Free resource · Word + Excel kit
AI Researcher Kit: 24 prompts to progress through 4 levels
The best use cases in research, from state of the art to commercialization, with ready-to-copy prompts and seven Excel tools. A step-by-step progression, without compromising scientific rigor or source traceability.
The AI Researchers Kit, in brief
- What it is
- A self-sufficient skills development kit for research teams: 24 use cases, each with its prompt, organized into four levels of mastery, and a binder of seven tools.
- For who
- Researchers, research engineers and doctoral students, R&D teams from research organizations, agencies and large industrial groups, technology transfer and valorization teams (SATT), and team leaders who want to lead it collectively.
- Format
- 74-page Word document: user guide, fundamentals, usage framework, 24 fact sheets, 4 progress challenges, 4 current affairs files, a section for facilitating the kit in teams, 6 appendices. 7-tab Excel workbook.
- What it contains
- Copy prompts and follow-up prompts, instructions for six assistants, missions for six agents and their safeguards, five-step reference verification protocol, AI usage log for transparency declaration, three fires for unpublished data and pre-patent work, reference register, monitoring plan, pre-submission review grid.
- Tools
- Copilot Chat and Microsoft 365 Copilot (often called "Copilot Premium"), ChatGPT, Claude, Gemini and Mistral, in their enterprise versions, your organization's internal assistant, and benchmarks on specialized bibliographic research tools.
- What is it based on?
- European guidelines on the responsible use of generative AI in research, the Wiley, Elsevier and Nature surveys on researchers and AI, publisher and funder policies, the AI Act, and AI training sessions conducted with research teams.
- Price
- Free, in exchange for a professional email address. No unsolicited emails.
- Update
- Version 1 of October 31, 2026.
- What it doesn't replace
- Neither your organization's rules of scientific integrity, nor the guidelines of your journals and funders, nor the supervised practice on your own subjects.
- Go Further
- AI training for research teams, AI innovator kit, responsible AI kit.
Researchers were among the first to experiment with generative AI, often working alone and cautiously, because their profession relies on evidence. Many use it to rephrase an abstract or translate a paragraph, but few use it to structure a literature review, challenge a protocol, or establish reliable monitoring. What's lacking isn't the desire, but a methodology that safeguards rigor: verifying each reference, distinguishing fact from suggestion, and knowing what cannot be entrusted to the tool.
This page presents what recent surveys say about researchers and AI, the four-level progression we use in training, the 24 selected use cases, and a sample worksheet. The complete kit, in Word format with its accompanying Excel workbook, can be downloaded for free at the bottom of the page.
1. Researchers and AI: what recent surveys show
Major surveys conducted among researchers, notably by Wiley, Elsevier, and the journal Nature in 2025, converge on the same point: the use of generative AI is progressing rapidly across all disciplines, but trust remains limited, and researchers are calling for clear rules, particularly regarding usage declarations and the reliability of references. Precise figures and their sources are included in the toolkit.
On the institutional side, the European Commission updated its guidelines on the responsible use of generative AI in research in 2026. These guidelines reiterate that researchers remain responsible for their publications, must declare any substantial use of AI, must not entrust unpublished work or manuscripts received for review to external tools, and must be wary of hidden instructions for AI within documents.
Three consequences for a research team: learn about their own subjects rather than generic examples; make checking references a reflex, because an invented reference is the first risk of the profession; and keep a usage log from the start, which makes the transparency declaration easy when publishing.
