The advertising industry is getting remarkably close to a moment it has been predicting for years: software will not simply help media buyers make decisions but will increasingly make some of those decisions itself.
Key takeaways
- AI buyer agents can already turn a brief into a live campaign: Magnite and Amnet report about 70% less setup time in their first EMEA agentic campaign.
- Speed doesn't fix supply. Pixalate classified 41% of the global open-programmatic mobile-app traffic it measured in Q2 2026 as invalid.
- An agent can only compare what is described. Without data on origin, resale, compliance and acquisition method, it buys questionable traffic faster.
- The goal is transparency, not a "verified" stamp: buyers, and their agents, need to see which claims come from the seller, which are verified and which come from past performance.
AI is moving from optimizing to buying
That transition is no longer confined to product demonstrations. AI is already deeply embedded in marketing organizations, with IAB Europe reporting that 86% of surveyed companies are using AI in marketing in some form. At the same time, a new generation of agentic advertising systems is beginning to move beyond campaign optimization into activities that look much more like media buying itself. Buyer agents can interpret campaign objectives, evaluate opportunities and communicate with seller-side systems, while emerging industry protocols are being developed specifically to allow those agents to transact with each other.
One of the more revealing examples came from Magnite and dentsu's Amnet, which reported completing an agentic advertising campaign in EMEA in which AI was used to streamline the process of setting up and executing the buy. According to the companies, campaign setup time was reduced by roughly 70%. That is the kind of efficiency gain that makes the direction of travel fairly easy to understand. If software can turn a media brief into an executable campaign in a fraction of the time required by conventional workflows, advertisers and agencies have a strong economic incentive to use it.
The next stage is already taking shape. IAB Tech Lab has been developing standards, such as AAMP 3.0 with OpenProposal, that allow buyer-side and seller-side AI agents to communicate around media opportunities, including proposals and campaign requirements. Amazon is introducing conversational agentic workflows for advertising, while Google and other large platforms continue to move more targeting, bidding, creative and placement decisions into automated systems. The practical result is that a media buyer increasingly has the option to describe the outcome required rather than manually configure every decision needed to reach it.
That could remove an enormous amount of friction from media buying. An advertiser looking for finance traffic, for example, may eventually be able to tell an agent that it wants high-intent consumers interested in refinancing in Texas, that qualified customer acquisition must remain below a particular threshold and that certain sources or acquisition methods should be excluded. The system could then search available inventory, compare opportunities, allocate budget and adjust the campaign as performance data arrives.
From an efficiency perspective, this is an obvious improvement. From a traffic-quality perspective, it creates a much harder problem.
Faster buying doesn't make traffic better
An AI system can compare thousands of opportunities faster than any human buyer. It can calculate expected returns, identify patterns in historical performance and optimize spending continuously. What it cannot do is determine the quality of information that does not exist or verify claims for which it has no reliable evidence. If a traffic source is badly described, incorrectly labelled or deliberately opaque, making the buying process faster does not make the underlying traffic better.
That distinction matters because digital advertising still has a significant supply-quality problem. Pixalate's Q2 2026 benchmarks classified 41% of the global open-programmatic mobile-app traffic it measured as invalid traffic. The figure should not be interpreted as meaning that 41% of all mobile advertising is fraudulent: invalid traffic is a broader category and Pixalate's dataset predominantly reflects buy-side open-auction programmatic inventory. But it demonstrates the scale of the problem automated buyers are entering.
There are similar issues elsewhere in the acquisition market that are harder to capture in a single IVT metric:
- A lead can belong to a real person and still be commercially worthless.
- Traffic can come from real users but originate from an acquisition method an advertiser would never knowingly approve.
- A publisher can supply genuine conversions that were heavily incentivized.
- A finance lead can have valid contact details but have been resold repeatedly before reaching the buyer.
- Casino traffic can convert while still creating compliance problems for the operator or agency purchasing it.
None of those problems disappears when AI becomes the buyer.
Automation repeats bad assumptions at scale
In some cases, automation may actually make them more important. One of the great advantages of agentic buying is scale: software can evaluate and purchase supply much faster than a human team. But the same mechanism that allows a good buying decision to be repeated thousands of times can also repeat a bad assumption thousands of times. If the agent has incomplete information about the origin or quality of supply, greater efficiency can simply mean buying questionable traffic more efficiently.
This is why the conversation about agentic advertising cannot stop at whether AI can negotiate a media buy or optimize a campaign. The more important question is what information the agent has available when it decides that one source is better than another.
Price and historical conversion rate are not enough.
Consider two publishers offering insurance leads in the same market at roughly the same CPL. One generates leads directly through its own content and search traffic, delivers them in real time, maintains clear consent records and does not resell the same consumer to multiple buyers. The other obtains leads through several intermediaries, provides little information about the original acquisition source and distributes the same records to multiple advertisers.
A human buyer with industry experience might discover those differences through due diligence, conversations with the supplier and a small test campaign. An AI agent can potentially do the same analysis much faster, but only if those differences are represented in data it can access.
