Publisher Elsevier alone received approximately 4.2 million manuscript submissions in 2025, of which 795,000 were published, while the remainder were rejected during the editorial process—many before ever reaching peer review.

The 2026 Future of Peer Review report found that editors now need to send out an average of five invitations before a single researcher agrees to review a manuscript.

The widespread adoption of generative AI by both authors and reviewers has sparked debate over whether—and how—the scholarly publishing system should incorporate the technology.

In an interview with Science Arena, physician and researcher Howard Bauchner, former editor in chief of the Journal of the American Medical Association (JAMA), advocates using AI as a pre-review screening tool.

“AI would evaluate the manuscript, and the editor would review that assessment before deciding how to proceed. They could reject the manuscript because the AI identified fundamental flaws, or return it to the authors for revision before sending it out for external review,” he suggests. 

The primary goal of AI is not simply to speed up the publication process, but to reduce the number of manuscripts that ultimately reach human reviewers, he says.

How AI-based editorial screening would work

Under this model, AI would be responsible for technical, verifiable, and repetitive tasks, including:

• Compliance with reporting guidelines; 
• Study registration; 
• Reference accuracy; 
• Detection of undeclared AI-generated text; 
• Image manipulation. 

The last of these—image manipulation—is particularly important in laboratory sciences and is rarely detected through conventional peer review.

Bauchner also points to what he calls ghost reviewers, who use AI even when journal policies prohibit it. Rather than allowing the practice to remain hidden and unregulated, he argues that AI use should be permitted, provided it is fully disclosed.

What AI can already do in editorial screening, according to Howard Bauchner

1. Compliance with reporting guidelines: Verifies that the manuscript meets the journal’s required reporting checklists. 

2. Study registration: Confirms that clinical trials and other eligible study designs were properly preregistered. 

3. Reference accuracy: Detects incorrect or nonexistent citations, including those fabricated by generative AI. 

4. Detection of undeclared AI-generated text: Flags passages that appear to have been written by language models without appropriate disclosure. 

5. Image manipulation: Identifies inappropriate alterations to scientific figures—a frequent oversight in human-only peer review.

The limits of artificial intelligence in manuscript evaluation

For now, AI cannot replace human reviewers in every aspect of manuscript assessment. Bauchner explains that AI is still unable to evaluate the broader scientific context of a study—how a manuscript fits within the ongoing research.

“And if you ask most reviewers, that’s exactly what they enjoy commenting on,” says the former JAMA editor in chief.

Jesús Mena-Chalco, a professor at the Federal University of ABC (UFABC) and a researcher in scientometrics—the field that examines the quantitative dimensions of science—identifies a similar limitation. “Determining whether a study is truly original and relevant is a key role of the human reviewer.”

This limitation is also reflected in the lack of critical judgment displayed by AI models. “Current models tend to be affirmative—they’re rarely critical,” Mena-Chalco explains.

The advantages AI offers in the editorial process are not without risks. Perhaps the most pressing is data confidentiality. Manuscripts contain unpublished findings that should not be absorbed into the language models used by reviewers—or by the authors themselves.

Even so, AI may help address a problem that it has itself helped to create: fabricated references generated by language models that invent inexistent citations.

Automatic reference verification—already envisioned as part of the proposed pre-review screening process—has therefore become increasingly important.

Bias: From training data to human reviewers

Like human reviewers, large language models (LLMs) reflect biases. In the case of AI, those biases stem from the datasets on which the models were trained.

Bauchner argues that AI systems could be instructed to disregard authors’ identities, helping eliminate biases associated with institutional affiliation, country of origin, or language.

Mena-Chalco adds that LLMs have one practical advantage over human reviewers: they do not become fatigued.

“A human reviewer who has to evaluate multiple manuscripts may ultimately be more biased than a computer that never gets tired.”

Both arguments, however, describe intended uses rather than guarantees. Biases embedded in training data persist regardless of the instructions the models receive.

Both researchers believe that openly integrating AI into the scholarly publishing ecosystem is only a matter of time. Mena-Chalco notes that many researchers already rely on these tools but remain reluctant to admit doing so.

“Many people feel guilty about using these systems because they worry their work will seem too easy. So where should the line be drawn?” he asks. 

For Mena-Chalco, the answer will come through normalization.

