What are the Sellers.json files?

As explained in our Ads.txt study, publishers use services called SSP (for "supply-side platform") to market inventory items (i.e. advertising impressions) on an ad exchange network. Each of these SSPs is therefore authorised by the publisher to sell inventory from its website on a certain number of ad exchanges, whether this SSP is directly integrated on the publisher's page or is external and therefore a simple reseller. Sometimes, an inventory item can pass through several SSPs in the chain to be sold, which is not described in the Ads.txt files.

The Sellers.json file describes the commercial relations between the different SSPs and the publishers, making it possible to identify all the intermediaries during the sale of inventory.

What are the additional elements compared to Ads.txt?

The types of frauds targeted by the Sellers.json file are quite similar to those targeted by Ads.txt (i.e. site usurpation and inventory resale), but this mechanism provides other tools to detect them.

Thus, it makes it possible:

• For publishers to check that the sales channel of their inventory meets their expectations.
• For buyers to identify the third parties involved in the inventory sale, thus avoiding dubious actors.
• For third parties to detect, through inconsistencies between the Ads.txt and Sellers.json registers, potential fraud.

Why do we care?

While Ads.txt files identify the relationships between online advertising companies and website publishers, Sellers.json files allow you to visualise the intermediary players involved in the sale of inventory. Each inventory item sold is associated with a unique identifier of the user, which is generally stored in a cookie. This identifier makes it possible to build advertising profiles of users, for example by tracing their browsing through the web. To better understand these practices, read the article on the RTB (in French), one of the most popular programmatic sales modalities.

By visualising the relationships between the different advertising networks that take place when an item of inventory is put up for sale, we lift the veil on an industry characterised by a high level of intermediation.

Methodology

We use the top 20 French web advertising companies publishing a Sellers.json file (see a random one) based on our Ads.txt study. For each of these players, if they list other SSP intermediaries, we add them to our list and carry on. Only files with a URL are considered valid.

This method enabled us to identify 6017 SSP, present on 217 985 websites.

We crawled this sample on to the 14th September 2020.

Disclaimer: these data are based on the values declared by the different actors studied. The results of this study are therefore dependent on the accuracy of these declared data.

The results

SSPs generally have only a limited number of direct agreements with publishers.

Here is the visualisation of the number of entities by typology that have links with the top 20 SSPs (these were identified during the study on Ads.txt). Using Sellers.json, we can see that these SSPs can declare relationships with 3 types of players:

• publishers;
• intermediaries;
• mixed players, fulfilling both roles.

Click to group the values by SSP: .

As can be seen in the visualisation, even the largest SSPs rarely have direct relationships with more than 1000 publishers. On average a SSP has relationships with 671 publishers and 72 intermediaries. There are of course exceptions to this rule. For example, a single SSP offers inventory from more than 130,000 publishers.

Here is a visualization of the relationships between 20 of the most present SSPs on the French web.

Move your mouse over the links to see the data flows to each SSP.

We can see that SSPs are characterised by a very high interconnection rate, making it possible to offer each element of publishers' inventory over a maximum number of advertising exchanges. There are 51 interconnection between the 20 most present SSPs.

This means that an extremely large number of SSPs can have access to the French Internet users’s data via these interconnections.

Here is a visualization of the relationships between 20 of the most present SSPs on the French web.

Move your mouse over the links to see the data flows to each SSP.

We can see that SSPs are characterised by a very high interconnection rate, making it possible to offer each element of publishers' inventory over a maximum number of advertising exchanges. There are 51 interconnection between the 20 most present SSPs.

This means that an extremely large number of SSPs can have access to the French Internet users’s data via these interconnections.

By using these relationships, the most important SSPs are all able to reach almost all publishers.

The main difference between them is the number of intermediaries they need to reach each publisher. The graph opposite counts the number of domains reached according to the number of intermediaries, with a maximum of 3 intermediaries.

Click to group the values: .

It can be observed that the majority of SSPs can reach and therefore sell inventory from more than 240 000 publishers. The main difference is that an SSP with fewer intermediaries will be able to offer the publisher's inventory at a lower cost.

A dense and opaque sales network

The SSP landscape is characterised by a high level of intermediation, interconnection and redundancy of players. The network for the sale of inventory is generally unknown to users, and sometimes even to publishers. As a result, even publishers who choose to authorise only a limited number of advertising agencies in their domains potentially expose their users' data to numerous third parties, as this data usually accompanies the sale of advertising space in the context of targeted advertising.

Below you can view these potential resale path for sites identified in this study as part of the French Alexa Top 50.

Select the name of the site whose partners you want to explore:
You can then click on the actors to view the links that connect them.
Choose

Article based on a scan made on 14 September 2020.

Find all the source data and the code of this page on the CNIL's Github.