An Austrian electoral watchdog said on Friday that Mettle Mesh, a social platform with 3.2 million users in the country, wrongly limited recommendations for 1.1 million public posts by election candidates. The error affected 1,284 candidates across all nine states, cutting their average recommendation reach by 38% during the campaign for the 8 November parliamentary vote.
The case began on 29 September when Ana Zupančič, a Vienna postal clerk who volunteers at a local election noticeboard, searched the platform for a candidate’s post about a public debate. She could open the post through a direct link, but it did not appear in the platform’s public recommendation feed. On a hunch, she checked other candidates’ profiles and found the same mismatch.
A chance check, then a pattern
Zupančič sent screen recordings to the Federal Campaign Integrity Office, the public body that monitors election rules and online campaign access. Auditors then compared recommendation data for public political posts with posts on other topics from 1 August to 15 October. They found that a software classifier had labelled the affected material “civic integrity” content and quietly reduced its distribution.
The office said the error was not confined to one party or region. Candidates from 30 parties were affected, including independents, and the problem grew worse in the final weeks of the review period. Deputy director Derin Özbey said the tally was the largest verified instance of election-related downranking recorded by the office since it began monitoring platforms in 2024.
“A citizen looking for one debate notice found a system-wide failure that the campaigns themselves could not see,” Özbey said. The office has ordered Mettle Mesh to restore recommendation access for eligible posts, preserve its election-period data and provide daily reach figures until polls close. It has not accused the company of deliberately favouring any party.
Company faces a deadline
Mettle Mesh said a change to its automated “civic integrity” filter had misclassified posts that mentioned voting, ballot access or campaign events. In a notice to users, the company said it had begun correcting the setting and would tell affected candidates how to request a review. It said the filter was designed to limit misleading election material, not lawful campaigning.
Colin H. Murray, a Vienna-based digital governance researcher who reviewed the watchdog’s figures, said the scale made the error unusually consequential. “A 38% reduction is not a rounding problem: it changes who gets seen in a campaign where attention is scarce,” he said. He cautioned that the figures measured recommendations, not total views, since users could still find posts through direct links.
The discovery has prompted parties to ask the office for a public account of how the classifier was tested. The office said it would publish a final report after the election and could seek a court order if the platform misses its 20 October deadline. Zupančič said she had expected a broken link, not a national-scale fault. “I just wanted the noticeboard to have the right time and place,” she said.