“In about 10 years, it will no longer be necessary to disclose AI use,” he predicts.

There are already signs that this transition is underway. The Public Library of Science (PLOS), one of the world’s largest nonprofit open-access scientific publishers, has implemented automated research integrity screening.

Data presented by the organization show that desk rejections increased from 13% in 2021 to 40% in 2025.

A study conducted by the editorial team of the journal Organization Science provides the first full-corpus empirical analysis of this phenomenon within a scientific journal.

Between 2021 and 2026, submissions to the journal increased by 42% following the launch of ChatGPT in November 2022. The rise was driven primarily by manuscripts produced with extensive AI assistance rather than by organic growth in the field.

Manuscripts containing more than 30% AI-generated content—as measured by the Pangram detection tool—were up to 30 percentage points more likely to be rejected during initial editorial screening. Over the same period, writing quality, measured using the Flesch Reading Ease score, declined by 1.28 standard deviations.

Review reports followed a similar pattern. More than 30% now show some degree of AI assistance, with reports that are more difficult to read and narrower in scope, placing greater emphasis on theory and less on data.

For researchers in countries where English is not the primary working language, however, this shift could produce tangible benefits well before that timeline.

What remains uncertain are the editorial policies that will govern this transition—and how they will affect the integrity of the peer review process that underpins scientific research.

O post Can AI help overstretched peer reviewers? apareceu primeiro em Science Arena.

The book I Told You So: Scientists Who Were Ridiculed, Exiled, and Imprisoned for Being Right, published in 2024 by St. Martin’s Press and written by Matt Kaplan, science correspondent for The Economist and former paleontology researcher at the University of California, Berkeley (USA). 

The book was based on dozens of interviews with active scientists and on the analysis of historical cases spanning four centuries of medicine, biology, and paleontology.

The text reconstructs what should be an elementary notion in the scientific world: that new data can overturn scientific consensus, and demonstrates case by case why this rarely happens without some personal cost to those who present the data.

WHY IS THIS BOOK RELEVANT?

I Told You So has historical ambition. Kaplan not only reports that scientists were persecuted: he diagnoses the mechanism. The central argument is that resistance to new ideas is not an occasional flaw in the scientific system, but a structural consequence of how paradigms are constructed, defended, and funded.

Semmelweis was dismissed, forcibly committed to an asylum, and died without ever seeing his hypothesis accepted. What Kaplan shows is that the resistance to his work did not stem from ignorance: it came from physicians who fully understood what he was saying but could not accept that they themselves were the vectors of the disease.

The book also documents the case of Mary Schweitzer, a paleontologist who, in the 1990s, identified structures consistent with red blood cells in an 80-million-year-old fossilized Tyrannosaurus rex bone.

The discovery contradicted the claim that soft tissues could not survive fossilization. Schweitzer was attacked by colleagues for two decades. Only in 2017, after publishing results obtained under tightly controlled conditions, did the field begin to reconsider its position. She is now considered the founder of molecular paleontology.

The book also documents how institutional biases are perpetuated: funding denied to researchers who challenge the consensus, journals rejecting articles for personal reasons, and advisors instilling in students the same certainties that they themselves inherited. 

WHAT MAKES THIS BOOK A MUST-READ?

There are two standout moments in the book: the first is the opening scene. In 2012, Kaplan witnessed a group of senior researchers surround PhD student Alison Moyer—one of Schweitzer’s students—and verbally attack her poster at a vertebrate paleontology conference. 

She had questioned whether the structures identified as melanosomes in fossil feathers might actually be bacteria—a scientifically grounded idea. The field responded with fury. The scene is unsettling because Kaplan was there, with his press badge tucked inside his pocket, and he did not look away.

The second is the story of Alexander Gordon, a Scottish physician who, in 1795, almost half a century before Semmelweis, published detailed evidence that midwives and physicians were transmitting childbed fever from patient to patient. Gordon created tables listing names, dates, and chains of transmission. 

He then published, along with the data, the names of the professionals he had identified as unwitting vectors. The population of Aberdeen turned against him. He fled to the Navy and died of tuberculosis at sea aged 47, without any of his conclusions being taken seriously.

Kaplan concludes that understanding how science has failed in the past is the only way to make it fail less in the future, and that the gap between discovery and acceptance has tangible consequences, measured in lives that could have been saved.

