The Media Rating Council said on November 10, 2026 that the first phase of its draft standards governing artificial intelligence in media measurement is planned for public comment in the first quarter of 2027, with a second phase to follow by mid-2027. The update, posted on the organization's LinkedIn page, points to a 13-page guidance document dated July 8, 2026, which maps seven existing MRC standards and guidelines onto nine AI principles that are already applied in MRC accreditation audits.

In Short

The Media Rating Council, the US body that accredits advertising measurement, plans to put its first draft rules on how AI is used in measuring ads out for public comment in early 2027, with a second batch by mid-2027. Until then, it says seven of its existing rulebooks already cover AI, and its auditors check accredited measurement companies against them today. Anyone relying on AI-driven numbers for fraud filtering, brand safety, audiences or auction pricing can now see which checks already apply and which risks have no standard yet.

Two documents, four months apart

The LinkedIn post, published on November 10 under the heading "MRC Standards On AI Use In Media Measurement," is brief. According to the MRC, the organization "is in the process of developing Standards relating to the rapidly evolving use of Artificial Intelligence and Machine Learning models in areas that intersect with media measurement processes and products." The new rules, the post says, "will be supplementary to existing Standards and Guidelines," which it describes as already covering data integration, empirical support, identity resolution, invalid traffic, brand safety and suitability, and auction transparency.

The scheduling sentence is the new information. "The MRC anticipates that draft Standards governing AI use in media measurement will be issued in Phases with the first Phase planned for public comment in Q1 of 2027 and the second Phase by mid 2027," according to the post.

That timetable sits later than the one in the guidance document the post links to. The document, titled "MRC Guidance on Existing Standards for AI Use in Media Measurement," carries the date July 8, 2026 on every page. In it, the MRC wrote that its standards effort "began in Q1 2026 and is expected to be completed in phases throughout 2026 and early 2027." The same document called the work "a top priority for 2026." The November 10 post places the first public draft after the end of 2026 and pushes the second phase to the middle of 2027. The post does not explain the change.

The post also lists the use cases the new standards will address: general model development and governance, invalid traffic and validity, brand safety and contextual considerations, "base measurement of structured and conversational ads in LLM," simulated and synthetic audiences, and auction transparency. The reference to ads inside large language model interfaces does not appear in the July document, which uses the broader label "Base Digital Measurement" for that focus area.

How the MRC frames the problem

The July document opens with definitions. According to the MRC, artificial intelligence "refers to the use of technologies to build machines and computers that have the ability to perform cognitive functions associated with human intelligence," while machine learning is a subset that "automatically enables a machine or system to learn and improve from experience." References to AI throughout the document apply to machine learning models unless stated otherwise.

The council separates what it calls Classical AI/ML from newer uses. Classical models use patterns in historical data to detect anomalies, forecast outcomes or classify content, typically performing one defined task. According to the MRC, it has "observed and evaluated, through audit, application of Classical AI/ML across a variety of measurement systems using existing Guidelines and Standards for several years."

The newer categories are where incremental guidance is needed. The document lists generative AI, which produces text, images, music and voices; agentic AI, which "acts autonomously and makes decisions" and takes actions "without human intervention"; assistive AI, the copilot tools designed to augment rather than replace people; and embedded AI, capabilities built into existing technology such as ERP systems or browsers. The MRC notes these categories may overlap.

Where does AI already touch measurement? The document names invalid traffic detection, agentic AI used both internally and in client-facing functions "such as auctions, campaign set up and reporting," and ad verification, brand safety and suitability processes. The MRC also observed the spread of AI among consumers and content creators. "The wide accessibility to AI-supported applications for a multitude of purposes has blurred the lines between organic and generative content as well as human user and agentic activity," according to the document.

That blurring is the measurement problem in one sentence. Audience counts, fraud filters and brand safety classifiers all depend on telling a person from a program and a human-made page from a generated one.

The council's stated position on disclosure is direct. According to the MRC, "adequate and transparent disclosures to measurement users regarding both the methodology and limitations of models used in the measurement process with impact on reported metrics is required and important."

