  • 2012: VAST 3.0 introduces ad pods, allowing sequential linear advertisements in a single break.
  • 2016, November: OpenRTB 2.5 published, carrying startdelay with generic values for mid-roll and post-roll.
  • 2017, February: Facebook confirms mid-roll advertisements, requiring 90-second videos with breaks no earlier than 20 seconds.
  • 2017, December: Facebook raises the threshold to three-minute videos with the first break at 60 seconds.
  • 2020, July: YouTube lowers mid-roll eligibility from 10 minutes to eight and enables mid-rolls by default, with a 29 July opt-out deadline.
  • 2022, spring: OpenRTB 2.6 adds podid, podseq, slotinpod, poddur, maxseq, rqddurs and mincpmpersec.
  • 2024, July: YouTube runs the experiment later cited as showing over 5% higher advertising revenue for channels combining automatic and manual slots.
  • 2024, July: IAB Tech Lab releases the VAST CTV Addendum, standardising registered ad IDs and high-resolution media across VAST 2, 3 and 4.
  • 2025, 12 May: YouTube shifts mid-roll placement toward natural break points and adds automatic slots to older videos.
  • 2025, 24 October: Magnite deploys machine learning in SpringServe for connected television pod construction.
  • 2026, February: Apple introduces HLS video podcasts with dynamic ad insertion, including host-read spots.
  • 2026, August: IAB Australia publishes server-guided ad insertion guidance, noting insertion standards do not govern impression counting.

Summary

Who: Publishers and creators schedule mid-roll breaks; ad servers and server-side insertion vendors fill them; demand-side platforms bid into them. IAB Tech Lab maintains VMAP, VAST and OpenRTB, the specifications that describe them.

What: An advertising break positioned inside content, after playback starts and before it ends, signalled by a time offset or a cue message and filled by one or more advertisements arranged as a pod.

When: Standardised for online video by VMAP in July 2012 and by VAST 3.0 pods the same year, extended for connected television podding by OpenRTB 2.6 in 2022, and reworked at platform level by YouTube's placement change on 12 May 2025.

Where: Long-form online video, connected television and streaming, podcasting and audio, and social video where duration thresholds permit. Short-form vertical video does not carry the format.

Why: The break reaches an audience that has already committed to the content and has a reason to return after the interruption, which is why mid-roll accounted for roughly two-thirds or more of US podcast advertising revenue across every year measured by IAB and PwC, and why placement quality now carries measurable revenue consequences on the largest video platform.