O post When science turns on those who get it right apareceu primeiro em Science Arena.

The main risk lies in what is known as “lazy prompting”—creation of imprecise prompts that delegate critical thinking to the machine. In literature review, the most common manifestation of this problem is uploading large volumes of PDF files at once without providing the model with sufficient context for accurate processing.

In an interview with Science Arena, neurologist João Brainer, a clinical researcher at Einstein Hospital Israelita and professor at the Federal University of São Paulo (UNIFESP), warned about the risks of this approach.

Hallucination and loss of context

The problem stems from the architecture of large language models (LLMs). As the volume of information included in a prompt increases, the system’s accuracy tends to decline.

Bulk document processing can compromise the analytical consistency of AI models. The result is often a low-reliability analysis characterized by three major flaws:

• Mixing authors and references from different studies;
• Loss of essential information diluted across large volumes of text;
• Generic conclusions.

“If you simply ask the system to take what already exists and replicate it, you’ll end up with a lot of nonspecific texts. In science, you need to understand and identify where the problem lies, understand the reality, talk to other people, and read similar articles,” explains Brainer.

According to the researcher, this lack of depth undermines the rigor required by cutting-edge science.

There is also a frequently overlooked technical limitation: most tools limit direct uploads to about ten articles at a time. 

Attempting to circumvent this restriction through general-purpose platforms or overly broad contexts can overwhelm the model’s ability to process information accurately.

An incremental approach as a solution

To preserve the quality of a literature review, Brainer recommends an incremental strategy: providing the AI with small batches of articles, for example, five PDFs at a time.

The proposed workflow is outlined below. When applied systematically, it enables cross-checking and helps reduce both model errors and inconsistencies in human analysis.

In his view, the researcher’s critical thinking and judgment remain essential for ensuring ethics, transparency, and meaningful scientific contributions.

How to process scientific articles with AI in incremental steps

Watch the full livestream with João Brainer below:

O post Why feeding AI with documents is not enough for high-quality scientific review apareceu primeiro em Science Arena.

The update, part of the European Commission’s Strategy for AI in Science, was published on the institution’s official research and jobs portal.

With this change, the ability to leverage AI is no longer just a technical advantage—it is now part of the skillset expected of all scientists, from early-career researchers to scientific and institutional leaders.

Institutions and funding agencies adopting the ResearchComp framework (which is voluntary) may apply it to activities ranging from the selection of fellowship recipients to the evaluation of research groups.

Four levels of proficiency

ResearchComp lists four progressive levels of AI competence: foundational, intermediate, advanced, and expert. Within these levels, skills range from a conceptual understanding of the technology’s applications and limitations to the ability to formulate institutional policies governing its use.

ResearchComp’s four levels of AI competence

Ethics as a guiding principle

Ethical issues surrounding the use of AI are relevant across all four proficiency levels. At a foundational level, researchers are expected to recognize the ethical implications and challenges of using the technology; at expert level, they are challenged with creating and applying guidelines to ensure responsible and equitable practices. Progression through the framework therefore requires not only technical expertise, but also institutional responsibility.

The framework’s 39 competences include data management, open science, citizen science, and scientific integrity.

Any researcher, including Brazilians, can access ResearchComp’s online self-assessment tool, also developed by the European Commission. 

The tool allows users to identify their own level in each of the seven areas of competence by answering a multiple-choice questionnaire. The results can then be used to identify training gaps and guide professional development.

ResearchComp and the self-assessment tool are available free of charge on the European Commission’s website: ResearchComp — European Competence Framework for Researchers. 

A direct link to the tool is also available on the website.

O post AI shifts from competitive edge to key competence for researchers apareceu primeiro em Science Arena.

The rapid growth of AI tools and how to use them with methodological rigor in scientific work were the focus of a virtual meeting hosted by Science Arena on May 28 to share practical tips on the ethical use of AI in producing scientific articles.

During the livestream, Brainer explained a fundamental premise: the output from AI tools reflects the quality of the human input. Providing vague instructions and data will result in a low-quality paper.

The secret lies in prompt engineering and transparency.

The guide below offers practical tips for writing scientific articles based on the experiences shared by João Brainer, who teaches a course on AI in science at Einstein. 

Watch the full livestream about the use of AI in scientific writing:

Start with a compelling introduction

The introduction should contextualize the problem and capture the reader’s interest, explains Brainer. To avoid generic text, provide the AI with your preliminary data and key references.