2. Four levels, saving time by delegating
Researchers' AI resources are almost all lists of tools or prompts, without any order of progression or verification method. This kit organizes use cases into four levels, each with its learning outcomes and a milestone challenge that verifies mastery of the level.
| Level | What you do | What you learn | Example of a record |
|---|---|---|---|
| 1. Save time | Short prompts, without sensitive data, a result in five minutes. | Formulate a request in four parts, demand "unknown" rather than an invention. | Constructing a bibliographic search equation |
| 2. Structure | Comprehensive prompts, on your articles, your protocols, your datasets. | Separate fact, inference, and suggestion; check each cited reference. | Map nearby teams and projects |
| 3. Customize and improve reliability | Permanent assistants on your corpus: Project, Gem, Copilot agent. | Organize the context once and for all, give the AI your sources and your rules. | State-of-the-art assistant on your corpus |
| 4. Delegate | Planned agents, thorough research, reproducible analysis. | To entrust monitoring or analysis, to set safeguards, to review outputs. | Publications and patent monitoring officer |
Level 3 is the one most teams skip, going directly from ad-hoc use to agents. Yet it's this level that makes AI reliable: an assistant that only works on your corpus, with the instruction to cite the page and write "unknown," invents far less than a generalist assistant.
3. The 24 use cases, family by family
Each row represents a family of situations, each column a level. You progress through the columns; you select from the rows, depending on the week's work. Each level covers the entire profession: no need to wait until the end of the kit to tackle the real issues.
The details of each case follow: the situation, the realistic payoff, the point to check before using the result, and what you obtain. The complete prompt, or the assistant's instructions and the agent's mission, is in the kit.
Win time
Constructing a bibliographic search equation
- You're starting out on a topic related to your own, or you need to update a state-of-the-art review. You know the concepts, but not necessarily all the synonyms, the controlled terms (MeSH, thesaurus), or the syntax of each database. As a result, you miss out on articles or end up with thousands of irrelevant ones.
- Savings: 15 to 30 minutes on constructing an equation, more if you change disciplines.
- To verify: The proposed thesaurus descriptors do exist in the database (check in the database search tool).
Deliverable: A commented Boolean equation, declined for two bases, with the concepts, their synonyms and the terms to exclude, ready to test.
Summarize an article to decide whether to read it
- Your online monitoring brings you more articles each week than you can possibly read. The author's summary is often too enticing or too dense to make a quick decision. You want to know in two minutes whether the article is worth reading in full.
- Savings: 5 to 10 minutes per article sorted, or one to two hours per week with sustained monitoring.
- To verify: The key figure is indeed in the article, with the same unit and the same meaning.
Deliverable: A half-page sorting sheet: question, method, main result, limitations noted by the authors, and a reasoned opinion on the relevance to your topic.
Generate alternative hypotheses
- You have a working hypothesis, and you risk only seeing the data that confirms it. Before launching an experiment or interpreting a result, you want to force yourself to consider other explanations, as a critical colleague would.
- Gain: 20 to 30 minutes of brainstorming, and above all, ideas you might not have considered on your own.
- To verify: Each explanation is physically or biologically plausible in your field.
Deliverable: A list of 5 to 8 alternative explanations classified by type (artifact, bias, confusion, competing mechanism), with for each one an observation that would allow it to be distinguished.
Explain a statistical result clearly
- You need to explain a confidence interval, a p-value, an odds ratio, or a mixed model to a colleague from another discipline, a doctoral student, or a committee. Or you are reading an article with statistics that are less familiar to you.
- Gain: 10 to 20 minutes per explanation, and less back and forth with your audience.
- To check: The definition of the p-value or interval is correct (not "probability that the hypothesis is true").
Deliverable: A clear explanation, including what the result allows us to say, what it does not allow us to say, and the most frequent misinterpretations.
Improving an abstract
- Your summary exceeds the word limit, mixes context and results, or fails to clearly state its contribution. Yet it is the most read part of the article, the part on which the editor and readers decide.
- Savings: 20 to 40 minutes per summary, more if you are writing in a language that is not your own.
- To be verified: No figures or statements have been added or changed.
Deliverable: A revised version that respects the requested structure and length, with a list of changes and points to be decided by you.
Simplify a result into three levels
- A published result must be presented to management, the communications department, an industry partner, or the general public. You know how to explain it to peers, but you have to find the right words for each audience, without betraying the science.
