🤖 AI

Full Disclosure, Less Trust? How the Level of Detail about AI Use in News Writing Affects Readers' Trust

This study reveals that while detailed AI disclosures in news reduce reader trust, they simultaneously increase source-checking behavior, suggesting a trade-off between transparency and trust rather than a universal "transparency dilemma."

Original authors: Pooja Prajod, Hannes Cools, Thomas Röggla, Karthikeya Puttur Venkatraj, Amber Kusters, Alia ElKattan, Pablo Cesar, Abdallah El Ali

Published 2026-06-10

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Pooja Prajod, Hannes Cools, Thomas Röggla, Karthikeya Puttur Venkatraj, Amber Kusters, Alia ElKattan, Pablo Cesar, Abdallah El Ali

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you walk into a bakery. You see a sign that says, "We use robots to help mix our dough."

Now, imagine two different versions of that sign:

The Short Sign: "Robots helped mix the dough."
The Long Sign: "Robots helped mix the dough, knead the bread, check the temperature, and suggest the recipe. A human baker still tasted it before it went in the oven. If you find a burnt crust, email us at baker@bakery.com."

This study asked a simple question: Does telling customers exactly how much the robot helped make them trust the bread less?

The Big Surprise: The "Transparency Trap"

For a while, experts worried that if news organizations admitted, "Hey, we used AI to write this," people would stop trusting them. This is called the "Transparency Dilemma." The fear was that honesty would backfire.

The researchers tested this by giving 40 people news articles (some about politics, some about lifestyle) with three different types of signs:

No Sign: No mention of AI at all.
Short Sign: A one-sentence note that AI was used.
Long Sign: A detailed paragraph explaining exactly what the AI did and how humans checked it.

Here is what they found:

The Short Sign was a "Goldilocks" solution. People trusted the news just as much as if there were no sign at all. It was honest, but it didn't scare them off.
The Long Sign was the problem. When people read the detailed, long explanation, their trust dropped. They were less likely to subscribe to the news outlet.
The "No Sign" group was the most relaxed. They didn't check the ingredients as much.

The "Skeptic's Checklist" vs. The "Wallet"

The study looked at two ways people reacted:

Checking the Source (The Skeptic's Checklist): When people saw any mention of AI (short or long), they became more curious. They were more likely to click a button to see where the story originally came from. It's like seeing a "Made with AI" label and thinking, "Okay, I want to double-check the recipe."
- Crucial Finding: The researchers found that people checked sources mostly because they were curious or interested in the topic, not necessarily because they didn't trust the news.
Subscribing (The Wallet): This is where trust mattered most. When people saw the Long Sign, they were less likely to pay for a subscription. The detailed explanation made them feel the news was less reliable, even though the content was the same.

The Great Paradox: What People Want vs. What They Do

This is the most interesting part of the story.

What they did: When given the choice, people trusted the news more when the sign was short.
What they said: When interviewed, two-thirds of the people said they actually preferred the Long Sign. They wanted the details! They felt it was more honest.

It's like a customer saying, "I want to know every single ingredient in this cake," but then saying, "Actually, I'll buy the cake with the short label because the long list of ingredients makes me nervous."

The "Menu" Solution

The researchers asked people what they would do if they could choose. Many people who liked the detailed info came up with a clever compromise: The "Detail-on-Demand" Menu.

Imagine a sign that says: "Robots helped make this. [Click here for the full recipe]."

If you are curious, you click and read the long, detailed version.
If you just want to read the news, you ignore the link and get your short, quick summary.

The Bottom Line

The paper concludes that honesty doesn't always hurt trust, but too much detail can.

Short disclosures are safe; they keep trust high while still being honest.
Long disclosures trigger a "transparency dilemma" where people feel overwhelmed and less trusting, even though they claim to want that information.
The Fix: Give people the short version by default, but let them click to see the long version if they really want to dig deeper.

In short: Don't overwhelm your readers with a novel about how the news was made. Just give them a quick note, and let them decide if they want to read the rest.

Technical Summary: Full Disclosure, Less Trust? How the Level of Detail about AI Use in News Writing Affects Readers' Trust

Problem Statement

As artificial intelligence (AI) becomes increasingly integrated into news production, there is a growing demand for transparency regarding its use. However, recent literature identifies a "transparency dilemma," where disclosing AI involvement paradoxically reduces readers' trust rather than enhancing it. While previous studies have examined the effects of generic "AI-generated" labels, there is a lack of understanding regarding how the level of detail in these disclosures (e.g., a simple one-line statement versus a comprehensive explanation) influences trust, source-checking behavior, and subscription decisions within the specific context of news reading. Furthermore, the impact of varying degrees of AI involvement (from minor editing to substantial generation) on these outcomes remains under-explored.