Suggested prompt: “Write an introduction of four to five paragraphs. In the first, introduce the problem of [insert your problem here], focus on [insert your insights], and use the following references [list references]. The objective of this text is [state your purpose]. Use appropriate, objective, and clear language.”

Methodology and hallucination-free results

For Brainer, methodology is the heart of the study. “You need to follow certain steps, to know what to write at each step and how, including ethical considerations and the statistical analysis plan.” A tip to ensure scientific rigor is to use standard guidelines for your research methods. 

Suggested prompt: Upload the guideline and supporting paper to your study, and ask the tool to help outline the steps of your methodology. “Ask me for detailed information on each item based on this methodology in the context of my study.” You can also use your data to ask the AI to create a statistical analysis plan tailored to the objectives of your paper.

Brainer emphasizes that to ensure AI tools check properly for inconsistencies, researchers should use emphatic terms to trigger a more in-depth review. 

Suggested prompt: “Make sure that what I have written is consistent with the type of study I am conducting; identify any ambiguities or inconsistencies.”

Do not be afraid to create three, four, or five prompts on the same subtopic; multiple checks will improve the quality of your paper.

RELATED: How is AI impacting science careers?

Intelligent organization of references and data

Verifying sources is the cornerstone of scientific legitimacy. Specialized tools such as Consensus, Elicit, SciSpace, and Perplexity can help survey the literature and manage references. 

Visual elements also require precise descriptions.

Suggested prompt for figure/graph captions: “Create a detailed explanatory caption for this dataset [insert graph data], highlighting the variables analyzed.”

Transparency and open science

For a manuscript to be considered credible by respected academic journals, Brainer argues that authors should explicitly indicate where they used AI and what for, either in the methodology section or the acknowledgments.

RELATED: “The risk is not AI itself, but delegating intellectual work to machines,” researcher says.

Data auditing is the best way for scientists to protect themselves against accusations of plagiarism or fraud. Brainer therefore also recommends depositing raw data and the prompts used on public platforms, such as Mendeley.

AI can save time during the writing process, but critical judgment, ethical responsibility, and final validation remain unequivocally human.

O post Scientific writing: a step-by-step guide to the ethical use of AI  apareceu primeiro em Science Arena.

According to the authors, the use of advanced language models can speed up scientific output and make it more structured and efficient. However, they are yet to achieve full autonomy, and cannot replace the scientist. 

The article points out that the human eye is key to guaranteeing critical interpretation of data and generation of original knowledge.

This discussion takes on particular relevance given the movement by state universities in São Paulo—the University of São Paulo (USP), University of Campinas (UNICAMP), and São Paulo State University (UNESP)—to regulate the use of AI in research. 

The guidelines highlight the need for methodological transparency and require that researchers detail not only use of the tool, but also how it is applied.

Traditional process under pressure from tech

Literature review is a structuring stage of scientific output, in which the researcher sets out what they know about the subject, identifies knowledge gaps, and guides new investigations. 

In the specific case of narrative reviews, there is more room for interpretation, synthesis, and construction of the case to be stated. It is at this point that AI comes into being as a support tool. 

To examine this scenario, the authors conducted a narrative review based on searches in the PubMed and Google Scholar databases, combining terms related to scientific writing, medicine, and AI. The strategy also included semantic exploration using digital tools, and secondary reference analysis.

Unlike systematic reviews, the selection process prioritized thematic relevance, enabling a comprehensive view of the ecosystem of available tools, albeit with lesser methodological reproducibility. 

Partially automated production chain

At the outset of the review process, the language models demonstrated their capacity to suggest research gaps, relying on both recent trend analysis and summarization of discussions in published scientific articles.

In the structuring phase, tools such as ChatGPT enable skeleton scripts to be drafted for the article, reducing initial planning work and contributing to logical coherence in the text.

Platforms such as Research Rabbit and Semantic Scholar can identify connections invisible to conventional searches, providing a more contextualized browsing experience.

In terms of organization, reference management tools integrated into AI resources allow not only for article storage, but also for structured annotation, classification, and recovery of information. 

This is a crucial aspect for narrative reviews, which rely on the cross-referencing of multiple sources.

