The story. A new supply chain study says more than 2,100 AI-generated recipe sites are monetising through programmatic ad tech, with fake author profiles and crowded ad layouts. The report names no crime. It does show how junk inventory blends in. (AdExchanger, 21 October 2026)
01What happened
- The scale. DecryptAds, a supply chain intelligence firm, found 2,116 recipe sites populated with AI-generated content. Ads from blue chip and Fortune 500 brands appeared on them.
- The same owners run more. The operators also run political news aggregators and forums for cars, anime and video games.
- How the list was built. DecryptAds started with about 300 ad tech companies in what it calls “risky geos”. It listed their publishers from ads.txt files and found 400 suspicious recipe sites by keyword.
- Then it widened the net. It confirmed AI avatars on those 400 by hand and built more than 190 technical footprints to find sites with the same owners.
- One SSP carries most of the sample. DecryptAds ties HBAgency to 1,631 of the 2,116 sites. HBAgency says every publisher is screened by an independent fraud provider and that it works directly with only one of the partners named.
- Others are named too. Ezoic is linked to 407 sites and Culture G to 34. MonetizeMore is linked to 11 and says none runs its demand.
- The evidence has limits. AdExchanger corrected the piece to say the HBAgency partner links came from account IDs in a publisher’s ads.txt file. Zach Edwards of DecryptAds says the report found no evidence of crimes.
- The industry is split on value. Ezoic says AI authorship does not by itself decide value. Google’s new helpful content update treats fabricated creator profiles as deception.
02What it means inside a GAM network
Most publishers will read this as an SSP problem, and it is partly that. The part that reaches a publisher’s own desk is the shared plumbing. ads.txt and sellers.json are the only public records of who sells whose inventory, and the study shows how far a chain of resellers can hide an owner.
The same files that expose these sites also protect or expose you. If your ads.txt lists a reseller that lists another reseller, your inventory sits in the same chain that carries slop. Buyers who filter on that chain will filter you out with it.
The HBAgency exchange is a useful warning. The report’s link rests on an ads.txt entry, the SSP disputes the chain, and the author says the data is too tangled to pin on one company with full confidence. A publisher cannot settle that argument from outside, so the safer course is to shorten your own chain.
One fair counterpoint: Ezoic argues that an accurate recipe that works serves a real reader. That may be true of a single page. It does not answer the layout problem, with overlapping placements built to game viewability, which is what buyers read as MFA.
This study counts sites, not crimes, and its attribution is contested.
03What publishers should do about it
04The bottom line
This study counts sites, not crimes, and its attribution is contested. The lesson for publishers is the same either way: the shorter and cleaner your seller chain, the less of this inventory sits next to yours.