Match rate is the percentage of records that one advertising system can link to identities held by another. An advertiser uploads 100,000 customer email addresses, the platform recognises 60,000 of them as its own account holders, and the result is 60%. The same arithmetic applies when a demand-side platform (DSP) tries to recognise the browsers a supply-side platform (SSP) sends it, or when a brand and a publisher compare customer files inside a clean room. The figure exists because identity is fragmented: every company assigns its own identifiers, and data becomes usable for targeting, suppression or measurement only once those identifiers have been joined.

How the percentage is produced

The most common use today is the customer list upload. Google's Customer Match, Meta's Custom Audiences and their equivalents follow one sequence. Contact details are exported from a customer relationship management (CRM) system, normalised, hashed and uploaded. The platform hashes its own account data in the same way and compares the two sets of codes.

Normalisation decides most outcomes. Google requires whitespace to be trimmed, text lowercased, phone numbers written in the E.164 international format and periods removed from the local part of gmail.com and googlemail.com addresses before hashing with SHA-256. One stray capital letter produces a different hash, and the record fails. PPC Land's clean room explainer names inconsistent normalisation as the most common cause of a collapsed match rate.

After processing, which can take 48 hours, Google reports the rate imprecisely by design: results are grouped into buckets, shown only when at least 100 rows match unique users, and not calculated at all for lists of mobile device identifiers. Its help centre describes the number as a check on formatting and states that it does not indicate list performance.

Match keys are the main lever. Google accepts email, phone, and first and last name with country and postal code, and now IP address with an interaction timestamp. Several email or phone columns per customer are allowed, because people accumulate addresses that a CRM stores in parallel. OpenAI followed the same reasoning when it added plural list fields for hashed emails and phone numbers to its conversion tools.

In real-time bidding the term describes cookie synchronisation. A DSP cannot read an exchange's cookie, so the two fire a pixel, learn each other's identifier for the same browser and store the pair in a match table. Google's Authorized Buyers documentation lets a hosted match table entry be treated as expired after 14 days and throttles pixel matching for bidders whose response rate falls below 90%. The share of bid requests arriving with an identifier the bidder recognises is its match rate on that exchange. OpenRTB, the IAB Tech Lab bidding protocol, carries the synced value in the user.buyeruid field, and its implementation guidance says that value is expected to come from a real-time cookie sync unless buyer and seller have agreed otherwise. A bidder facing an unknown user can decline with no-bid reason code 8.

Audience management platforms use a third formula. Adobe Audience Manager, as PPC Land's addressable reach explainer sets out, divides the addressable audience by the total segment population: a one-million-device segment with 700,000 devices synced to a destination reports 70%. Safari traffic drops out entirely, since Intelligent Tracking Prevention blocks the syncs.

Advertisers and agencies own lists and data hygiene. Onboarding vendors such as LiveRamp, Experian and TransUnion translate offline records into online identifiers. SSPs run sync pixels, publishers pass authenticated identifiers such as Unified ID 2.0, and walled gardens perform the join internally and disclose only the outcome.

Origin and evolution

Facebook introduced Custom Audiences in 2012, according to Time, allowing advertisers to reach users by email address. Google followed on September 28, 2015 with Customer Match across Search, YouTube TrueView and Gmail. At launch it accepted email addresses only, and lists needed at least 1,000 of them, according to Gulf News. Phone numbers and mailing addresses came later, and Google has since cut audience thresholds to 100 users across its networks.

How much of the number platforms reveal has shifted repeatedly. Facebook suspended reach estimates for Custom Audiences in March 2018, after Northeastern University researchers showed the tool could be used to infer attributes of individuals on an uploaded list, according to AdExchanger. The estimates returned on July 2, 2019, with more complex rounding. Google went the other way in 2021, displaying Customer Match rates instantly alongside upload history reaching back to April 1, 2020.

In programmatic trading the change was definitional. During 2024 buyers found that some sellers were filling buyeruid with identifiers inferred by device graphs rather than produced by syncs, a practice known as ID bridging. An IAB Tech Lab workstream of more than 80 participants from 40 companies revised OpenRTB 2.6, publishing a three-page definition of cookie syncing and adding inserter, matcher and match method attributes to the extended identifiers object. Where an identifier was produced by probabilistic or graph matching, the guidance requires that match method value to be passed downstream, even if a later hop uses a genuine cookie exchange. PPC Land's unlinkability explainer details the provenance fields.

Ingestion routes changed in 2026. Google stopped accepting Customer Match uploads through the Google Ads API on April 1, forcing migration to the Data Manager API. On May 28 it added IP addresses as a Customer Match input, promising higher rates from the third quarter. IP matching is unavailable for users in the European Economic Area, the UK and Switzerland.

Why it matters

A match rate caps everything built on a first-party file. At 40%, a 50,000-customer list leaves 20,000 people eligible for targeting, exclusion or lookalike modelling before any other filter is applied. Suppression lists suffer equally, so existing customers keep seeing acquisition ads.

The metric also carries the commercial argument over where data should be applied. Equativ's analysis of October 24, 2024 put match rate losses between platforms at 40% to 70%, with each third-party cookie sync degrading the signal. That range has become the standard case for sell-side curation. It was invoked when Mediavine brought audience data from 18,000 publishers into Index Exchange in April 2026, and PPC Land's supply path explainer uses it to show why shorter routes preserve addressability. The number originates with a company selling curation.

Server-side auctions show the same cost. A server cannot read bidder cookies, so Prebid Server keeps its own synced identifiers, treating them as stale after seven days, and Index Exchange documentation says its browser adapter matches better than its server route.

