Frequency capping is a delivery rule that prevents an advertisement from reaching the same person more than a set number of times inside a set period. A buyer enters a count and a window, three impressions per day or ten per week, and whichever system holds the counter refuses further impressions to that identifier once the count is met. The rule exists because digital media is sold in impressions rather than audiences. Left alone, an auction keeps buying the cheapest impression available, and cheap impressions belong disproportionately to the heaviest users of the web, who absorb much of a campaign.

How a cap is built

Three components define a cap: a count, a time window, and the entity the counter attaches to. In Display & Video 360, Google's demand-side platform, caps sit at line item, insertion order, or campaign level, and the tightest applicable cap wins. An insertion order cap constrains every line item beneath it. Google's documentation gives the worked example: an insertion order capped at four impressions and a line item capped at three per day are both held to four.

The Display & Video 360 API expresses this as a FrequencyCap object with four fields - unlimited, timeUnit, timeUnitCount, and maxImpressions. Time units run from minutes through days, weeks and months to lifetime. Lifetime never meant forever: Google's documentation defines it as three months and states that lifetime caps are being deprecated, with a 30-day maximum after February 28, 2025. Exposure data is retained for 90 days.

Sell-side structure is similar, the vocabulary different. Google Ad Manager allows up to ten caps on one line item, so publishers can stack rules: one per hour plus three per day means both bind. Preferred Deal line items cannot be capped. Roadblocked line items count page views rather than impressions, so a roadblock carrying three creatives capped at six delivers eighteen counted impressions.

Enforcement happens before the auction, not after. A demand-side platform reads the identifier in a bid request, checks its exposure store, and declines to bid once the cap is met.

Two counters that never meet

The buy-side counter belongs to the platform doing the buying, the sell-side counter to the publisher's ad server. Neither sees the other, nor a third platform buying the same brand against the same person. That architecture, not any vendor's shortcoming, is why a viewer meets the same commercial twice in one break.

Vendors have built for the gap. Innovid launched Harmony Frequency on July 31, 2024 as a cross-platform frequency view. Amazon took the in-platform route, launching Frequency Groups in October 2023 to apply one cap across separate orders, then adding deduplicated cross-order reporting on December 23, 2024. Amazon internal data from September 2023, self-reported in that announcement, put the average unique reach gain at 6%.

Identity decides whether the cap works

A cap is only as good as the identifier it counts against, which makes it hostage to every shift in the identity layer. Connected television is the clearest case. Display & Video 360 caps across apps on one device when the Identifier for Advertising appears in the bid request, with publisher provided IDs, exchange provided IDs and session IDs accepted as alternatives, in line with IAB Tech Lab's OTT IFA guidelines from December 2018. Where none is available, capped line items exclude the impression outright, turning a delivery control into a targeting exclusion.

Ad Manager applies a comparable rule: time-based caps require a user identifier or a publisher provided ID, and without one the line item cannot serve. Pod- and stream-based video caps serve anyway but ignore the cap. Where only first-party cookies exist, capping works within one domain; in apps without a device identifier, within one app.

Where identifiers are missing, platforms model. Google states that some insertion orders and line items use machine learning and historical data to approximate caps when cookies or device IDs are unavailable, and warns that delivery fluctuates as a result. Household approaches lean on IP addresses, and recent research undercuts that footing. A Stanford-led paper reported that 5% of client addresses generate 55% of web requests and that 44% carry at least two distinct user agents. A Truthset study released on November 5, 2025 put IP-to-postal linkage accuracy at 13%.

The privacy-preserving alternatives did not survive. Google's Private Aggregation API supported frequency capping analysis without exposing individual behavior, and was among nine technologies Chrome retired on October 17, 2025. IAB Tech Lab's ID-Less Solutions guidance of July 2025 lists frequency and recency capping among the use cases needing rebuilding from aggregate signals alone.

Where the number came from

The cap is a digital control, but the number entered into it is inherited from television research. Herbert Krugman, then at General Electric, published "Why Three Exposures May Be Enough" in the Journal of Advertising Research in 1972, and Michael Naples assembled the surrounding evidence in Effective Frequency for the Association of National Advertisers in 1979. Three exposures per purchase cycle became the planning default.

Erwin Ephron's recency argument, published in the same journal in 1997, inverted the logic: since the moment of purchase readiness cannot be predicted, continuous presence matters more than accumulated exposures. Digital caps encode Ephron's position rather than Krugman's. A cap is a ceiling, not a target, and it spends the saved impressions on someone new.

What buyers can now measure

Frequency management was long asserted rather than quantified. Google Ads introduced Cross-Media Reach in June 2024 for deduplicated reach and frequency across video campaigns, and Display & Video 360 added 8+ and 9+ reach buckets on July 7, 2025, filling gaps between the 1+ through 7+ and 10+ tiers.

The bigger change landed on August 26, 2025, when Display & Video 360 replaced its frequency metrics. The retired measure counted programmatic guaranteed bid requests passed due to frequency. Its replacements express savings in money rather than requests, through Savings Reinvested from Frequency Cap, alongside a Cost per reach metric calculated as total media cost divided by impression reach.

Claims about the size of the prize come mainly from vendors. EDO, launching Ad EnGage Optimize on June 4, 2026, said its research found advertisers could recover up to 24% of budgets through frequency optimization alone. Audience data is more independent: TiVo's fourth-quarter 2025 research found excessive frequency to be the leading viewer annoyance, named by 35.2% of respondents ahead of repetitive or irrelevant creative.

Where caps break

Guaranteed inventory outranks the cap. In Display & Video 360, when a campaign holding both guaranteed and non-guaranteed line items hits its limit, the non-guaranteed one stops immediately while the guaranteed one stops only if the reservation can still be met. For YouTube programmatic guaranteed deals, campaign caps are described as best effort, and cost-per-hour Masthead buys ignore higher-level caps entirely.