Methodology

The authors conducted a 3×2×2 mixed factorial laboratory study with 40 participants (final sample $N=34$ after excluding those who missed disclosures). The study design manipulated three independent variables:

Disclosure Level (Between-subjects): Three conditions were tested:
- No Disclosure: No mention of AI.
- One-line Disclosure: A brief statement indicating AI use (e.g., "An AI tool was used to add an extra layer...").
- Detailed Disclosure: A comprehensive statement describing specific production steps, human oversight, and error reporting contacts.
News Type (Within-subjects): Participants read articles from two domains: Politics (hard news) and Lifestyle (soft news).
AI Involvement (Within-subjects): Two levels of AI integration were simulated using ChatGPT-4o:
- Low Involvement: Minor edits to source material.
- High Involvement: Substantial content generation based on source material.

Stimuli and Procedure:
Participants read four articles per session across three sessions. They completed the News Media Trust questionnaire (measuring fairness, bias, and accuracy) at both the article and outlet levels. Additionally, two decision-making tasks were employed:

Source-Checking: Participants spent limited "tokens" to reveal the original source of an article, serving as a proxy for skepticism or verification behavior.
Subscription: Participants decided whether to subscribe to the fictional news outlet.

Semi-structured interviews were conducted to gather qualitative insights into the motivations behind these behaviors and preferences for disclosure formats.

Key Results

Quantitative Findings

Trust Scores: Detailed disclosures resulted in a statistically significant decline in trust compared to both one-line and no-disclosure conditions, particularly for lifestyle articles. One-line disclosures yielded trust scores comparable to the no-disclosure condition.
Source-Checking Behavior: Both one-line and detailed disclosures increased token spending (source-checking) compared to no disclosure. The effect was most pronounced for detailed disclosures. Notably, qualitative analysis revealed that source-checking was primarily driven by topic interest/curiosity rather than a lack of trust.
Subscription Decisions: Subscription rates were lowest in the detailed disclosure condition, especially for high-AI involvement articles. Unlike source-checking, subscription behavior was strongly correlated with trust levels.
AI Involvement: While high AI involvement generally led to more source-checking and lower subscriptions, the statistical differences were less pronounced than those driven by the level of detail in the disclosure.

Qualitative Findings

Preferences: Approximately two-thirds of participants expressed a preference for detailed disclosures, citing a desire for transparency and accountability (e.g., error reporting contacts).
Detail-on-Demand: Among participants who preferred one-line disclosures (or found detailed ones too long), the majority suggested a "detail-on-demand" format (e.g., a short summary with a link to expand), allowing readers to choose their depth of information.
Motivations: Interest in the topic was the primary driver for checking sources, whereas trust was the decisive factor for subscription.

Key Contributions

Granularity of Disclosure: The study demonstrates that the "transparency dilemma" is not uniform; it is exacerbated specifically by detailed disclosures, whereas concise (one-line) disclosures do not significantly erode trust compared to no disclosure.
Behavioral Distinction: The research distinguishes between two types of reader behaviors: source-checking (driven by curiosity/interest) and subscription (driven by trust). This clarifies that increased verification behavior does not necessarily equate to a total loss of trust in the outlet.
Trade-off Identification: The paper identifies a trade-off between the readers' desire for transparency (favoring detailed info) and the preservation of trust (favoring concise or no info).
Design Recommendations: The authors propose that news organizations can mitigate the transparency dilemma by adopting concise disclosures or "detail-on-demand" interfaces, rather than avoiding disclosure entirely.

Significance and Claims

The paper claims that not all AI disclosures lead to a transparency dilemma. Instead, the negative impact on trust is contingent on the form and framing of the disclosure.

Practical Implication: News organizations can fulfill transparency obligations without necessarily sacrificing audience trust by utilizing concise disclosures or interactive "detail-on-demand" designs.
Policy Implication: The findings suggest that regulatory frameworks (such as the EU AI Act) should consider the nuance of disclosure formats. Mandating overly detailed technical disclosures for all content may be counterproductive, potentially leading to information overload and reduced engagement.
Theoretical Contribution: The study reframes the transparency dilemma not as an inevitable consequence of AI use, but as a design challenge involving the balance between accessibility, informativeness, and user behavior.

The authors conclude that while detailed disclosures align with reader expectations for transparency, they currently incur a trust cost. Future work should explore diverse populations and multimedia formats to further validate these findings.