Eigen RadarAI
Analysis

Study maps how real and generated visuals combine in Türkiye’s fact-checking cases

A study of Türkiye’s Teyit.org archive groups 52 selected visual cases by their narratives and combinations of authentic and generated material. Two cases involved genuine images incorrectly called synthetic. The exploratory analysis describes disputed content and verification methods; its single-platform sample does not establish how common these forms are across the internet.

Artificial Intelligence··Morning
An illuminated inspection lightbox, optical loupe, photographic slide frames, digital camera and closed archival boxes on a bright media-research worktable in Türkiye.

Genuine images among the disputed cases

Genuine photographs appear alongside generated material in a new study of Türkiye’s Teyit.org archive. Abdullah Yıldırmaz and Yaşar Öztürk examined 52 selected visual cases, including two in which authentic images had been wrongly described as AI-generated. The other 50 involved synthetic content. Published on November 15, this single study treats mistaken allegations of generation as part of the visual-verification problem, alongside the identification of fabricated scenes.[1]

Narratives and mixed production

The classification separates geopolitical scenes, depictions of political figures, culturally sensitive material and extraordinary events presented without their context. A second set of categories concerns production: genuine and generated elements can sit beside one another, be assembled into one composition or move from a photograph into a video. A real statement can also acquire an invented illustration. These are descriptions of how the selected items were constructed, rather than a ranking of detection products.[1]

A purposively selected archive

The researchers drew the sample from 76 candidates documented between January 2024 and May 2026. One coder examined it twice, four weeks apart; independent-coder agreement was not calculated. Their four-part verification schema combines detection tools, origin signals, platform labels and human examination. It is the authors’ reconstruction, not Teyit’s published technical architecture. Purposive selection from one platform limits the findings to an exploratory account rather than an estimate of online prevalence.[1]

References

  1. News sourceBlack Sea Journal of Engineering and ScienceTurkish study proposes a case typology for AI-related visual content↩1↩2↩3