Finally, AI can act as a multifunctional assistant during the writing process—producing drafts, adjusting academic style, improving clarity, and even simulating critical reviews. 

The study points out that this integration can significantly save time spent on writing while enhancing formal manuscript quality.

Structural limits

Although tech advancement is praiseworthy, the authors emphasize that automation can also present relevant risks, notably the production of technically correct but intellectually superficial texts, without significant analytical contribution. 

The lack of original interpretations, generation of incorrect information (known as “hallucinations”), and the inaccuracy of references demand careful verification by researchers. 

The study concluded that the way forward is not the total automation of science, but the consolidation of a new knowledge production regime: hybrid, streamlined, and dependent upon qualified human supervision. 

AI tools to help researchers 

ChatGPT

Identifies research gaps, prepares drafts, rewrites article sections (e.g. introduction and discussion), suggests adjustments to register, and improves readability. It can also evaluate articles in a similar manner to peer review.

Research Rabbit

Identifies articles, authors, and related topics through analysis of citations and coauthorship. Facilitates literature browsing with a visual interface, and integrates with Zotero to enable citation management.

Semantic Scholar

Uses natural language processing and machine learning to research articles, understand content in context, and rank articles by relevance.

Elicit

Uses advanced research strategies to locate and compile relevant articles. Provides an exclusive interface to efficiently classify bibliographical research results.

Zotero

Free reference management system (RMS) to collect, organize, cite, and share research sources. Includes browser integration, PDF annotation, markup, citation, bibliographical management, and support.

Mendeley

Web-based reference management system (RMS) similar to Zotero, with integration to Microsoft Word online for management of citations and references in the cloud.

TinyWow

Offers PDF summary tools for quick article analysis and relevance assessment.

Scribbr

Offers PDF summaries to help manage large volumes of literature, providing concise article summaries.

Quillbot

Summarizes scientific articles for initial screening to determine their relevance and importance for narrative reviews.

Grammarly

AI-driven communication assistant to help with spelling, grammar, and style, ensuring clarity and accuracy. Can be integrated with text processes to provide real-time assistance.

Jenni.ai

Combines spelling and grammar correction, text enhancement, and assistance in compiling bibliographical citations. Improves writing by suggesting phrase development and integrating support resources.

Yomu

Integrates writing assistance resources, with text improvement and help with bibliographical references. Supports the writing process with resources similar to those of Jenni.ai.

Claude

Advanced conversational AI to assist with writing, summarizing, and brainstorming scientific texts. Recognized for its focus on security and treatment of extensive materials.

Perplexity

AI-based search engine providing concise answers and sources; useful for obtaining updated references during literature reviews.

DeepSeek Chat

Large language model (LLM) designed for in-depth reasoning, code-based tasks, scientific questions and answers, problem solving, and drafting of technical content.

Llama/Mistral via DuckDuckGo/Groq

Chat platform with DuckDuckGo AI offers access to Llama and Mistral via Groq – useful for quick scientific consultations and summaries with short response times. 

How to use AI tools in scientific production?

Reference

Del Giglio A, Costa MU. Utilizing Artificial Intelligence to create narrative literature reviews. einstein (São Paulo). 2026;24:eRW1165. https://dx.doi.org/10.31744/einstein_journal/2026RW1165

O post ChatGPT and other AI redefine scientific literature review apareceu primeiro em Science Arena.

Rejecting both uncritical enthusiasm and technological panic, she sees AI not as a threat to academia, but as a mirror reflecting long-standing issues: gaps in methodological training, unresolved questions of authorship, and a culture that still treats scientific integrity as a bureaucratic protocol rather than a core value.

The arrival of AI tools in everyday research has prompted organizations such as the Brazilian National Council for Scientific & Technological Development (CNPq) to publish unprecedented regulatory guidelines—a sign that the technology is already part of academic practice and demands clear governance standards.

For Scussel, the greatest risk is not the technology itself, but researchers who delegate essentially intellectual tasks to machines.

In this interview with Science Arena, she explains how to select AI tools critically, why cross-validation is nonnegotiable, and what separates responsible AI use from careless application.

Science Arena — What are the biggest practical challenges researchers face when using AI without compromising the quality of their work?

Fernanda Scussel – AI has highlighted fundamental problems in academia that already existed but were neglected, such as questions of authorship, authenticity, and weaknesses in how research methodology is taught in Brazil.