- Savings: 30 to 45 minutes for all three versions, then fewer back-and-forths with the communications department.
- To be verified: No version promises an application, a timeframe, or a benefit that the article does not demonstrate.
Deliverable: Three versions of the same result (general public, decision-maker, specialist in another discipline), each with its key message and limitations.
Structure
Comparative table with sourced information from a batch of items
- You have selected 5 to 15 articles on a topic. To write your literature review, you need to compare them: methods, populations, results, limitations. Doing this manually takes a day, and extraction tables are exactly the type of task where AI helps, provided you can check each cell.
- Profit: Half a day to a full day on a batch of 10 to 15 items, depending on the level of control.
- To check: Check at least 20% of the boxes against the articles, giving priority to numerical results and sample sizes.
Deliverable: A homogeneous extraction table, each cell referring to the page or section of the article, with empty cells marked "unknown" and a summary distinguishing facts and inferences.
Map nearby teams and projects
- You are preparing a collaborative project, a funding application, or a technology transfer strategy. You need to know who is working on what: academic teams, industrial laboratories, patent applicants. The information exists, but it is scattered across publications, team websites, and patent databases.
- Gain: Half a day to a full day on initial mapping, to be supplemented by specialized databases.
- To be verified: Each DOI and each patent number exists and corresponds to the actor cited (Espacenet, Google Patents, doi.org).
Deliverable: A map of the actors (teams, organizations, companies) with their representative work, each entry supported by a verifiable source, and a list of what remains to be confirmed.
Challenging a protocol (variables, biases, controls)
- Your protocol is written. Before submitting it to the ethics committee, the funding body, or launching the experiments, you want a critical review: poorly defined variables, uncontrolled biases, missing controls, statistical power. A colleague who is available would be ideal, but they don't always have the time.
- Gain: 1 to 2 hours of critical review, and no back-and-forth with the committee or the funder.
- To check: The remarks are based on your text, and not on an imagined "standard" protocol.
Deliverable: A structured review by risk category, with for each point the relevant section, the severity and a proposed correction, plus the questions that the protocol does not answer.
Explore a dataset and identify anomalies
- You receive a data file (instrument export, cohort database, survey results). Before any analysis, you must understand it: variables, types, missing values, duplicates, outliers, inconsistencies between columns. This is tedious and is where many errors originate.
- Savings: 1 to 3 hours on the first cleaning, depending on the size and condition of the file.
- To check: Rerun two or three of the findings yourself (number of missing values, duplicates) in your usual tool.
Deliverable: An exploration report: description of each variable, missing values, anomalies identified with the lines concerned, and a list of questions to ask the person who produced the data.
Prepare the response for the reviewers
- You have received the reviewers' reports on YOUR manuscript: about ten comments, some accurate, others debatable, sometimes contradictory. You must address everything, respond politely and precisely, and follow the changes in the text.
- Savings: 2 to 4 hours on organizing and writing responses, excluding new analyses.
- To be verified: Every comment from the reviewers has been addressed; none have been merged or omitted.
Deliverable: A tracking table for each comment, a proposed response for each comment, and a list of changes to be made to the manuscript, which you complete and validate.
Prepare a grant application (ANR, Horizon Europe)
- A call for proposals is relevant to your work. The call document is 30 pages long, and the form requires a specific structure, weighted evaluation criteria, and page limits. You must verify its suitability, develop a plan, and ensure nothing is omitted.
- Gain: Half a day to a full day on the analysis of the call and the plan, depending on the length of the texts.
- To be verified: The deadline, eligibility and page limits are checked on the official website of the funder (the texts change from one edition to another).
Deliverable: An analysis of the suitability of your idea to the call, a plan of the file aligned with the evaluation criteria, and a checklist of documents and constraints, each element referring to the text of the call.