Nine principles

The MRC developed the framework with the audit firms that conduct its accreditation reviews. According to the document, it was "developed in alignment with industry AI Standards frameworks" and contains nine principles "which represent areas in need of evaluation (and potential risks) when intersecting with measurement systems and the media being measured." The document does not name which external frameworks it aligned with.

The nine, with the MRC's descriptions:

  • Explainability: model and algorithm outputs "are explainable and allow for validation."
  • Compliance: models and algorithms comply with applicable MRC standards and guidelines.
  • Fairness: potential model bias "is assessed for group and individual fairness in representation and model outputs."
  • Transparency: the purpose and design of the model are disclosed to users "in sufficient detail to understand the outputs produced by the model."
  • Security: systems, data transfers and storage are protected against unauthorized access.
  • Data protection: the system's data use "aligns with permitted rights and confidentiality."
  • Reliability: models produce "valid and accurate outputs based on stated expectations for the model."
  • Accountability: inputs, designs, processes, thresholds, limitations and assumptions are known to the measurement service "and quantified, as applicable."
  • Adaptability: the model can adapt to technological change and to changes in media consumption, content delivery and consumer behavior.

According to the MRC, the effort involved its staff and "supporting and sponsoring organizations such as the ANA, 4As, IAB Tech Lab, media buyers and sellers, measurement services and other media measurement users." The document also states that while existing MRC rules "provide a good foundation for evaluating AI use in media measurement, that existing guidance is not currently consolidated or explicitly expressed as applicable to this area."

Seven rulebooks, 58 mappings

The core of the July document is a cross-reference. The MRC, working with its audit firms, built a list of the standards and guidelines currently used to evaluate AI in accredited services. Seven documents appear:

  • MRC Minimum Standards
  • Guidelines for the Conduct of Ad Verification
  • MRC Supplement to IAB Guidelines for the Conduct of Ad Verification: Enhanced Content Level Context and Brand Safety
  • Invalid Traffic Detection and Filtration Standards Addendum
  • MRC Guidelines Concerning Data Integration
  • MRC Outcomes and Data Quality Standards
  • MRC Digital Advertising Auction Transparency Standards

The status of this mapping matters more than its format. According to the MRC, it "currently applies these aspects in audit to AI use in measurement and expects audited measurement services to comply with relevant requirements where applicable." This is not a proposal. It describes what auditors check now.

Several requirements cut across all seven documents, according to the summary table. All of them require data quality control, model empirical support and associated disclosure. All require "clear disclosure to users of methods, data sources, procedures and results including bias and limitations." All require quality control, validation and monitoring of models and methods. And all require "clear disclosure of errors, limitations and guidance for applicable use."

A three-page appendix then breaks the mapping down section by section. By PPC Land's count, it contains 58 entries: 23 drawn from the Minimum Standards, 11 from the Outcomes and Data Quality Standards, seven from the auction transparency standards, six from the brand safety supplement, four each from the ad verification and data integration guidelines, and three from the invalid traffic addendum.

Minimum Standards

The Minimum Standards entries read like a disclosure checklist for any model that touches reported numbers. Section A.10 covers "pilot testing and validation of new AI models and disclosure before deployment." Section A.11 covers the "provenance of training data to prevent undue influence or bias." Section B.2 requires "disclosure and quantification of material methodology change impacts," and B.9 requires "disclosure and quantification of AI-driven adjustments and explanation of purpose."

Statistical candor is spelled out as well. Section B.8 maps to "disclosure of confidence intervals or error rates for model outputs" and "disclosure of model accuracy and error margins and calculations." Section B.12 covers "disclosure of anomalies or incidents that may have impacted model outputs." Section C.3 requires that users have the latest models and that "errors and fixes are promptly communicated."

Brand safety

The brand safety supplement carries the most detailed machine learning language in the appendix. Section 3.4 maps to "end-to-end ML lifecycle controls, inclusive of training data selection, parameter/feature choice, data preparation, model selection, training, evaluation, tuning and periodic refresh." The same section requires that training and evaluation datasets are distinct, that data is "cleaned/normalized/deduplicated," and that "material bias is reduced and disclosed."