A lesson from financial markets
The problem therefore begins to resemble one that financial markets solved long ago. Faster trading is useful only when the instruments being traded are clearly defined and market participants can access reliable information about them. Advertising is moving toward increasingly automated transactions while much of the underlying supply remains fragmented, inconsistently described and difficult to compare.
This gap becomes particularly obvious outside the largest closed advertising platforms. Google, Meta and Amazon have enormous amounts of internal information about their own inventory and users. Independent media markets are much less standardized. Supply is distributed across publishers, networks, affiliates, agencies, marketplaces, apps, commerce media platforms and direct commercial relationships. Two sellers can use the same description for traffic that was generated in completely different ways.
Supply has to become legible
For agentic buying to expand across that wider market, supply will need to become more legible.
That does not mean every publisher must reveal its entire acquisition strategy or expose commercially sensitive information. It does mean that some characteristics of traffic are likely to become increasingly important to buyers:
- where the traffic originates;
- whether the seller is the direct publisher or an intermediary;
- which GEO and devices are involved;
- whether the inventory has been resold;
- what compliance restrictions apply;
- how users were acquired;
- what historical evidence exists about downstream performance.
Different verticals will require different information. An insurance buyer may care about consent provenance and lead age. A casino operator may care about licensing, GEO and acquisition methods. An app advertiser may care about invalid traffic and attribution fraud. An ecommerce company may care about purchase behavior and first-party audience signals. There will not be one universal definition of quality, because quality depends on what the advertiser is trying to achieve.
Why "verified traffic" needs care
This is one reason the phrase "verified traffic" needs to be used carefully. No marketplace or verification provider can simply stamp a source as good and make every future impression, click or lead trustworthy. Quality has multiple dimensions, and a source that works well for one buyer may perform poorly for another.
A more useful goal is transparency.
If an advertiser, or an AI acting for that advertiser, can see enough reliable information about a source to make an informed decision, the market becomes easier to evaluate. Verification can then be applied to specific claims rather than used as a vague guarantee of quality. A publisher might verify its identity, ownership of a domain, historical performance or direct relationship with a particular source of inventory. A buyer can decide which of those signals matters for its campaign.
This distinction is important because agentic media buying changes the economic value of structured information.
When a human buyer evaluates ten publishers, missing information can sometimes be resolved with ten conversations. When an AI agent evaluates ten thousand media opportunities, that model stops working. Information about the supply needs to be available in a form the system can interpret before the transaction takes place.
The marketplace that simply contains the largest number of traffic offers may therefore be less useful than one containing fewer opportunities that are substantially easier to understand.
Why Traffberry exists
This is where the development connects directly with the reason Traffberry exists. The problem in the traffic market is not simply that advertisers need more places to buy traffic or publishers need another place to sell it. There is already an enormous amount of supply. The harder problem is discovery and transparency: understanding what is available, who is offering it and whether it fits the buyer's requirements.
AI can dramatically reduce the cost of searching and evaluating that market, but it cannot manufacture transparency after the fact. If supply remains opaque, the agent is still making decisions inside an opaque market.
A marketplace designed for the next generation of media buying therefore has to think beyond listings. A traffic opportunity needs enough context for a buyer to understand what sits behind it, while publishers need ways to demonstrate the characteristics that distinguish their supply from cheaper or less transparent alternatives, starting with a detailed listing.
That becomes even more important if the buyer on the other side of the marketplace is eventually an AI agent, as we described in The Media Buyer Is Becoming an AI Agent.
An agent looking for traffic should be able to understand whether a seller is a direct publisher or reseller, what audience and geography are available, what acquisition methods are involved, what restrictions apply and which quality signals have been verified. It should also be able to distinguish between information supplied by the seller, information verified independently and performance information generated through previous transactions. Our ad network profiles follow the same rule on a small scale: every fact is shown with its source.
Humans stay, the information bar rises
This does not require removing humans from the transaction. Large media deals, regulated verticals and strategic partnerships will continue to involve judgment, negotiation and relationships. Agentic systems are more likely to remove repetitive work around discovery, comparison, campaign setup and optimization than to eliminate the people responsible for acquisition strategy.
What they will change is the amount of information the market needs to expose.
For years, digital advertising solved scale by automating the transaction. Programmatic systems made it possible to buy billions of impressions without negotiating every placement individually. AI now promises to automate another layer of the process by reducing the amount of manual work required to decide what to buy in the first place.
The next problem is whether the underlying supply is ready for that level of automation.
If an AI agent can evaluate a thousand traffic sources in seconds but cannot determine where the traffic originated, whether it has been resold, whether it complies with the advertiser's requirements or whether its reported performance represents genuine commercial value, speed is not the limiting factor anymore.
Information is.
That may be the more important lesson from the first wave of agentic advertising. The industry is rapidly developing machines capable of buying media. It now needs to make the media they are buying considerably easier to understand.
Sources
- IAB Europe: The Impact of AI on Digital Advertising Report 2026
- Magnite: Magnite launches first agentic campaign in EMEA with Amnet France (September 2026)
- IAB Tech Lab: IAB Tech Lab introduces AAMP 3.0 with OpenProposal
- Pixalate: Q2 2026 Ad Fraud Benchmarks Report
- Amazon Ads: Introducing Amazon Ads Agent