An analysis of June 2026 bidding activity by DataBeat found that AI agents buying ad inventory entered 86% fewer auctions than conventional demand-side platforms, and cleared their impressions at $6.13 CPM against $6.95 for conventional buyers, a difference of 13.4%. PPC Land reported the finding on PPC Land on 16 August 2026 and argued that programmatic buying is drifting back toward the model television has used for decades.
Those two numbers pull in opposite directions once you sit with them. A buyer that enters far fewer auctions and still pays less per thousand impressions looks efficient. A buyer that meets far less competition also pays less for reasons that have nothing to do with skill. The report gives both readings room, and so should anyone planning a 2027 media budget around agent-driven buying.
What DataBeat measured in June 2026
DataBeat's June 2026 analysis, as reported by PPC Land on 16 August 2026, drew on a dataset of more than 35 billion monthly impressions and over 200 bidders, tied to more than $55 million in monthly revenue. Inside that pool, PPC Land separates two classes of demand. Conventional demand means a demand-side platform running bid logic across a very wide slice of available impressions. Agentic demand means AI agents that buy inventory themselves, deciding what to enter and what to skip.
The differences the analysis reports are narrow in number and wide in implication:
- Auction participation: agentic buyers entered 86% fewer auctions than conventional demand.
- Clearing price: $6.13 CPM for agentic buyers against $6.95 for conventional buyers, a 13.4% gap.
- Fill rate: 0.204% for agentic buyers against 0.183% for conventional, an edge of 11.5%.
- Fees: PPC Land contrasts agentic guaranteed products at roughly 1% with an incumbent platform take rate of roughly 20%.
Read the first two together and the picture is a buyer that is highly selective about where it bids and still gets its impressions cheaper. That is the headline. The second reading, and PPC Land makes this the spine of its argument, is that a bidder present in only a small share of auctions is also a bidder who is not competing against much of anyone.
Fewer auctions at a lower price can be read two ways, and the difference matters
This section is analysis rather than a finding from the DataBeat data. Price in an open auction is produced by rival bidders. The second-highest bid, or the price floor, is what stops the winner from paying whatever it likes. When 86% of auction participation disappears from one class of buyer, the price that buyer clears at is being set in a thinner room. Lower cost per thousand impressions in that setting is not automatically a saving. It can be the sound of price discovery getting quieter.
The optimistic reading is equally available and equally reasonable. Agents that skip most of the inventory pool may be skipping the parts that were never going to perform, which is what a competent media buyer does by hand. A 11.5% higher fill rate on 86% fewer entries is consistent with better targeting rather than weaker competition. Nothing in the reported figures separates the two explanations, and the report does not claim to.
The practical consequence for a marketer is the same either way. A CPM that falls without a matching movement in outcomes is a number to check, not a number to celebrate. If agent-bought inventory clears cheaper and converts worse, the cheaper price was a cost.
The comparison to television buying, in numbers
PPC Land's central claim is structural. Television locks audience volume months ahead through bilateral negotiation between buyer and seller, settles against a currency both sides accept, and verifies statistically and after the fact rather than impression by impression. Programmatic did the opposite. It treated a single impression as the unit of sale, put it in an open auction, and made per-transaction verification possible for anyone through ads.txt and supply chain objects.
The table below sets the reported figures against each other. Every row comes from the DataBeat analysis as PPC Land reported it on 16 August 2026.
| Metric, June 2026 | Agentic buyers | Conventional buyers |
|---|---|---|
| Auction participation | 86% fewer auctions entered | Baseline |
| Clearing price | $6.13 CPM | $6.95 CPM |
| Fill rate | 0.204% | 0.183% |
PPC Land's framing of the shift is direct: "Trust that competition used to supply automatically must now be asserted, recorded and audited." That sentence is the whole argument in one line. In a busy auction, nobody has to certify that a price is honest, because a rival bidder disproves a dishonest one within milliseconds. Take the rivals away and honesty becomes a claim that has to be filed somewhere.
Why AAMP 2.3 in July 2026 is the evidence for that argument
The IAB Tech Lab shipped version 2.3 of its agentic protocol in July 2026, adding a pricing provenance field intended to stop agents fabricating bid prices, alongside server-side trust verification on price-moving paths and a vendor approval gate tied to the IAB Diligence Platform.
Notice what the pricing provenance field implies. In a competitive auction that control is unnecessary, because a fabricated price loses to a real one and the market corrects itself without a protocol. A standards body only writes a field to record where a price came from when the market has stopped proving it. The control is not a criticism of agents. It is an admission about the room they are bidding in.