Human review is not optional under this reading. According to the MRC's interpretation of sections 3.4 and 3.5, "Known model weaknesses and edge cases should direct manual review, and reviewer findings should feed back into model improvement." Sections 3.4, 3.7 and 9 require "non-technical disclosure of model methods, dataset sizes, data sources, and level of reliance on ML versus human review."

Auctions: feature importance and model cards

The auction entries treat the auction itself as an algorithm. According to the appendix, sections 1.1, 4, 5.2, 5.3 and 6.1 of the auction transparency standards call for governance of auction mechanisms "as an algorithmic decisioning process over allocation and pricing, with transparent disclosure of the logic, parameters, and rules."

Section 3.3 goes further than any other entry in the document. It maps to two specific outputs. The first is feature importance: "Provide a dynamically generated list of the top N features or variables that contributed the most to the ad's placement and price." The second is a model card: "Document the training data, intended uses, known limitations, and performance metrics of ML models."

Section 2.1.4 supports "independent explainability review of auction behavior through third-party measurement and access to relevant bid or protocol information where possible." It is the entry in the appendix that most clearly points to access for outside parties, rather than disclosure by the service itself.

Invalid traffic

The invalid traffic entries focus on governance rather than disclosure. Sections 3.1, 3.2 and 3.3 of the addendum map to creating "independent, technically qualified governance, R&D teams with authority to oversee AI." Sections 3.3 and 5 call for governance and continuous improvement processes for AI detection systems, "balancing security requirements with transparency and accountability." That balance reflects a tension built into fraud detection: explaining a filter in full can teach fraudsters how to evade it.

The summary table adds that the invalid traffic standards "require protection of techniques from reverse engineering" and a "risk and control approach to monitor evolving threats."

Data integration and outcomes

The data integration entries address modeled and fused data. According to the appendix, services should "monitor match/linkage rates and avoid overusing synthetic/imputed data," and "quantify and flag variability or unreliable outputs." The outcomes entries call for human review "to correct AI outputs and enhance fairness," the use of "explainable AI techniques to clarify prediction drivers," and data retention policies "that balance accountability and privacy."

Where existing rules run out

The July document is candid that the existing rules leave gaps. According to the MRC, there are "several areas where there are either no defined metrics or there is a need for specific guidance on minimum requirements and evaluation criteria."

It names six priority areas for the new standards: general, invalid traffic, brand safety, base digital measurement, identity and big data, and auctions. According to the document, these were chosen based on "the incidence of AI use in measurement or impact on behaviors being measured" and the need for requirements "not covered in existing Guidelines or Standards."

The emerging use cases are listed by medium. For television, the MRC cites further big data and fusion integration, "privacy responsive methodologies, absent availability of PII," device attestation for verifying supply sources, and content-level classification at scale. For audio, it names richer audience data in podcast measurement and cross-channel outcome measurement such as brand lift, attribution and creative ROI. For digital, it points to cross-media measurement and "agentic workflows developing to support ad buying, selling and measurement." For out-of-home, it lists approximations of attention, parity with digital techniques in forecasting and audience modeling, and mobility and location intelligence.

Agents, crawlers and grey areas

The invalid traffic section of the risk list is the longest. The MRC names the "implications of agentic AI advances (near term) and Artificial General Intelligence (far term)." It flags the "increased complexity of 'General IVT' activity as AI-based crawlers increase (declared)," the "increased presence of non-declared or spoofed user agent-based crawlers," and the "increased sophistication of SIVT mimicking human behavior patterns." It also lists the "proliferation of Made for Advertising (MFA) where applicable to IVT."

A separate heading covers what the MRC calls IVT "grey-areas." Three items appear. The first is "distinguishing 'human-involved' agentic-AI exposure in measurement," which the document labels "agentic validity." If a person instructs an agent to research a product and the agent loads a page with an ad on it, has a person been exposed to that ad? Existing invalid traffic rules were not written with that question in mind.