Customize and make reliable
State-of-the-art assistant on your corpus
- Your team has compiled a reference library on its topic (50 to 200 articles). Every new doctoral or postdoctoral student must familiarize themselves with it, and they all ask the same questions. You want a teaching assistant who can answer questions based solely on this corpus, with references to the articles.
- Benefit: Several hours per newcomer spent on appropriating literature, and faster answers to recurring questions.
- To check: Entry test: ask a question outside the corpus and check that the assistant replies that they cannot answer.
Deliverable: An assistant who answers questions about the state of the art of the team by citing articles from the corpus (author, year, section), who signals questions outside the corpus and who never supplements with unverified external references.
Patent reading assistant (claims)
- Competitors' patents, potential partners' patents, documents cited in research reports: researchers are increasingly required to read patents. The language of claims is confusing, the scope depends on each word, and the distinction between independent and dependent claims is not intuitive.
- Savings: 30 minutes to 1 hour per patent on understanding the claims.
- To verify: The claims are quoted word for word (compare with the text published on Espacenet or Google Patents).
Deliverable: An assistant that breaks down the claims of a published patent into elements, explains the scope of each claim in plain language, and links each element to the description, without ever concluding on infringement or validity.
Methodological assistant with your analysis plan
- Your project has a validated, sometimes pre-recorded, protocol and statistical analysis plan. For months, the team asks itself questions: Can we add this variable? What should we do with the missing data? Is this analysis planned or exploratory? It's essential to maintain consistency with what has been decided.
- Benefit: Faster and more consistent decision-making throughout the project, and a more complete history of deviations.
- To check: The documents uploaded are indeed the latest validated versions (number and date in the instructions).
Deliverable: An assistant who knows your protocol and analysis plan, responds by referring to it, and clearly signals any discrepancy between a request and what was planned (confirmatory or exploratory analysis, deviation to be documented).
Analysis with code execution
- You want to go beyond exploration: produce descriptive statistics, graphs, and test a simple model on anonymized data. Conversational tools can now execute Python code. However, the analysis must be traceable and you must be able to repeat it without AI.
- Savings: 1 to 3 hours on a descriptive analysis or a simple model, depending on your programming skills.
- To verify: Rerun the final script in your own environment (R, Python, lab software) and get the same results.
Deliverable: An analysis assistant that always works by showing its code, announces its assumptions before calculating, checks the conditions for applying the tests, and delivers a reusable script with the results.
Targeted Review Standards Assistant
- Each journal has its own guidelines: length, structure, style of references, mandatory declarations (funding, conflicts of interest, data availability, use of AI), and figure formats. A manuscript rejected for non-compliance can waste weeks.
- Gain: 1 to 2 hours per submission on compliance, and fewer technical reviews.
- To check: The instructions uploaded are up to date (compare the date with the online page of the journal).
Deliverable: An assistant who is familiar with the author guidelines of your target journal and checks your manuscript section by section, flagging each deviation from the relevant rule.
Technology sheet for the SATT or the valorization service
- A technology developed in the laboratory is protected (patent filed) or published, and the Technology Transfer Office (SATT) or the technology transfer office must present it to industry. A one- or two-page fact sheet is required, outlining the problem solved, its advantages, its maturity, applications, and intellectual property. Researchers often describe the science; industry expects a tangible benefit.
- Gain: One to two hours per sheet on the first version, with more homogeneous sheets from one technology to another.
- To verify: Each advertised benefit is supported by a published or protected result, with its source.
Deliverable: An assistant that produces consistent technology sheets, in the format of your structure, from public or already protected documents, with a justified level of maturity and without any unsubstantiated promises.
Delegate
Guided literature review, checked references
- You must produce a substantial literature review (thesis chapter, expert report, project introduction). A thorough research assistant can scan dozens of sources in minutes. But a well-written 20-page report can contain inaccurate references or misquotations, invisible on a first reading.
