The next speaker in this panel at the 2026 International Communication Association conference in Cape Town is Kilian Bühling, whose focus is on reconstructing deleted Telegram messages to prevent potential biases in data analysis. Message deletion and data loss is a common issue in working with digital trace data; the volume of deleted messages in a data sample increases over time, so data collected a substantial time after an even will miss quite a lot of messages.
This challenges the validity of analyses of such data, as well as the reproducibility of data analyses by other researchers; indeed, it is also difficult to fully assess the completeness of the datasets we collect. It is therefore important to gather data as soon as possible after an event, and ideally also to share data across research groups (where permissible) to enhance completeness.
The present study focusses on Telegram, whose public channels and group chats are widely accessible; information diffuses here via forwarding of messages between channels, and this is also how many research projects discover new channels via snowball sampling. Telegram is popular with mainstream users in Russia and Ukraine, while it is a more fringe and alternative platform in other parts of Europe and North America.
Telegram messages contain channel IDs and message IDs; the latter are consecutive from 1 for each channel, and this enables the discovery of how many messages are missing from a channel dataset. Forwarded messages also contain such IDs, and the deletion of original messages in their channels of origin does not proliferate across forwards.
This means that messages deleted in a given channel might still exist elsewhere if they were forwarded before deleted; they are digital zombies, and researchers may be able to exploit this to restore such deleted messages to their original channel dataset. But this has obvious ethical implications, as it seems to conflict with the right to be forgotten, as implemented in the EU GDPR. However, there may be overriding considerations here, depending on the agenda of the research project.
An application of this recovery approach to several datasets enabled the project to identify a small but significant set of deleted messages (around 1-2%); these are not any messages, but those which were considered by other users to be so important to be worth forwarding, which affords them extra importance. This approach may therefore be an important addition to current Telegram research methods.