The second is the "measurement and monetization of zero-click," in which answers are delivered without a visit to the source. The third concerns "Answer Engine Optimization (AEO) proliferation and model training data scraping impacts on property level IVT classifications," which the document glosses as "publisher collateral damage." The concern, as the wording suggests, is that a site heavily crawled by AI systems could see its own traffic classified as invalid at the property level, penalizing the publisher for activity it did not invite.

Synthetic audiences and content labels

Two other risk areas address content and audiences. Under AI content labeling and rights management, the MRC lists the "current limitations associated with AI content labeling," the "ability to distinguish organic vs. generative content in ML training data," and industry frameworks such as C2PA. A brand safety classifier trained on a corpus that increasingly contains generated text and images learns from that corpus, whether or not anyone labeled it.

Under simulated and synthetic audiences, the document calls for an "evaluation process for simulated/synthetic audience accuracy and completeness," attention to the "influence of training data on model outputs and fidelity of data used in model predictions," and criteria for determining demographic characteristics and "minimum viewership metadata."

Auctions and black boxes

The final risk heading covers "transparency in AI buying systems (auctions and seller operated buying agents)." The MRC lists disclosure requirements for AI use in auction environments, including agentic AI where the auction transparency standard does not already capture it. It names "risks related to algorithmic collusion, algorithmic poisoning, and cold start issues" and, in two words, "blackbox placements."

The AI governance list also names a dependency risk: "dependency on AI and AI vendors to support code development and measurement." Measurement services that rely on third-party models to write their code or run their classifiers inherit those vendors' limitations.

Questions the MRC wants measurement users to ask

The July document closes with a set of questions for buyers, sellers and intermediaries that subscribe to measurement services. According to the MRC, the aim in the interim is to let users "ask the right questions and set appropriate expectations for quality measurement" while the new standards are written.

The questions cover where AI sits in a measurement, buying or reporting workflow, and how much of a result comes from direct observation versus model-based techniques. On reliability, the MRC lists: "What is the basis of 'truth' in the development of models?" Further questions address how training data is vetted, how models are validated in production (with human in the loop given as an example), and which KPIs are used for validation.

On change management, the questions ask how often methodologies are updated through retraining or new model releases, and what notice clients receive about the timing and effect of those updates on their analytics. On data, the MRC raises whether a client's information is used to train the vendor's models. And on validation, the list asks whether the process is examined and accredited by the Media Rating Council, and what observations third parties have made about the validity of the methodology.

Why this matters for the marketing industry

The MRC sets the bar that measurement vendors, platforms and verification companies must clear to call a metric accredited. Its documents typically move through drafts, public comment periods and final versions, often followed by a grace period. When the council updated its Invalid Traffic Detection and Filtration Standards in June 2020, it gave accredited measurers a one-year grace period to comply. Its draft standards for digital ad auction transparency went out for comment in September 2025, and its October 2025 policy barred property-level verification from being sold as brand safety, with a grace period that ended in April 2026. The MRC and IAB finalized attention measurement guidelines in November 2025 after a draft that May.

Against that record, the new timetable means no MRC standard written specifically for AI will exist in final form during 2026, and the second phase will not reach the public before mid-2027. Comment periods and any grace periods come after that. The MRC has not published a date for final standards. Until then, the July mapping is the operative text.

The pressure the document describes is already visible in the data. Cloudflare figures published in July 2026 showed AI crawl-to-referral ratios ranging from 118:1 to nearly 50,000:1, the kind of declared crawler volume the MRC now treats as a complication for general invalid traffic. Undeclared agents are a separate problem: DataDome documented a single Grok query triggering 16 requests from 12 IP addresses with spoofed user agents, which matches the MRC's concern about "non-declared or spoofed user agent-based crawlers."

On the fraud and quality side, IAS reported in July 2026 that invalid traffic fell to 1.134% and MFA to 0.199%, both still above its forecast. Earlier in 2026, DoubleVerify's Fraud Lab described AutoBait, a network of more than 200 MFA domains whose AI-generated articles cost under $2.25 each to produce. The MRC files MFA proliferation under invalid traffic "where applicable to IVT," a placement that matters to publishers whose inventory could be affected by property-level classifications.

