Dataset

Where a publisher’s ads.txt and their ad network’s own seller list disagree, across 462 checked relationships

Two files are meant to say the same thing and nobody publishes how often they do not. A publisher hosts a file called ads.txt naming every company allowed to sell their advertising space, with the account number each one holds for them. The advertising network publishes a matching file, sellers.json, naming every account it may sell for and the site that account belongs to. The specification says the account number is the same in both. A matching account number is one check; the seller domain and relationship also need review. Direct browser requests may be restricted by cross-origin rules. We ran it across 462 relationships between one publisher and one network, drawn from 500 domains sampled at random from a 16,085-domain frame from Mediavine and Raptive. We compared the results against six network rosters. Of 462 relationships, 426 matched, 35 were roster-only and one had an account-number mismatch. These public-file differences do not establish lost revenue, buyer behavior or unauthorized trading.

• 16,085 sites in the frame
• 500 domains drawn
• 462 checked relationships
• 426 confirmed by both files

Coverage

2026-09-02

Last updated

2026-09-02

Licence

CC BY 4.0

DOI

10.5281/zenodo.22280372

What a row is

One relationship between a publisher and an advertising network: whether both files agree about it, and what the publisher’s own file looked like when it was read.

Subjects

ads.txt, sellers.json, authorised digital sellers, programmatic advertising, ad fraud, domain spoofing, supply chain transparency, IAB Tech Lab

Download the public data sample462 rows · 10 columns · CSV · no account, no email

How to cite this

Use it commercially, change it, build on it. The only condition is credit, and a link satisfies it.

What you should know about this data

• Neither file carries a date. Nothing in an ads.txt or a sellers.json says when an account was added or last confirmed, so a disagreement cannot be blamed on either side. What can be said is what a buyer reading both files on the day would have seen.
• There are two denominators and they must not be mixed. A row is one publisher against one network, so a publisher on two networks contributes two rows. The number of publishers behind this table is 457, not the number of rows.
• A file that could not be fetched says nothing about a publisher. The 43 domains that refused are excluded from the denominator rather than counted as having no file.
• These are small, advertising-funded content businesses named by six advertising networks. They are not the web, and a publisher with no ad network is not in the frame at all.
• The domain is not published. Every field the comparison used is, so the counts can be recomputed; which site each row is cannot be recovered from it.

How it was measured

500 domains were drawn at random from a 16,085-domain frame in Mediavine and Raptive seller lists, using a published seed so the same draw can be repeated. 457 domains were reachable and 43 refused or could not be reached; the ones that could not be reached are excluded rather than counted as having no file. Each domain was then compared against the seller lists of all six networks, one publisher-and-network pair at a time, matching on the account number, because that is the field the specification says the two files share. A pair is confirmed when the publisher’s file names the network and the network files that number against that site. It is roster-only when the network claims the site and the publisher’s file says nothing about it, and not-in-roster when the file names the network and that number appears on the network’s list nowhere. Rows are in the order the sample was drawn.

The summary

The first 60 of 462 rows. The download has all of them. Free to use with credit, commercially included.

What each column means

network

The advertising network in this relationship

agreement

confirmed, roster-only or not-in-roster, as the method defines them

why_unconfirmed

For a roster-only row, what the publisher’s file did instead

other_network_named

The network the file named instead, where it named one of the six

site_under_another_number

yes when the network holds this site under an account number the site’s own file does not carry

ads_txt_status

The HTTP status the publisher’s address answered with

ads_txt_present

yes when that address served a usable ads.txt

records_declared

How many seller records the publisher’s file carries

unparsed_lines

Lines in the file that are neither a seller record nor a variable declaration

served_as_html

yes when the body began as a web page rather than as records

← Every dataset