Made-for-advertising detection is a post-buy problem today. A classifier reviews placements after the campaign has run, flags the MFA inventory, and the buyer adjusts future campaigns to avoid it. That workflow has a fundamental assumption: that there is a moment after the buy and before the next buy where a human or a tool can intervene. In agentic advertising, where deals are negotiated before inventory is bought, the post-buy audit runs after an agreement that was already made autonomously. Catching MFA post-buy in an agentic workflow does not prevent the buy. It audits a fait accompli.
TL;DR. In programmatic advertising, MFA detection happens post-buy: classifiers review placements after the campaign runs and feed findings back into future decisions. In agentic advertising, the deal is negotiated before the buy, which means quality evaluation has to happen at the point of negotiation if it is going to prevent the problem rather than record it afterwards. The information asymmetry this creates is structural: a seller's agent knows its inventory quality, while a buyer's agent relies on signals the marketplace provides. Without explicit quality standards encoded as machine-readable deal terms the buyer's agent can evaluate before agreeing, there is no mechanism for agentic buying to prevent MFA at the point of decision. Transparency in agentic advertising means ensuring the buyer's agent had the right information when it made the deal, not just that someone could audit what it did afterwards.
What is MFA, and how is it currently handled?
Made-for-advertising sites are domains built primarily to attract programmatic spend rather than to serve a genuine audience. They carry content, often AI-generated or low-quality, that exists to satisfy the minimum threshold for running ads. The Association of National Advertisers' research on programmatic supply chain waste found that MFA inventory accounted for a significant share of programmatic budgets, capturing spend that buyers believed was going to legitimate publisher environments.
The IAB and ANA have both published MFA standards that define the category with increasing specificity: high ad-to-content ratios, inorganic traffic, absence of direct audience relationships. Brand safety tools apply those definitions post-buy, flagging domains that meet the MFA criteria after impressions have already run. Buyers use those flags to update block lists, exclude domains in future campaigns, and push platform partners to improve supply quality.
This workflow is imperfect but functional in a world where the buy happens through an auction and the block list can be updated between campaigns. The workflow's assumption is that post-buy detection feeds back into future decisions, tightening the excluded inventory set over time.
Why does the auction model allow post-buy correction, and why does the deal model not?
Programmatic auctions are discrete events. Each impression is bid on independently, and a block list applied to one campaign takes effect on the next impression. The feedback loop between detection and prevention is fast: catch MFA on Monday, exclude the domain by Tuesday.
An agent-negotiated deal is a forward commitment. The buyer's agent agrees to buy a volume of inventory from a seller's agent under agreed terms, and the deal is binding before the first impression runs. If that inventory turns out to be MFA, the deal is already made. Post-buy detection identifies the problem after the commitment is locked, and the buyer's remedies are limited to what the contract allows, which current standard contract terms were not written to address for agentic scenarios.
The deal model moves the accountability question to the point of negotiation. Did the buyer's agent have sufficient information about the inventory quality when it made the deal? If yes, and the agent agreed to MFA inventory anyway, the buyer has a configuration problem: the agent should have been given harder quality constraints. If no, the buyer has an information access problem: the marketplace or the seller's agent did not provide the signals needed for an informed decision.
What information does a buyer's agent actually have at the point of negotiation?
This is the information asymmetry at the heart of the problem. A seller's agent represents inventory the seller controls and characterises. A buyer's agent must evaluate that inventory against quality standards using signals the seller provides or the marketplace makes available.
In a standard programmatic auction, brand safety tools run on the full inventory pool and buyers apply pre-configured block lists. The burden of quality assessment falls on the buyer's side, using third-party data and the buyer's own historical exclusions. In an agent-negotiated deal, the same asymmetry applies: the seller's agent knows what the inventory is, and the buyer's agent has to infer quality from whatever signals are available at deal time.
For MFA specifically, the buyer's agent would need to evaluate the deal against defined MFA criteria before agreeing. That requires machine-readable MFA signals to be present in the deal terms or in the marketplace's inventory metadata. Currently, MFA classification happens post-buy by tools reading live impressions. There is no standard mechanism for a buyer's agent to query an inventory source's MFA status as a deal term before the agreement is made.
What would solve the information asymmetry?
Encoding inventory quality standards as explicit, machine-readable deal terms the buyer's agent can evaluate before agreeing. This means several things in practice.
The IAB's MFA definition, currently used as a post-buy classification framework, could function as a pre-deal quality standard: a buyer's agent sets an MFA compliance requirement as a deal term, and the seller's agent either confirms compliance or the deal does not complete. That moves detection from a post-buy classifier to a deal-level condition, where the buyer's agent has evaluated it before committing spend.
Shailley Singh of IAB Tech Lab, writing in August 2026, described the architecture requirements for agentic transactions as including "systems of record that minimise hallucination or misinterpretation of context and that both sides can reference and audit," as reported in an ADOTAT investigation published that month. The same requirement that applies to price terms applies to quality terms: a deal condition only holds if both sides have agreed to the same standard, in a record both can verify.
The harder problem is enforcement. A seller's agent declaring MFA compliance as a deal term may or may not be accurate, and a buyer's agent cannot audit that claim before the inventory runs. This is why the deal record and a shared audit trail are both necessary: not just that quality terms were agreed, but that the actual inventory delivered can be compared against those terms after the fact, with the comparison running against a record both sides signed.
What does this mean for how buyers should think about transparency?
Transparency in agentic advertising is not just about what the agent did. It is about what the agent knew when it made the decision.
A buyer whose agent bought MFA inventory has not necessarily been failed by a lack of post-buy transparency. The buy was fully visible in the logs. What the buyer was failed by was a lack of pre-buy information: the quality data the agent needed to make a different decision was not available at the point of negotiation.
Post-buy audit remains necessary, because no pre-deal quality evaluation will be perfect. But its role changes: in agentic buying, post-buy audit is an integrity check on whether what the seller claimed matched what was delivered, not the primary mechanism for quality control. The primary mechanism has to be the quality terms in the deal itself, agreed before the agent commits the spend.
________________
This article references the Association of National Advertisers' research on MFA and programmatic supply chain waste, the IAB's published MFA standards and definition framework, and statements by Shailley Singh of IAB Tech Lab published in ADOTAT in August 2026.