AI Governance Institute
← News
Research2026-10-25

Research Finds AI Text Harder to Read Than Human Writing, Raising Output Quality Risk

Source

Why is AI-generated language so hard to understand?

arXiv / University of Texas Austin (Mahowald lab)

What happened

Researchers at the University of Texas at Austin's Mahowald lab published Why is AI-generated language so hard to understand?, a psycholinguistic study. It tested nine frontier and near-frontier large language models. The study found that AI-produced text systematically uses sentence structures that are harder to follow than equivalent human writing. Specific patterns include inanimate subjects paired with action verbs, unusual word combinations, and long gaps between related parts of a sentence. The researchers attribute this to training on math, code, and reasoning tasks, which rewards packing in information rather than communicating clearly. The findings have direct bearing on regulations and frameworks that require AI-generated content to be understandable, accessible, and non-deceptive for end users.

Why it matters

  • ·Regulations including the EU AI Act (Regulation (EU) 2024/1689) and several US state laws require that AI-generated content be intelligible to the people it affects. Systematically hard-to-read AI output may create a compliance gap that teams have not yet assessed.
  • ·Human oversight of AI decisions depends on reviewers being able to understand what the AI produced. If AI text is structurally harder to process, oversight staff may miss errors or approve outputs without fully comprehending them, weakening any human-in-the-loop control.
  • ·Organizations that rely on AI to draft customer communications, disclosures, or regulated documents face reputational and legal risk if those outputs are later found to be difficult to understand. This is especially acute in healthcare, financial services, and legal contexts where plain-language obligations apply.

Stay ahead of stories like this

Get every Global AI governance development like this one, plus the rest of the week's developments. Every Thursday.

Free every Thursday. Unsubscribe anytime.

Governance controls affected

SAF-002AI Output ValidationHOC-003AI Output Review WorkflowHOC-004Automation Bias PreventionALC-004AI Explainability DocumentationMGV-010AI Output Pre-Publication Verification for High-Stakes Claims

What to do now

  • ☐Identify every workflow where AI generates text that end users or regulators will read, including customer notices, disclosures, reports, and communications, and flag them for readability review.
  • ☐Ask your AI quality assurance or compliance team whether current output review processes include a check for readability and plain-language standards, not just factual accuracy.
  • ☐Review any plain-language or intelligibility obligations in your sector regulations and confirm that AI-generated content in regulated documents meets those standards before publication.
  • ☐Brief human reviewers who sign off on AI-generated text about the tendency of AI models to produce structurally complex sentences, so they know to actively simplify, not just approve.
  • ☐Where AI output feeds directly to customers or patients without human editing, consider adding an automated readability scoring step as a pre-publication gate.

What to watch next

Regulators in the EU and several US states are increasingly scrutinizing whether AI-generated content meets accessibility and intelligibility standards, particularly in high-risk sectors such as healthcare and financial services. As the EU AI Act (Regulation (EU) 2024/1689) conformity assessment obligations mature, expect auditors to ask for evidence that AI outputs were tested for comprehensibility, not just accuracy. Standards bodies may also begin developing benchmarks for AI text readability, which would convert this research finding into a formal compliance requirement.

Related Coverage

Corporate Policy2026-11-14

FAA's $875M AI Traffic System Rolls Out Without Controller Union Input

The Federal Aviation Administration has deployed its $875 million SMART air traffic management system to all 22 air route traffic control centers across the US. This expands from an initial pilot covering Washington, DC airspace. The system uses cloud-based AI to combine weather, flight path, traffic flow, and staffing data to recommend routes and arrival sequencing. Human FAA specialists validate recommendations before they reach local controllers, who retain final authority over whether to accept them. The National Air Traffic Controllers Association, representing more than 15,000 members, stated it was not consulted in the design, testing, or implementation of SMART.

Corporate Policy2026-11-14

HSI Demands Auditable AI Outputs and Source Citations for All Criminal Investigations

Homeland Security Investigations (HSI) has stated that any AI tool used in federal criminal investigations must produce auditable outputs with citable source documents. All investigative materials are subject to legal discovery. Peter Hatch, HSI's principal assistant director for intelligence, made the requirements explicit at a public event and said vendors who cannot meet them will not be considered. HSI currently operates nearly a dozen deployed AI use cases and has five more in pre-deployment phases.

Research2026-11-12

AI Chatbots Surfaced Fake Retail Sites in up to 15% of Shopping Queries

IP protection firm Podqi tested ChatGPT, Gemini, and Perplexity across 6,795 shopping-related queries covering 151 brands. All three chatbots surfaced risky URLs in a material share of responses. Risky sites included fake impersonator pages, knockoff retailers, and expired domains reregistered by third parties. Perplexity showed risky sites in 15 percent of answers, ChatGPT in 8 percent, and Gemini in 4 percent.