The "conversational ads in LLM" line in the November 10 post points to ad formats that have drawn a crowd of measurement vendors this year. OpenAI named 22 companies working on ChatGPT Ads in October 2026, 20 of them in measurement roles, and IAS said it was among the first media quality partners in a closed ChatGPT Ads pilot with no method or pilot size disclosed. The post lists base measurement of these ads among the areas the new standards will address, which means the basic definitions, such as what counts as a served or viewed ad inside a chatbot answer, are still to be written.

Content labeling has advanced on the platform side. Google added "How this ad was made" AI disclosure panels to My Ad Center on July 9, 2026, ahead of EU AI Act transparency obligations that applied from August 2, 2026. The MRC document treats labeling frameworks as a measurement input: if classifiers cannot reliably tell generated content from organic content, brand safety and contextual metrics built on them carry an error the user cannot see.

Other standards bodies are working on adjacent pieces. The IAB Tech Lab published final programmatic auction definitions in June 2026, with 15 defined terms and a 12-step OpenRTB workflow, which gives the MRC's auction transparency work a shared vocabulary. The IAB released a draft cross-channel measurement decision guide on October 20, 2026, with comments closing November 4, covering how to weigh marketing mix models, incrementality tests and attribution. And on search, IAB Australia urged a shift from traffic volume to citation-based metrics in December 2025 as instant answers reduce clicks, the same zero-click problem the MRC now lists as an invalid traffic grey area.

Accreditation itself already reaches into AI-adjacent processes. HUMAN Security's viewability measurement earned MRC accreditation in April 2026, with invalid traffic filtering as part of the scope. Under the July guidance, AI models used in accredited processes of that kind are evaluated against the disclosure, validation and governance requirements the appendix lists, where applicable.

What the documents leave out

Several things are absent from both the July document and the November 10 post.

Neither names the members of the MRC AI Standards Working Group beyond the sponsoring organizations, or the audit firms that helped build the framework. The July document refers to alignment with "industry AI Standards frameworks" without naming them.

The appendix sets no numeric thresholds. It requires disclosure of confidence intervals, error rates and data sufficiency thresholds, but it does not say what level of model accuracy or what minimum sample size would be acceptable. Those numbers, if they come, will presumably sit in the new standards.

The document does not say how many accredited services use AI in their measurement, which ones, or what auditors have found so far. It describes the MRC's observations in general terms only.

Finally, the post does not explain why the timeline moved from completion "throughout 2026 and early 2027" to a first public draft in the first quarter of 2027 and a second phase by mid-2027, nor does it say when final standards might take effect. The MRC's July document invites organizations that want to take part to contact MRC staff.

Timeline

Summary

Who: The Media Rating Council, the US body that sets media measurement standards and accredits services against them, working with its accreditation audit firms and sponsoring organizations including the ANA, 4As and IAB Tech Lab. The rules affect accredited measurement services and the buyers, sellers and intermediaries that rely on their metrics.

What: In a LinkedIn post, the MRC said draft standards governing AI use in media measurement will be issued in phases, the first for public comment in Q1 2027 and the second by mid-2027. The post links to a July 8, 2026 guidance document that maps seven existing MRC standards and guidelines onto nine AI principles in 58 appendix entries, lists the gaps the new standards will cover, and sets out questions measurement users can put to vendors.

When: The post was published on November 10, 2026. The guidance document is dated July 8, 2026. The effort began in Q1 2026, and the July document had expected completion "throughout 2026 and early 2027."

Where: The guidance applies to MRC-accredited measurement across TV, audio, digital and out-of-home, in the United States and wherever MRC accreditation is used as a benchmark.

Why: According to the MRC, AI has spread into invalid traffic detection, brand safety classification, auctions, audience modeling and reporting, while AI use by consumers and creators has blurred the line between human and agent activity and between organic and generated content. Existing rules cover much of this but were never consolidated for AI, and areas such as agentic traffic, zero-click consumption, synthetic audiences, AI content labeling and ads inside LLM interfaces have no defined requirements yet.