The Maze: Advertising researcher DecryptAds identified 2,116 recipe websites with misleading or apparently AI-generated author profiles and links to online ad sellers. Its study exposes a problem for brands buying ads across websites: an approved seller list does not establish whether each listed firm actually sold the space.
The Details:
The sample followed technical clues, not a census. DecryptAds examined advertising firms it classified as geographically risky, then used food-related keywords to find an initial pool of about 400 sites. Its October 18 report says researchers checked at least three pages per site for AI-style author photos and text indicators. More than 190 technical fingerprints, including shared advertising account identifiers and website infrastructure, expanded the pool. Analysis ran from July to October 2026. That process can uncover related operators; it cannot tell advertisers what proportion of the whole web is affected. Nor does an artificial-looking portrait establish how every recipe was produced or whether anyone used it successfully.
Outsourcing makes the selling chain harder to read. In automated advertising, software buys space on websites through sales intermediaries. A publisher may hire one partner to manage those relationships. The partner can add its own roster of sellers across client sites. The public file called ads.txt lists authorized sellers. Another file, sellers.json, helps identify the businesses behind seller accounts. Those records offer investigators useful connections. They do not certify the usefulness of the content. DecryptAds also acknowledges that a seller can remain listed after a publisher switches it off, and that a named manager does not necessarily control every account in a file.
The rebuttals change what the map proves. Advertising sales business HBAgency disputed several direct commercial connections inferred from listed accounts. Publisher advertising specialist MonetizeMore said it was not actively supplying its own or partners’ ads to the 11 sites linked to it, and described one legacy reporting connection. These responses were published on October 20. They are company statements, not audited findings, but they matter: a historical account entry and a current payment are different things. Advertising platform Ezoic also argued that useful AI-assisted recipes can attract real readers. The reporting supplies no total advertising spend or measured conversion loss for the 2,116-site sample.
MarketMaze Take: The useful output is a queue for investigation, not a bill for proven waste. For advertisers anywhere buying through the same open-web sales chain, authorization, content quality and actual transactions are separate checks. Collapsing them into one verdict makes both buying decisions and accusations less reliable.
An approved seller is only the first question. A file can show that a business is allowed to sell space without showing that a particular page helps a shopper or deserves a brand’s budget. The crowded placements and misleading author identities give buyers a concrete reason to inspect those pages. AI use alone does not settle their value.
Follow the purchase before assigning the loss. Seller records can narrow an audit, but transaction evidence is needed to establish which intermediaries actually sold the space. The vendor rebuttals make that distinction material here. The researcher acknowledged attribution uncertainty and said the report did not establish crimes by the sites or their advertising partners.
Sources: DecryptAds, AdExchanger, IAB Tech Lab
