Method

We work with tech companies

Glossary

Data void

Michael Golebiewski introduced the term in 2018 and developed it with danah boyd in a 2019 report for Data & Society.

A data void is a gap in reliable information for a specific search query. This leaves room for disinformation. An assistant with web search looks for pages relevant to the question. Pages closest to the question's wording often share the same framing

Source

Golebiewski, M. and boyd, d. (2019). Data voids: where missing data can easily be exploited. Data & Society

LLM grooming

The mass production and duplication of false narratives online with the intent of manipulating the output of large language models (LLMs), the technology behind AI assistants. The same narrative is published on a large number of pages, as in the Pravda network (also called Portal Kombat), whose automated sites republish state and pro-Kremlin content in dozens of languages. VIGINUM, the French government service against foreign digital interference, and the organisation CheckFirst have documented this network.

The American Sunlight Project, a United States non-profit organisation that researches and exposes disinformation, introduced the term in 2025. An AI assistant that searches the web can find such pages, cite them as sources and repeat the narrative in its answer

Sources

American Sunlight Project (2025). A pro-Russia content network foreshadows the automated future of info ops

VIGINUM (2024). Portal Kombat: a structured and coordinated pro-Russian propaganda network. Secrétariat général de la défense et de la sécurité nationale

CheckFirst (2025). “Pravda” Network: Worldwide Expansion and LLM, Wikipedia Pollution

FIMI

Foreign information manipulation and interference, a pattern of behaviour that is mostly not illegal but threatens or may negatively affect values, procedures and political processes. Such activity is manipulative, intentional and coordinated, meaning that separate actions follow a common plan. The actors can be state or non-state, including their proxies (persons or groups acting for them), inside and outside their own territory.

The definition comes from the European External Action Service, the diplomatic service of the European Union, and concerns behaviour, such as coordination and hidden origin, and not only false content. For AI assistants this means that judging a cited source requires asking who publishes it and how, and not only whether its claims are true

Source

European External Action Service (2024). 2nd EEAS report on foreign information manipulation and interference threats

AI prebunking

The prompt and pre-emptive filling of data voids on manipulative topics with authoritative information, where a data void is a search query for which little or no authoritative content exists. Official sources and quality press publish clear information on such a topic early, in the words people actually use and not only in the language of refutation.

Prebunking in general means warning or informing people before they meet a false claim, while debunking means correcting the claim afterwards. The project's study calls the same practice "filling a data void". An AI assistant with web search selects the pages closest to the wording of the question, so it can find the authoritative page before it finds the disinformation

Sources

Dosimeter (2026). Election-day AI poisoning in Germany, glossary

Roozenbeek, J., van der Linden, S., Goldberg, B., Rathje, S. and Lewandowsky, S. (2022). Psychological inoculation improves resilience against misinformation on social media. Science Advances, 8(34)

Stochasticity

The property of an AI assistant to answer the same question differently each time. A large language model (LLM), the technology behind the assistant, does not store ready answers: it writes every answer anew, word by word, and chooses each next word among several probable ones, partly by chance. If the same question is asked ten times, the assistant returns ten texts that differ in wording and often in the sources they cite

Source

Makhortykh, M., Sydorova, M., Baghumyan, A., Vziatysheva, V. and Kuznetsova, E. (2024). Stochastic lies: how LLM-powered chatbots deal with Russian disinformation about the war in Ukraine. Harvard Kennedy School Misinformation Review

Temperature

Temperature sets the degree of stochasticity, the property of a large language model (LLM) to return different text and often different sources for the same question. The company that runs an AI assistant sets the temperature, and a user cannot change it in the app

Source

Makhortykh, M., Sydorova, M., Baghumyan, A., Vziatysheva, V. and Kuznetsova, E. (2024). Stochastic lies: how LLM-powered chatbots deal with Russian disinformation about the war in Ukraine. Harvard Kennedy School Misinformation Review