A study published in PNAS Nexus found something that surprised many in the science communication community: AI-generated plain language summaries of scientific papers are more understandable and more trusted by the public than summaries written by human experts.

The study, covered by The Conversation, tested whether GPT-4 could produce scientific summaries that are simpler and more accessible than those written by the researchers themselves. The results showed that the AI-generated summaries improved both public understanding of the science and public perceptions of the scientists.

This is not a small finding. It challenges a fundamental assumption in science communication: that the best person to explain research is the researcher who conducted it. The data suggests otherwise — at least when the goal is plain language accessibility.

The plain language summary problem in science

Scientific papers are written for other scientists. They use specialized vocabulary, dense sentence structures, and implicit context that only domain experts share. The average research paper scores at a college reading level or above — far beyond what the average citizen can follow.

This creates a communication gap:

• Researchers write for peer review and publication, optimizing

for precision and completeness

• The public needs the main findings explained in language they

can understand

• Plain language summaries (PLSs) bridge the gap — but writing

them is time-consuming, requires a different skill set than research, and competes with the researcher's other priorities

As Science Editor notes, "a PLS is a short summary of a scientific article written in nontechnical language that makes the main idea of the paper easier to understand for a nonexpert audience." JAMIA Open adds that PLSs are "particularly important in medical research, where patients need to understand studies that affect their healthcare decisions."

The problem: most researchers are not trained in plain language writing, and the summaries they produce tend to score at a high reading level because the researcher cannot fully step outside their domain expertise.

What the PNAS Nexus study found

The PNAS Nexus study, titled "From complexity to clarity: How AI enhances perceptions of scientists and the public's understanding of science," made several key findings:

1. AI summaries were simpler. GPT-4 produced summaries with

lower reading levels than human-written versions, using shorter sentences and simpler vocabulary

1. AI summaries improved understanding. Readers who read the

AI-generated summaries scored higher on comprehension tests than those who read human-written summaries

1. AI summaries improved trust. Readers who read the AI-generated

summaries rated the scientists as more approachable and trustworthy than those who read human-written summaries

1. The effect was consistent. The simplification effect held

across different scientific domains, suggesting it is not domain-specific but a general property of AI-generated plain language text

The study's conclusion: "linguistic simplicity, facilitated by AI, can significantly influence public perceptions of scientists and crucially, also improve the public understanding of science."

Why AI outperforms humans at plain language summarization

The PNAS Nexus findings have a straightforward explanation: researchers are domain experts, not communication experts. When a researcher writes a plain language summary, they:

• Retain domain-specific terminology because it feels precise
• Include too much detail because they know what is important
• Use complex sentence structures because they are accustomed to

academic writing

• Struggle to simplify because their own understanding is so deep that

they cannot reconstruct the perspective of a non-expert

AI models have a different advantage. They have been trained on vast corpora of text at multiple reading levels — children's books, news articles, educational materials, and academic papers. They can adjust their register deliberately, choosing simpler vocabulary and shorter sentences without losing the core meaning. They do not have the researcher's curse of knowing too much to simplify effectively.

The graded approach: one paper, multiple audiences

A 2024 ACM paper by Tal August et al., "Know Your Audience: The benefits and pitfalls of generating plain language summaries beyond the 'general' audience," made an important contribution: the observation that "the general audience" does not exist. Different audiences need different levels of simplification:

• Policy makers need the key findings and implications at a

professional but non-technical level (grade 10-12)

• Educators need the findings explained in terms they can teach

(grade 8-10)

• General public needs the main idea in plain language (grade 6-8)
• Broad public reach (children, elderly, non-native speakers,

people with low literacy) needs the main idea at a very simple level (grade 5)

This is the graded simplification approach. Rather than producing a single plain language summary, the researcher produces multiple versions at different reading levels, each serving a different audience.

How ELI5 AI serves science communication

ELI5 AI is a free, no-login text simplifier that takes any scientific text — an abstract, a paper summary, a research finding — and rewrites it at a 5th-grade reading level with four levels side-by-side: 5-year-old, 10-year-old, teenager, and adult. For researchers and science communicators, it implements the graded approach:

1. Paste the abstract or paper summary into ELI5 AI
2. Get four reading levels — the same content rewritten for

different audiences

1. Publish the appropriate level for each audience — the adult

level for policy makers, the 10-year-old level for general public audiences, the 5-year-old level for broad accessibility

1. Keep the original — the academic paper is unchanged. The

simplified versions are additive

This takes seconds, costs nothing, and produces summaries that the PNAS Nexus research shows are more understandable and more trusted than human-written versions. It also solves the time problem: writing a single plain language summary can take a researcher hours. ELI5 AI produces four versions in under a minute.

What this means for science communication

The PNAS Nexus study has implications that extend beyond individual papers:

1. Journals should require AI-assisted PLSs

An increasing number of journals require plain language summaries. The PNAS Nexus data suggests that AI-assisted summaries are more effective than human-written ones. Journals should encourage or require researchers to use AI tools to produce PLSs, with human review for accuracy.

2. Researchers should treat readability as part of publication

Writing the paper is not the end of the communication process. The findings need to reach the public who funds the research, benefits from it, and makes decisions based on it. A plain language summary is part of the publication, not an afterthought.

3. Science communication should be graded, not one-size-fits-all

The ACM paper's finding that "the general audience" does not exist should change how PLSs are produced. One summary is not enough. Graded summaries at multiple reading levels serve more people more effectively.

4. Trust in science depends on accessibility

The PNAS Nexus study found that AI-generated summaries improved not just understanding but trust. When people can understand what scientists found, they trust the scientists more. Readability is not just a communication metric — it is a trust metric.

What researchers should do now

• Generate a plain language summary for every paper. Use

ELI5 AI to produce four reading levels from your abstract. It takes seconds and costs nothing

• Publish the summaries alongside the paper. On your lab

website, in the preprint, in the journal's supplementary materials, and in press releases

• Match the summary to the audience. Send the adult level to

policy makers. Post the 10-year-old level on social media. Use the 5-year-old level for community outreach and schools

• Check accuracy. AI-generated summaries are more readable but

should be checked for factual accuracy by the researcher before publication. The PNAS Nexus study found accuracy was preserved, but researchers should always verify

• Track engagement. Compare engagement metrics for content with

and without plain language summaries. The data will show whether PLSs increase reach, sharing, and public engagement with your research

The bottom line

The PNAS Nexus study provides empirical evidence for what science communicators have long suspected: AI can write plain language summaries that are more understandable and more trusted than those written by the researchers themselves. This does not replace the researcher — the science is still theirs. But it does mean that the communication of that science can and should be delegated to tools that are better at it.

The researchers who adopt AI-assisted plain language summaries will see their work reach more people, be understood by more audiences, and — the data shows — be trusted more. The researchers who do not will continue writing summaries that fewer people read and fewer still understand.

Try it: paste any research abstract, paper summary, or scientific text into ELI5 AI and get four reading levels side-by-side — no account required.