For anyone tracking how AI systems are changing the plumbing beneath media buying rather than just the surface, this is the more interesting signal than the CPM gap. The same principle governs how AI systems select and cite sources: when a machine intermediates the decision, the audit trail becomes the product.
Connected TV is the cautionary case, and the numbers are already in
PPC Land offers connected TV as the precedent for what happens when a market goes programmatic without carrying verification with it. 85% of CTV inventory now trades programmatically, up from 75% the year before. Only 40% of CTV bid requests carry usable programme-level signals. 43% of CTV buyers say they doubt where their advertisements actually ran. The report also cites 34 of every 100 monitored impressions running outside streaming content, and CTV fraud schemes up 140% year on year.
That is a market with high automation and low visibility, and it got there in a few years. The concern PPC Land raises is that agentic buying could reproduce the same shape across the rest of programmatic: mechanised transaction, verification left behind.
Who is affected, and who is not, yet
PPC Land cites agency adoption of agentic AI at 46% as of April 2026, but the split by task is where the caution sits. Ideation stands at 86.9% and research at 84.0%, while media planning is at 29.1% and media buying strategy at 22.1%. Most agencies using agentic AI are using it to think, not to spend.
So the DataBeat figures describe a real but early slice of the market. A brand running search and social through the standard platforms is not affected today. A brand buying open programmatic display or CTV through a partner experimenting with agentic execution may already be inside these numbers without having been told.
What to check in your own account this week
- Ask your programmatic partner in writing whether any part of your buying is executed by agentic systems, and on what share of spend.
- Pull CPM alongside a downstream outcome for the same period. A falling CPM with flat or falling conversions is the pattern worth escalating.
- For any CTV line, ask what percentage of delivered impressions carried programme-level reporting. If the answer is unavailable, that is itself the answer.
- Check that your ads.txt and sellers.json entries are current, since per-transaction verification only works when the records behind it are accurate.
What the source did not say
The report does not say that agentic buying performs worse. It does not report conversion rates, viewability or brand-safety outcomes by buyer class, so the 13.4% price gap cannot be converted into an efficiency verdict in either direction. It gives no forecast for when agentic share of programmatic auctions will reach any particular level. It does not name the platforms behind the agentic demand in the dataset. And it says nothing about Thailand or Southeast Asia specifically, so any read on this market is inference, not reporting.
What this means for Thai marketers
Open programmatic is a smaller share of Thai media budgets than it is in the United States, with much of the money sitting in the walled platforms and in LINE. That buys Thai advertisers time, and the sensible use of it is to write the verification questions into partner agreements before agentic execution arrives rather than after, since retrofitting a reporting requirement onto a signed insertion order is a slower conversation.
Thai CTV is growing fast on the local streaming services, and the 40% programme-level signal figure PPC Land cites is a global one. Nobody has published the Thai equivalent. Asking a local CTV seller for programme-level delivery reporting is a reasonable request today and a much harder one once the category has settled into whatever reporting standard it defaults to. Teams that already treat measurement discipline as part of the buy, the same discipline that makes paid search accounts auditable line by line, will find these questions easy to ask.
FAQ
Does this mean AI agents buy media better than a DSP?
The reported data does not answer that. It shows agentic buyers entering 86% fewer auctions and clearing at a 13.4% lower price with a slightly higher fill rate, but it reports no conversion or quality outcome by buyer class, so neither efficiency nor waste is demonstrated.
Is any of this live in Thailand?
The source did not state this. The DataBeat analysis covers June 2026 bidding activity across a global dataset without a country breakdown, and PPC Land made no claim about Thai inventory or Thai buyers.
Do I need to change anything in my campaigns right now?
Not for most advertisers, no. If your spend runs through search, social or direct buys, nothing here touches your account today, and the useful action is a question to your programmatic partner rather than a change to a campaign setting.
Why does a pricing provenance field matter so much?
Because it records where a bid price came from, which an open auction never needed to do. The IAB Tech Lab added it to AAMP 2.3 in July 2026, and the reason a standards body writes that field is that rival bidders are no longer establishing the price automatically.
When will agentic buying be a majority of programmatic?
The source gave no timeline. It reports agency adoption of agentic AI at 46% as of April 2026 with media buying strategy at 22.1%, but offers no projection for when agentic demand will represent a majority of auction participation.
If you are buying programmatic or CTV in Thailand and want a second pair of eyes on what your partner is actually reporting back, Relevant Audience can walk through your delivery data with you and tell you which questions your current reporting cannot answer.