- Gain: Several days of research and initial synthesis, but allow one to two days for verification for a complete report.
- To check: Apply the protocol in Appendix 3 to all the references you keep, starting with the ten structuring statements.
Deliverable: A comprehensive research report structured according to your plan, accompanied by a table of all references with DOI and verification status, and a list of claims to be checked as a priority.
Publications and patent monitoring officer
- You're following a specific topic, competing teams, and a few patent families. Alerts from bibliographic databases and patent offices arrive in a flood, and you don't have time to sort through them all. You want a curated and sourced weekly digest.
- Savings: 1 to 3 hours per week sorting alerts, depending on the volume.
- To check: During the first two weeks, compare the rating with the raw alerts: is anything important being missed?
Deliverable: A scheduled agent who produces a two-page weekly monitoring note: new publications and new patents published, classified by interest, each with its link and a sentence on what is changing.
Prepare the data management plan (DMP)
- Most funding bodies (ANR, Horizon Europe in particular) require a data management plan, with an initial version at the start of the project and subsequent updates. This plan must describe the data produced, formats, metadata, storage, security, sharing, and archiving, in accordance with FAIR principles and legal requirements.
- Gain: Half a day to a full day on the first version, mainly thanks to the collection of scattered information.
- To be verified: Each piece of information in the PGD refers to a document in the project, or is marked UNKNOWN.
Deliverable: A mission Cowork which gathers information scattered throughout the project documents, pre-fills the funder's PGD model, and delivers a draft with a list of questions to submit to the right people (DPO, CIO, data warehouse, partners).
Reproducible analysis pipeline (Claude Code, notebook)
- Your analyses reside in scattered scripts, copy-pasted files, and spreadsheets. Six months later, no one can reproduce Figure 3. Journals increasingly require code and data. You want a clean, versioned, documented pipeline that a colleague can run from start to finish.
- Gain: Several days of structuring and documenting the code, and above all, significant time saved with each review or request for a reviewer.
- To check: Review each script before running it on real data: the logic corresponds to the analysis plan.
Deliverable: With Claude Code (or an equivalent notebook environment), a structured project: modular scripts, fixed environment, control tests, reporting notebook and documentation, which you review and validate step by step.
Reviewing a manuscript for consistency before submission
- The manuscript was written collaboratively over several months, with successive versions of the analyses. Before submission, everything must be cross-checked: figures in the abstract and results, tables and text, figures and captions, cited references and bibliography, and mandatory declarations. This is the proofreading that no one has time to do thoroughly.
- Gain: Half a day to a full day of cross-checking, and less risk of erratum.
- To be checked: Each inconsistency reported is checked in the document (the agent may also be using the wrong version).
Deliverable: A mission that produces a report of inconsistencies classified by severity, each point located (section, line, table), which the corresponding author addresses before submission.
Project Call Monitoring Officer
- Calls for proposals are multiplying: ANR, Horizon Europe, regional programs, foundations, industry calls. Deadlines are often discovered too late, and everyone on the team monitors their own sources. You want a monthly overview of relevant calls, with their deadlines and their suitability to your research themes.
- Gain: 1 to 2 hours per month of source monitoring, and fewer calls discovered too late.
- Please note: Each deadline is checked on the official page before any decision is made (dates may be subject to change).
Deliverable: A scheduled agent who produces each month a table of open or announced calls, filtered according to themes and team eligibility, with dates, official link and an initial suitability assessment.
4. An example of a worksheet: constructing a bibliographic search equation
All the worksheets follow the same structure: the situation, what you get, the prompt to copy, a follow-up prompt, the checkpoints, the realistic gain, and the safeguard. Here is the prompt for worksheet 1.1, to be pasted as is into your assistant, replacing the highlighted sections.
You are a scientific documentalist specializing in [discipline].
I am looking for literature on: [Research question in one or two sentences].