The challenge is not only using the tool ethically, but also examining academic culture as a whole.

The focus needs to shift away from the technology itself and back to the research process. In other words, AI should be viewed as a tool for specific stages of research—not as a magical solution capable of writing dissertations or reading papers on its own. 

Many researchers feel overwhelmed by the sheer number of tools available. What criteria do you use to determine what is actually useful?

The main criterion is suitability for the task. Researchers should first define the problem they need to solve and only then choose the technology. Another important factor is testing: I recommend exploring no more than three or four tools and focusing on the one you feel most comfortable using.

Good results come from making the most of a specific tool, because consistent use enables machine learning and allows the model to better understand the user’s needs over time.

What validation strategies are paramount to ensure that a tool is trustworthy?

We need to understand that AI platforms are companies, and they contain what I call a “flattering element” designed to encourage engagement—they want to please you.

That means researchers must develop their own validation criteria. One crucial strategy is cross-checking, to verify whether what the AI claims actually matches what the author of a paper wrote.

Another critical error is delegating essential tasks, such as reading, to the machine. Use AI to accelerate the process—for example, to help navigate a complex text—but never relinquish intellectual responsibility for the content.

How do you stay up to date and distinguish genuine innovation from mere hype?

You need to be very careful with hype because it creates an environment of anxiety. If you are already using a tool that delivers reliable and satisfactory results, there is no need to migrate to the “tool of the week” simply because of external pressure.

Maintaining focus on a single tool also strengthens the machine-learning process I mentioned earlier and reduces the anxiety of constantly chasing the next technique.

As a professor, I test many different options, but for most researchers, the ideal approach is to tune out the noise and concentrate on what works for their own process.

What are the most common mistakes people make when using these tools?

The first is not knowing how to create an effective prompt. You need to know how to “direct” the AI—that is, provide context and establish limits.

The second is the lack of interaction: many people simply copy and paste the first response they receive, which results in generic writing and increases the risk of plagiarism.

Finally, there is a lack of technical understanding, such as ignoring the “context window,” which can lead to hallucinations when too much information is inserted at once.

How to structure an effective prompt for scientific research

What are the greatest opportunities AI offers today for democratizing Brazilian science?

AI has practically eliminated the English-language barrier, for example, and to me that has been a major breakthrough. Researchers not fluent in English can now understand complex texts and participate in classes on a more equal footing.

In your view, what are the next steps for this technology?

We now need to focus on training educators so they are better prepared to incorporate these tools into their own practices and teach students how to use them correctly.

We need to discuss all of this responsibly and without taboo. AI is here to stay, and therefore it deserves serious attention and a thoughtful approach.

O post “The risk isn’t AI, but delegating intellectual work to the machine,” researcher says apareceu primeiro em Science Arena.

According to Brainer, the biggest challenge faced by scientists today is strategic: understanding which tool is best suited to each stage of the research process and how to get the most out of them without compromising scientific integrity.

To help researchers navigate this rapidly expanding ecosystem, Science Arena compiled a list of the main tools recommended by Brainer, including details of their main functions, strengths, and weaknesses. See below:

CONSENSUS

Link: https://consensus.app/ 

Key function: Literature searches, focusing on direct, accurate, evidence-based results.

Strengths: The platform has partnered with major scientific publishers (such as Wiley, Sage, ACS, and others). As a result, it can extract data from papers in their entirety, rather than just the abstracts. It also provides the exact page number and a link to the source article, significantly reducing the risk of hallucinations or data manipulation.

Weakness: Because it depends on publishing partnerships, Consensus’s search is limited to specific databases, potentially excluding important titles.

CORE

Link: https://core.ac.uk/

Key function: Indexing open access scientific literature. 

Strengths: CORE is a free platform that aggregates data from repositories around the world. It is particularly useful for conducting broad literature searches, focusing on the latest developments in open-access publications.

Weakness: By design, CORE does not search paid-access journals or articles protected by paywalls.

OPEN EVIDENCE

Link: https://www.openevidence.com/ 

Key function: Searching for high-impact medical literature.

Strengths: The tool is extremely thorough and reliable for the medical and biomedical fields, targeting the databases of established journals, such as The New England Journal of Medicine (NEJM) and The Journal of the American Medical Association (JAMA). 