Platforms concede that caps are approximate. When LinkedIn launched frequency capping on July 2, 2025 at 3 to 30 impressions per member account across seven days, its documentation warned that delivery may exceed the advertiser-set cap because delivery is not guaranteed. Ad Manager notes that a new browser session can trigger a cap one impression late, the first impression having been spent setting the cookie, and that its forecasting ignores monthly and lifetime caps.

Access is uneven. Reddit has restricted cap controls to advertisers with account managers, leaving self-serve buyers with API fields that do not function.

The dispute over the right number

Whether caps should be tighter is contested inside Google. Pablo Perez and Jesús Martín Calvo of the company's Marketing Research & Insights team published an analysis on February 3, 2026 arguing that the first impact delivers the most value and that reach should be maximized over repetition. Their evidence is YouTube Brand Lift data published in 2017 showing second impacts indexing at 1.5 to 1.8, below the 2x that would mark equal value. They also hold that S-shaped response curves in marketing mix models reflect weak measurement at low spend rather than real consumer behavior.

The counter-position is that frequency correlates with outcomes in aggregate models. The researchers accept the observation and reject the inference, treating higher frequency in high-return channels as a consequence of correct investment rather than a goal.

Terms that get confused with it

Target frequency is the inverse of a cap. Display & Video 360 announced monthly YouTube target frequency on December 2, 2024, optimizing delivery toward four to eight exposures rather than stopping at a ceiling.

Pacing governs how fast a budget is spent across a flight and says nothing about individuals.

Competitive separation keeps rival brands apart within a page or an ad pod. In Ad Manager it runs on exclusion labels rather than the frequency counter, though both are break-level controls that pod-level auctions were built to expose to buyers.

Suppression removes a person from targeting on a condition such as a completed purchase. A cap counts exposures; suppression reads a state.

Recent developments

Amazon has moved fastest on reporting. After dynamic reach and frequency reporting at unBoxed in November 2025 and cross-campaign deduplication that February, it launched cross-account reporting on April 2, 2026, combining Sponsored Ads and Amazon DSP data.

Google added reach overlap dimensions to Unique Reach Overlap reports on April 10, 2026, for refining frequency caps and reallocating budget. Bulk workflows loosened too: partial Structured Data File uploads, announced May 5, 2026, carry a frequency cap column group, so caps can be edited without the full 120-column line item file.

The structural problem remains European connected television. IAB Europe's programmatic guide, covered in April 2026, notes that HbbTV environments often lack persistent identifiers, leaving contextual and household approaches where individual capping cannot run.

Timeline

  • 1972: Herbert Krugman publishes "Why Three Exposures May Be Enough" in the Journal of Advertising Research
  • 1979: Michael Naples publishes Effective Frequency for the Association of National Advertisers
  • 1997: Erwin Ephron publishes his recency planning argument, prioritizing reach over accumulated exposures
  • December 2018: IAB Tech Lab publishes OTT IFA guidelines, defining identifier types used for connected television capping
  • October 2023: Amazon launches Frequency Groups, applying a single cap across separate orders
  • July 31, 2024: Innovid launches Harmony Frequency for cross-platform frequency management
  • December 2, 2024: Display & Video 360 announces monthly target frequency for YouTube, optimizing toward four to eight exposures
  • December 23, 2024: Amazon adds deduplicated cross-order reach and frequency reporting
  • February 28, 2025: Cut-off after which Display & Video 360 lifetime caps give way to a 30-day maximum
  • July 2, 2025: LinkedIn begins rolling out frequency capping at 3 to 30 impressions per member account over seven days
  • July 7, 2025: Display & Video 360 adds 8+ and 9+ impression reach buckets to frequency reporting
  • August 26, 2025: Display & Video 360 replaces bid-request frequency metrics with cost-based savings metrics and Cost per reach
  • October 17, 2025: Google retires nine Privacy Sandbox APIs, including Private Aggregation, which supported privacy-preserving frequency analysis
  • November 5, 2025: Truthset study for CIMM and Go Addressable puts IP-to-postal linkage accuracy at 13%
  • February 3, 2026: Google researchers publish an analysis arguing first impacts deliver the most value
  • April 2, 2026: Amazon launches cross-account reach and frequency reporting
  • April 10, 2026: Google announces reach overlap dimensions for Display & Video 360 and Campaign Manager 360
  • May 5, 2026: Display & Video 360 introduces partial Structured Data File uploads with a frequency cap column group
  • June 4, 2026: EDO launches Ad EnGage Optimize, automating frequency capping decisions on live television campaigns

Summary

Who: Buyers configure caps inside demand-side platforms and social advertising interfaces, publishers configure them inside ad servers, and identity providers, exchanges, and measurement vendors determine whether those counters can recognize the same person twice.

What: A delivery rule limiting the number of impressions one identifier can receive from a campaign, insertion order, line item, or creative within a defined window, enforced by refusing bids or ad selections once the count is reached.

When: The underlying frequency literature dates to 1972 and 1979 and digital enforcement arrived with ad servers in the 1990s, but the live change is measurement, with cost-based savings metrics arriving in August 2025 and deduplicated cross-account reporting in April 2026.

Where: Across display, video, audio, mobile app, and connected television inventory, with enforcement quality varying by identifier availability and weakest in HbbTV and other environments without persistent IDs.

Why: Repetition beyond a small number of exposures adds little measured value while consuming budget that could reach new people, and viewers name excessive frequency as their leading complaint about ad-supported streaming. The open dispute is not whether caps are needed but where the ceiling belongs, with Google's own researchers arguing that the first impact is worth most and that reach should be favored over repetition.