Construct a bibliographic search equation:
1. Break the question down into 2 to 4 concepts.
2. For each concept, list the synonyms, spelling variants, acronyms and, if it exists, the thesaurus entry. [MeSH / other].
3. Assemble the equation using AND, OR, NOT, quotation marks and truncations.
4. Adapt the syntax for [base 1, e.g., PubMed] et [base 2, e.g., Scopus].
5. Propose 2 exclusion terms to reduce noise, explaining why.
Format: table of concepts, then the two equations in separate blocks.
Don't invent thesaurus descriptors: if you're not sure a term exists, write "to be verified".
Before using the result: the proposed thesaurus descriptors do exist in the database (check in the database search tool); the syntax works: run the equation and look at the first 20 results; three key articles that you already know are found by the equation (recall test); no important concept from your question has been missed.
5. The three non-negotiable rules
No references without verification. Each reference suggested by AI is checked (identifier, metadata, content, article status) before being entered into a manuscript. The kit provides the five-step protocol and the log to track it.
Unpublished works remain in the validated tool. Unpublished data, pre-patent results, personal research data, and manuscripts received for review never go into a public tool. The kit classifies them as red flags.
The researcher remains the author, and says so. The AI makes the proposal; the researcher interprets, decides, and signs it off. Substantial use is declared in the format requested by the journal or funding body: the usage log of the kit serves to write it.
6. What the kit contains
Version 1, published on October 31, 2026. The kit is a 74-page Word document. It opens with a user guide (the matrix of 24 fact sheets, the one-hour-per-week pace, where to begin), what the surveys say, the fundamentals (the four-building block method, requiring references and the "unknown" category, verifying each reference, the equivalent functions of assistants and specialized tools), and the usage framework (the three fires for research data, scientific integrity, and the declaration of use). Next come the four levels, each with its own introduction, six fact sheets, and a progress challenge. Four current affairs files for 2026-2027 address European guidelines, publisher and funding policies, the AI Act and research, and intellectual property before patent filing. A section is intended for the team leader who wants to facilitate the kit, and six appendices complete the package. The Excel workbook brings together seven tools: self-positioning, library of 24 prompts, reference verification register, AI usage log, monitoring plan, pre-submission review grid and time saved measurement.
Download the AI Researchers Kit (Word and Excel, free)
74 pages, 24 fact sheets across 4 levels, 4 progress challenges, 4 current affairs dossiers, and a binder of 7 tools. Written by NextStart.AI based on European guidelines, surveys by Wiley, Elsevier, and Nature, publisher and funding policies, and AI training sessions conducted with research teams.
- Fundamentals of the scientific prompt and three-pronged usage framework
- 24 fact sheets: state of the art, monitoring, protocols, analysis, writing, dissemination
- Six assistants and six agents ready to configure, with their safeguards.
- Reference log, usage journal, monitoring and review in Excel
7. Frequently Asked Questions
Do we already need to use AI to operate the kit?
No. The self-assessment grid indicates where to start, and level 1 begins with short prompts, without sensitive data. A researcher who is already comfortable starts at level 2 or 3, after completing the challenge of the previous level to verify their understanding.
How to avoid invented references?
By not allowing any reference to be included in a manuscript without verification. The kit provides a five-step protocol (identifier, metadata, content, article status, traceability) and an Excel spreadsheet for tracking it.
Can unpublished data be entrusted to AI?
Only in a tool validated by your organization for this purpose, and never for results pending patent applications or manuscripts received for review. The framework of the kit classifies research data into three categories.
Should the use of AI be declared in an article?
Most publishers and funders require it for any substantial use, albeit in different formats. The kit summarizes their policies and provides a usage guide for drafting the declaration in the correct format.
What tools does the kit work with?
Copilot Chat and Microsoft 365 Copilot (often called "Copilot Premium"), ChatGPT, Claude, Gemini and Mistral, in their enterprise versions, or your organization's internal assistant, with references to specialized bibliographic research tools.
