What do users actually want from AI disclosures?
Most people expect clear, specific explanations about how AI uses their data, but companies often fall short. A 2022 study surveyed 835 data subjects and analyzed the AI disclosures of 100 companies and organizations, finding that current GDPR-mandated notices are 'vague, incomplete and lack transparency' [1]. The study showed that explanations drawn up to meet generic legal requirements differ widely and do not match what data subjects expect or need [1]. This gap between user expectations and company practice is the core problem transparency reports must solve.
Users want more than just a notice that AI is being used; they want to understand its logic and consequences. The same study found that Articles 15 and 22 of the GDPR, which require companies to inform users about automated decision-making, are often interpreted so loosely that the resulting disclosures are ineffective [1]. For a transparency report to be useful, it must go beyond legal boilerplate and address the specific questions users have about how their data is processed and what the outcomes mean for them.
Does structured transparency actually increase public trust?
Yes, and the effect is substantial. A 2026 mixed-methods study of 1,247 adults across six countries found that disclosure of transparency practices accounted for the largest variation in public trust (beta = .38, p < .001) — meaning that the presence and quality of transparency explained more of the difference in trust levels than any other factor measured [2]. The study also found that structured transparency mechanisms, formal oversight of editorial content, and explicit policies holding algorithms accountable significantly increased trust across all demographic groups [2].
The same study revealed that the format of disclosure matters: simply having a disclosure is not enough. The researchers recommend that regulators mandate 'minimum content standards for algorithmic disclosure rather than mere disclosure presence' [2]. This means a useful transparency report should follow a consistent structure, include specific details about how the AI works and is overseen, and be paired with media literacy initiatives to help users understand what they are reading [2].
What level of detail should a transparency report include?
Beyond general explanations, useful transparency reports should include timestamped, auditable records of AI behavior and anomalies. A 2025 framework for AI transparency in scientific research recommends archiving 'forensic anomaly reports, parsing failures, memory drift documentation, procedural validation gaps, and corrective actions' [3]. Each entry should be independently auditable and version-controlled to maintain integrity [3].
While this level of detail is designed for scientific research, the principle applies broadly: transparency reports should document not just what the AI is supposed to do, but what it actually did — including errors, unexpected behaviors, and how those were addressed. This forensic approach turns a static disclosure into an ongoing accountability record, which directly addresses the vagueness and incompleteness identified in the 2022 study of GDPR disclosures [1].
About These Sources
This answer is built on 3 studies (2 peer-reviewed, 1 preprint) — published from 2022 to 2026, 2 from 2024 or later, collectively cited 76 times — selected as the most relevant from 3 studies that passed quality screening, drawn from 54 papers retrieved from a database of over 500 million.
Sources used in this answer
“Please understand we cannot provide further information”: evaluating content and transparency of GDPR-mandated AI disclosures
A 2022 study surveyed 835 data subjects and analyzed 100 companies' AI disclosures, finding that GDPR-mandated notices are often vague, incomplete, and fail to meet user expectations for clarity and detail [1].
Ethical AI Communication and Public Trust: Examining the Role of Transparency in Digital Media
A 2026 survey of 1,247 adults across six countries found that structured transparency practices accounted for the largest variation in public trust (beta = .38, p < .001), and that minimum content standards for disclosures are more effective than mere presence of a notice [2].
AI Transparency and Anomaly Reports
A 2025 framework for AI transparency in scientific research recommends archiving timestamped, auditable forensic reports including anomaly records, parsing failures, and corrective actions to maintain scientific integrity [3].