Brief IA
›

Elections: Google AI Insights Unequal and Biased According to AlgorithmWatch

Models & LLM·Tom Levy·

Elections: Google AI Insights Unequal and Biased According to AlgorithmWatch

Elections: Google AI Insights Unequal and Biased According to AlgorithmWatch
Key Takeaways
1AlgorithmWatch analyzed 4,480 queries about regional elections in Germany
2Google’s AI snippets appear unevenly and rely on few sources, with YouTube being the most cited
3The tone of the snippets is often positive, with marked differences depending on the parties
4Google invokes neutrality and does not specify its display criteria
Why it matters : Google’s AI results can influence political perception, highlighting the importance of transparency and fairness in their operation.

AlgorithmWatch's work on queries related to regional elections in Germany describes Google's AI snippets as often positive in tone and displayed unevenly depending on the topics or parties. The cited sources focus on a limited number of areas, with YouTube leading the way. Google emphasizes the neutrality of its features and does not detail its criteria for appearance.

A Often Positive Tone, Changing Evaluations Depending on Parties

According to AlgorithmWatch, political positions are frequently presented with positive and vague qualifiers such as "pragmatic," "close to citizens," or "stable," which often echo the parties' own descriptions. In a sample focused on Saxony-Anhalt, 81.5% of the snippets concerning the CDU are rated as clearly positive, compared to 50% for the SPD, 10.7% for the Greens, and 0% for the AfD. AlgorithmWatch observes that a greater share of party-affiliated domains tends to correlate with more positive representations, although this does not hold uniformly across parties. The framing can also vary for the same individual: AfD candidate Ulrich Siegmund is described with labels of varying severity, ranging from far-right and national-conservative representative to right-wing populist. AlgorithmWatch indicates that it did not identify a pattern behind these variations, some occurring at short intervals. Unlike previous studies, clear factual errors were noted only very rarely.

Snippets Less Present on Electoral Topics and Unevenly Triggered

For electoral-related queries, AI snippets are present in 39.1% of situations, while they appear in 65.3% of cases for non-political queries. Queries concerning polls yield a snippet in very rare cases, whereas those about parties or personalities show snippets in a proportion between 40% and 50%. Queries about the AfD result in a snippet less frequently, at 24%, while other parties range between 45% and 58%. Overall, Saxony-Anhalt generates fewer snippets than Berlin and Mecklenburg-Vorpommern. Google does not specify the exact conditions under which a snippet is displayed.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

YouTube Dominates Links Proposed by Google's AI

Nearly 50% of all links come from a limited number of ten domains, which restricts the diversity of sources. YouTube is the most frequently cited source. Among media outlets, the German public broadcasters NDR, rbb, MDR, and ARD, as well as Zeit Online, dominate, with six media sites concentrating 75% of the links from media. Alongside Wikipedia and government sites, party pages also appear, without labels distinguishing them from neutral information sources. AlgorithmWatch interprets YouTube's prominence as a possible sign of favoritism towards a Google service, which the company denies.

Google Claims Neutrality and Remains Vague on Its Criteria

A Google spokesperson asserts that the AI search features are designed to be neutral and objective and that they do not appear for every search. The company states that snippets appear when it believes there is a strong added value, without providing the criteria applied in practice to AlgorithmWatch's request. In 2024, Google announced the blocking of responses generated by Gemini for election-related questions, a measure deemed very effective at the time by AlgorithmWatch. It has not been possible to establish whether comparable measures still apply in 2026, and Google has not commented on this point.

A Study Protocol Based on DSA Access and 4,480 Queries

AlgorithmWatch gained access to Google's Search Researcher Result API through the data access provision outlined in Article 40(12) of the DSA. The group executed 4,480 queries related to the 2026 elections in Saxony-Anhalt and Mecklenburg-Vorpommern and the Berlin House of Representatives election, then analyzed the displayed AI snippets and their sources. The organization concludes that snippets appear inconsistently, with a limited source base and sometimes partisan descriptions, noting particularly a lower frequency of appearance for queries related to the AfD compared to other parties.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.