Can the kit be distributed throughout the entire laboratory?
Yes, freely within your organization, in its original form; the prompts and the binder can be adapted without restriction.
8. Going further
The kit is self-sufficient for a motivated researcher. It goes faster, and further, when the team uses it together on their own topics: that's the purpose of the AI training for research teams from NextStart.AI, in 1.5 hours or half a day, prepared with your scientific advisor and facilitated using tools validated by your organization. For teams that transform their results into projects, the AI innovator kit extends the Valorisation family.
The entire series of free kits. To drive adoption: AI charter template, AI training plan template, AI Ambassadors Kit, Manager's toolkit for dealing with AI, AI-powered time-saving kit, Adoption Roadmap Kit, AI maturity grid, AI Measurement and ROI Kit, AI Governance Kit, Cowork kit, 50 Copilot M365 prompts, Copilot Studio kit, Deploy Microsoft 365 Copilot kit, Copilot Kit Cat, Copilot Excel kit, Copilot Teams and Outlook kit, Copilot Word and PowerPoint kit, Configure Claude kit, Deploying Claude in the Workplace Kit, Claude Agents Kit, Claude Skills kit, Claude Plugins Kit, Claude Code kit, Claude Design kit et Vibe coding kitFor the AI project sponsor: AI Act kit, GDPR kit, Security and data kit, Shadow AI kit, AI and Social Dialogue Kit, frugal AI kit, AI Sovereignty Kit et responsible AI kitBy profession: AI HR kit, AI marketing kit, AI finance kit, AI kit for lawyers, AI kit for salespeople, AI toolkit for project managers, AI innovator kit, communicator's AI kit, AI Shopping Kit et AI Kit for Researchers.
9. sources
The figures and references cited on this page and in the kit are based on the following sources, verified on October 31, 2026.
- Wiley, AI Adoption Jumps to 84% Among Researchers as Expectations Undergo Significant 'Reality Check' (ExplanAItions 2025), October 2025.
- Elsevier Researcher of the Future (Confidence in Research 2025), November 2025.
- Times Higher Education, Use of AI in research widespread but distrust still high (Oxford University Press survey), May 2024.
- Nature, Is it OK for AI to write science papers? Nature survey shows researchers are split, May 2025.
- Scientific Reports, Fabrication and errors in the bibliographic citations generated by ChatGPT (Walters and Wilder), September 2023.
- ERA Austria (European Research Area), Commission publishes update of ERA Living Guidelines on responsible use of generative AI in research, May 2026.
- European Commission, ERA Living Guidelines on the responsible use of generative AI in research updated, October 2026.
- European Commission, ERA Forum, Living guidelines on the responsible use of generative AI in research, fourth version, October 2026.
- arXiv, Hidden Prompts in Manuscripts Exploit AI-Assisted Peer Review (Zhicheng Lin), July 2025.
- Elsevier Generative AI policies for journals, June 2026.
- Springer Nature (Nature Portfolio), Editorial policies: Artificial Intelligence (AI), accessed in October 2026.
- Wiley, AI guidelines for researchers, October 2025.
- European Research Council, ERC clarifies limits on AI use in grant evaluationMarch 2026.
- National Research Agency (ANR), AI usage guidelines for depositors and appraisers, June 2026.
- European Commission, AI Act Service Desk AI Act, article 2: scope of application, accessed in October 2026.
- Lewis Silkin, The Digital Omnibus on AI enters into force today, July 2026.
- European Patent Office, European Patent Convention, Article 54: new feature, accessed in October 2026.
- Patentology (Mark Summerfield), AI Tools and the Patent 'Prior Disclosure' Problem, August 2026.
- Inserm (via Enssib), Guide to good practices in artificial intelligence at Inserm, February 2025.
- IH2EF, OFIS Symposium: Issues of Scientific Integrity and Artificial Intelligence, June 2025.