Weakness: By focusing on specialized databases, the scope of the results is limited, potentially excluding important journals or fields.

SCISPACE

Link: https://scispace.com/ 

Key function: Supporting academic writing with citation optimization and reference management.

Strengths: Authors can select a section of their text and ask the tool for a source within its database that supports the claim. The platform has a direct interface with Zotero and Mendeley, two popular reference management tools, and can automatically format tables and citations in more than 2,600 styles, including Vancouver and AMA. It can also be used to adjust the tone of a text, to make it more persuasive, pragmatic, or conversational, for example.

Weakness: Over-reliance on rewriting features can weaken the text’s originality and the author’s voice.

PERPLEXITY AI

Link: https://www.perplexity.ai/ 

Key function: Connecting ideas and cross-referencing data from different scientific articles.

Strengths: Perplexity performs advanced methodological correlations by cross-referencing author data and multiple articles, quickly creating complex theoretical overviews.

Weakness: João Brainer issues a strong warning about data privacy. The platform’s integrated browser collects user reading and browsing data in real time. For scientists working with patents, business ideas, or unpublished theses, this can create risks related to information leaks and loss of intellectual property before publication.

Popular Large Language Models (LLMs)

Key function: Preliminary screening, organizing ideas, and refining manuscripts.

Strengths: Each of the most popular large language models (LLMs) offers its own advantage in the research process: Brainer highlights Claude as the most rigorous and refined model for dealing with the density of purely academic texts; Gemini excels at searching for references online and providing links for fact-checking; and ChatGPT is especially useful for identifying trends and gaps in the literature.

Weaknesses: Because they are designed for general purposes, LLMs function based on linguistic probability. This means that when asked to process data without contextual limitations, there is a greater risk of hallucinations. They require detailed and exhaustive input; poorly worded or superficial prompts tend to lead to inaccurate and irrelevant responses.

Number one tip: combine tools and build a PDF database

Brainer’s main practical recommendation is not to expect any single tool to solve every aspect of a research project. It takes some effort to achieve consistent results. 

“You can combine the broad searches of PubMed AI CORE with the more in-depth refinement provided by Consensus and the reference management of SciSpace,” says the expert.

“You should also create your own archive of PDFs you have collected,” advises the researcher.

Brainer predicts that in response to the current flood of AI-generated papers, major journals will begin requiring authors to submit the original source files they consulted, both for auditing purposes and to ensure scientific integrity.

Watch the full discussion with João Brainer below:

O post AI guide for science: Key tools for producing novel research papers apareceu primeiro em Science Arena.

In an interview with Science Arena, Glauco Arbix, scientific coordinator of the Responsible AI Chair at the University of São Paulo (USP), highlighted that science itself already exists within a context of instability. This is because society is constantly facing turbulent times, from geopolitical to economic issues. 

When it comes to technology, he noted that “all new technologies, especially transformative ones, sometimes carry issues as big as or even bigger than expectations.” The sociologist from USP added: “You hope that life will improve—at the same time, the adoption of technology can generate problems.” 

According to Arbix, artificial intelligence optimizes research processes, but brings with it issues that require extra care from scientists, such as the risks associated with accelerated production of scientific articles without critical scrutiny. 

AI and bias in scientific research

Research bias is a problem intensified by AI. According to Arbix, this occurs because the data obtained through the tool are mainly produced and stored by companies based in English-speaking countries. 

“You can’t prepare to work with Africa or South America,” he says. “Research becomes more homogenized and, therefore, less innovative,” he added. 

Another point raised is the need to pay attention to how questions are posed to these platforms because, depending on the query, the tool will respond in the way the researcher wants and not based on evidence.

Inequality in access to artificial intelligence

Although AI is increasingly improving, not all countries have equal access to these tools. The USA and China are both way ahead of the others when it comes to AI development. 

According to Arbix, the second group of countries with greater access to this technology includes nations such as Germany, France, and the UK. 

“We have major deficiencies in data centers, in workforce qualification, and in access to the advances themselves,” he warned. “If we do not have widespread AI literacy programs to enable people to demystify and use it, if we do not change the way education is being delivered, we will reproduce inequality.”

To read the full content on the uncertainties generated by the use of artificial intelligence in science, see the interview in this feature from Science Arena.

O post Artificial intelligence: how does it generate uncertainty in science? See what the expert says apareceu primeiro em Science Arena.