Measurement depends on the join too. Enhanced conversions, conversion APIs and clean-room attribution all link hashed identifiers, and PPC Land's enhanced conversions explainer notes that Google's rates rely on customers being signed in.

Limitations and disputes

The central weakness is that a match rate measures recognition, not correctness. Research commissioned by the Coalition for Innovative Media Measurement (CIMM) and Go Addressable, conducted by Truthset and published on November 5, 2025, found IP-to-postal linkages accurate 13% of the time and IP-to-email 16%, with six providers agreeing on the same household in 6.4% of cases. The study observed that data providers often favour match rate and scale over recency. FreeWheel concluded that IP-based targeting can miss 87% of households. A high rate can rest on wrong links.

Self-reported figures add a second layer. Viant has claimed a 95% Household ID match rate across US households through TransUnion, and Yahoo counts 232 million logged-in ConnectID users; neither is independently audited. Pinterest reported 5% to 15% higher match rates after integrating Epsilon's clean room in November 2025, again a vendor figure. IAB Spain's April 2026 guide urged buyers to audit vendor match-rate accuracy before signing.

Denominators differ. Google counts every row matching an account, including people who have opted out of personalised ads or been inactive for 30 days, and its developer documentation says reachable coverage can range from 1% to 99% of a list. Cross-platform comparisons mislead for the same reason, and Google lists them among common causes of apparent problems.

Privacy is the final dispute. The 2018 Facebook episode demonstrated that precise match feedback can expose individuals, which is why platforms bucket, round or withhold the figure.

Not the same as

List size is Google's estimate of matched users actually reachable on a given property. It is usually smaller than the match rate implies.

Match accuracy is whether a link is correct, the property Truthset grades. Accuracy can be low while the rate is high.

Event Match Quality is Meta's score from 0 to 10 for how reliably server events carry identifiers that match accounts, a per-event grade rather than a list percentage.

Log reconciliation uses the same phrase differently. The ISBA and PwC study of May 6, 2020 matched 31 million of 267 million impressions between DSP and SSP records, and a January 2023 follow-up reported 58%. Those figures describe auditability, not audiences.

Recent developments

Identity inputs keep widening. Roku became the first publisher in Google's Confidential Publisher Match in March 2026, pitched as a way to raise match rates for Display & Video 360 buyers on its inventory. OpenAI made automatic advanced matching, which sends hashed form data from websites, the default on existing ChatGPT pixels from August 17, 2026.

Accuracy is gaining its own market. Truthset connected its graded audiences to Unified ID 2.0 in The Trade Desk's marketplace on May 7, 2026, and FreeWheel adopted the grades for connected television matching. Google's answer, introduced on September 10, 2026, is a Data Strength Uplift Metric counting conversions recovered by first-party setups, a platform-computed figure that, like the match rate itself, cannot be checked from outside.

Timeline

  • 2012: Facebook launches Custom Audiences, matching advertiser email lists to user accounts
  • September 28, 2015: Google announces Customer Match for Search, YouTube and Gmail, email only
  • March 2018: Facebook suspends Custom Audience reach estimates after a privacy vulnerability is reported
  • July 2, 2019: Facebook reinstates reach estimates with revised rounding
  • May 6, 2020: ISBA and PwC publish their supply chain study matching 31 million of 267 million impressions
  • 2021: Google makes Customer Match rates visible instantly, with upload history from April 1, 2020
  • January 2023: ISBA and PwC follow-up reports a 58% match rate
  • October 24, 2024: Equativ publishes analysis putting cross-platform match rate losses at 40% to 70%
  • December 2024: IAB Tech Lab publishes OpenRTB 2.6 identifier provenance updates
  • November 5, 2025: CIMM and Go Addressable publish Truthset research on 13% IP-to-postal accuracy
  • November 2025: Pinterest integrates Epsilon's clean room
  • December 2025: Google reduces audience list thresholds to 100 users
  • February 28, 2026: FreeWheel warns IP-based targeting can miss 87% of households
  • March 2026: Roku joins Google's Confidential Publisher Match
  • April 1, 2026: Customer Match uploads through the Google Ads API stop working
  • April 2026: IAB Spain SSP guide urges audits of match-rate accuracy
  • May 7, 2026: Truthset connects graded audiences to Unified ID 2.0
  • May 28, 2026: Google adds IP addresses to Customer Match, excluding the EEA, UK and Switzerland
  • August 17, 2026: Automatic advanced matching becomes the default on existing ChatGPT pixels
  • September 10, 2026: Google introduces the Data Strength Uplift Metric

Summary

Who. Advertisers and agencies upload and clean the data; platforms such as Google, Meta and OpenAI compute the rate; onboarding vendors, DSPs, SSPs and clean room operators run the joins; the IAB Tech Lab defines how synced identifiers are declared in bid requests.

What. Match rate is the percentage of records one system can link to identities held by another, reported for customer list uploads, cookie syncs between exchanges and bidders, audience destinations and clean room overlaps. It measures recognition, not accuracy.

When. Customer list matching dates to Facebook Custom Audiences in 2012 and Google Customer Match on September 28, 2015. OpenRTB provenance rules followed in 2024, and Google rebuilt its ingestion route and added IP matching in 2026.

Where. Inside walled garden audience tools, in the match tables exchanges and bidders maintain, in audience management platforms, and in clean rooms where brands and media owners compare files.

Why. Fragmented identifiers make a join necessary before first-party data can be used. The resulting percentage caps addressable audiences, anchors the argument for sell-side curation, and is easily overstated when rates are self-reported or links are wrong.