Who gets shown which political content?

RE.ELS runs controlled accounts and records everything each one is shown. Between two accounts, only one trait differs: where they live, their declared age, what they were seeded to watch. What changes in the feed is the recommender’s choice.

Platform
Mainly TikTok, for now
Measures
What feeds serve, not what gets posted
Method
Open, and published with every study
Scroll through one evening of two feeds

One persona, one phone. A fresh account with a declared age, here 18–24, on a phone of its own. We scroll its feed the way a person would and record every video it is served. Yellow ones were classified as political.

Its twin changes one trait. Set up the same way, in the same city, on the same evening. Only the declared age differs: 55+. Neither account searched for anything or followed anyone.

The difference is the finding. 9 political videos of 24 against 3 of 24. With everything else held equal, the space between the two lines is the recommender’s choice.

Then we repeat it. One evening is an anecdote. A real study repeats the comparison over many personas and sessions and reports exposure with its uncertainty. The bands: thirty illustrative sessions each, middle 90% of outcomes.

Scroll for the next step

One evening, two feedsIllustrative data, not a measured result
Persona A, declared 18–24Persona B, declared 55+Classified as political

How a study runs

Most monitoring counts what was published and estimates who could have seen it. RE.ELS records what each feed actually delivered, and to whom.

  1. Set up personas

    Fresh accounts, one per phone, on a home connection in the persona’s city. Between two personas exactly one trait differs: place, declared age or viewing history.

  2. Scroll and record

    Each persona scrolls at a human pace. Every video served is recorded with its place in the feed and a timestamp. Sessions our checks flag are dropped.

  3. Classify and check

    Political or not, then leaning or party, and topic. Human coders score a sample against the automatic labels.

  4. Compare

    Exposure between persona groups, or before and after a vote, reported with its uncertainty.

RE.ELS makes no attribution claims. Who is behind a campaign, and whether accounts are coordinated, is a question for other kinds of analysis.

Work with us

We run these audits, and we design them for others.

Commission a run

On a given election, market or platform. You get the data and the write-up.

Partner on a study

With newsrooms, universities, regulators and civil-society groups.

Hire us directly

Measurement design, statistics and review of your own audit.

Build your tooling

Capture, classification and validation, so your team runs it in-house.

info@dotted.spaceWrite to us

Who we are

Two ETH Zurich graduates who treat feed audits as measurement problems first.

Jacopo Bottoni

Astrophysics, ETH Zurich. Statistical methods, inference from noisy observations.

Michael Galli

Systems biology, ETH Zurich. Experimental design and classification.

RE.ELS stays a research project with an open method. The work around it is what we offer.

More about us on dotted.space →

dotted space GmbH is a start-up company of ETH Zurich, supported by UPortunity, the ETH start-up accelerator, and the ETH Student Project House.

A Start-up Company of ETH Zurich UPortunity — ETH Start-up Accelerator enabled by UBS ETH Student Project House