The most relevant finding of Mena-Chalco’s analysis, however, is not the volume of articles published, but the profile of their authors. PhDs account for around 35% of the journal’s authors, a higher percentage than masters (17%) and bachelors (12%). 

Furthermore, 82% of the published manuscripts had at least one PhD among their authors. “They are not inexperienced,” says the professor at UFABC.

How to verify whether a journal is predatory

Accurately defining what constitutes a predatory journal is not a simple task. One of the first responses from the scientific community was to create “blocklists,” but the approach has its limitations. “What are the criteria, and who defines them? In some of these lists, you might find a journal from an emerging country that may not be the best publication or not have the best website, but it is not trying to deceive anyone,” says Lorraine Estelle, a collaborator with the Think.Check.Submit initiative.

The initiative chose a different approach: a guide with questions to help researchers evaluate a journal before submitting a manuscript. If the answer to most of the questions is yes, the likelihood of it being a predatory journal is low.

Think.Check.Submit CHECKLIST 

Author profile contradicts predominant perception

The idea that predatory journals mainly attract novice researchers is backed by scientific literature, but is has been challenged by more recent studies. 

A 2015 article analyzing 324 manuscripts in seven journals considered predatory concluded that most of the authors were ”young and inexperienced researchers from developing countries.” Later analyses reached different conclusions.

Another study, which examined over 2 million Brazilian publications between 2000 and 2015 and cross-referenced data from the Lattes platform with information from bibliometric lists and indexes, such as the Directory of Open Access Journals (DOAJ), the Journal Citation Reports (JCR), and the Scimago Journal & Country Rank (SJR), identified significant participation by experienced scientists in predatory journals. 

The conclusion: the longer the interval between the PhD defense and the publication date, the greater the probability that the article was published in a fraudulent journal.

Marcelo Perlin, a professor at the Federal University of Rio Grande do Sul (UFRGS) and coauthor of this latest study, acknowledges that it is still difficult to identify the exact profile of authors who publish in predatory journals, but highlights pressure and incentives as central factors.

Mena-Chalco, of UFABC, suggests that the pressure begins as early as during graduate studies. “In many graduate programs, students need to demonstrate that they have published in a journal before being able to defend their dissertation,” he explains. “In general, publishing in prestigious journals takes time and rarely happens on the first attempt. Therefore, faced with this pressure and the ease of publishing in predatory journals, individuals submit manuscripts to this type of publication.”

Risks to science and gaps in regulation

The proportion of predatory publications in Brazil is still low. In the study by Perlin, it did not reach 1% of total national publications. But Fhillipe Campos, a researcher at the Brazilian Institute of Information in Science and Technology (IBICT), warns that the problem should not be measured solely by volume. 

“For example, the lack of peer review [an essential stage in the evaluation of manuscripts, waived by most predatory journals] represents a direct risk to the integrity of scientific production,” he warns. 

“It may be that just a few predatory journals handle a large number of articles precisely because of the ease of accepting submissions. This could be a problem,” adds André Appel, also from IBICT.

The sector is essentially self-regulated by the scientific community, according to Campos. In Brazil, Qualis—the journal evaluation system maintained by the Brazilian Federal Agency for Support and Evaluation of Graduate Education (CAPES)—was intended to function as a quality filter. However, the system has shown flaws. 

The historical Qualis model prioritized the volume of publications and used bibliometric databases, such as Scopus and Google Scholar to measure impact. Predatory journals that publish a lot of articles could therefore accumulate favorable metrics.

The journal analyzed by Mena-Chalco exemplifies this: according to information published on the journal’s website, it received an A2 classification, the second-highest tier in Qualis, in the evaluation for the 2017–2020 four-year cycle.

Consulted for the report, CAPES acknowledged that the historical model was based on past practices related to each evaluation cycle, and admitted to the growing difficulty of separating legitimate journals from predatory ones, given that “journals from the same publishing group behave very differently in terms of editorial practices.”

Since 2025, the model has been revised. The new Qualis focuses on the quality of articles, rather than journals. Publications in journals with predatory characteristics, such as the promise of rapid publication, lack of information on peer review, and aggressive marketing, are now automatically classified in category C, the lowest tier.

O post Experienced researchers frequent among authors in predatory journals apareceu primeiro em Science Arena.