1. How to Judge a Research Tool: Criteria and Trade-offs

The first thing university libraries tell students is that there is no single best tool — and they mean it structurally, not as a courtesy. The University of Illinois qualitative-research guide states it flatly: “There is no ‘best’ software or technology for qualitative data analysis (QDA). Most programs designed for QDA offer comparable features.” [1] The University of Alberta Library likewise “does not recommend or endorse any particular product” [2]. What the libraries do instead is hand you a decision framework, and the frameworks converge on the same handful of questions.

Cost model comes first, because it decides everything else. The decisive question is not how much does it cost but who pays. UCL’s guide walks through this explicitly: EndNote Desktop is free on all UCL computers and downloadable to home machines because the institution holds a subscription, while Mendeley and Zotero are free to use but charge for storage above a cap. [3] Oxford makes the same point: “EndNote 2025 is free to University members through the institutional licence.” [4] This matters more than any feature comparison, because a tool that is free at your institution and $139.99 without it is a different tool in each case.

Then: offline capability, operating system, and word processor. UCL notes that “EndNote does not work with Linux” while Zotero runs on Windows, macOS and Linux, and that only EndNote is compatible with Apple Pages. [3] Oxford’s comparison tables add that EndNote Online is the odd one out on word processors — it supports Microsoft Word only — while Zotero covers Word, LibreOffice and Google Docs. [4] If your thesis template is a Word document and your laptop is a Chromebook, these constraints settle the choice before features get a vote.

Durability — will the tool still exist, and can you leave it? This is the criterion students consistently underweight and the one with the longest tail of consequences. A reference library is a multi-year asset; a tool that gets discontinued mid-degree strands it. The evidence here is unambiguous and unusually concrete: Mendeley retired its mobile apps in March 2021, Elsevier stopped downloads of Mendeley Desktop in September 2022, and a planned 2024 shutdown of Desktop sign-ins was reversed only in July 2025 after user pressure. [5] The University of Alberta’s summary table lists Mendeley’s drawbacks as a “small number of reference types”, “limited group sharing for free accounts” and “no Google Docs integration”. [2] A widely-circulated 2025 critique of Zotero makes the exit problem concrete from the other direction: its author’s “most substantial problem” is that “it is much easier to put information into Zotero than it is to get it out.” [6] Both are real risks. The practical test is whether the tool can export your entire library in a format something else can read — Zotero’s BibTeX export is the reason that critique still calls it “absolutely the right tool” for most people.

Then discipline fit, learning curve, collaboration, and citation-standard compliance. Collaborative writing needs to survive co-authors: Mendeley’s free tier caps groups at five groups of twenty-five members, while Zotero’s groups draw storage from the group owner’s account. [4] Qualitative software adds two criteria that do not appear in reference-manager comparisons at all — whether the tool can store data on a vendor’s server at all (a research-ethics question, not a convenience question) and whether it is available through your institution’s licence rather than only for purchase.

Two structural facts cut across all of these criteria. First, price and feature claims decay fast: Alberta warns that “software features change regularly”, and the corpus bears it out — a 2018 journal comparison, a 2021 institutional table, a 2017 forum thread and a 2026 encyclopaedia entry all disagree about the same products’ prices. Every cost figure in this report therefore carries its date. Second, the 2026 landscape has a genuinely new axis: the AI layer, and with it a set of integrity obligations that did not exist when most of these guides were written.

2. The Core Toolkit: Literature Discovery, Reference Management, and Full-Text Access

Discovery: the comprehensiveness/order trade-off

The choice between Google Scholar, Web of Science and Scopus is not a choice between good and bad; it is a trade between “more comprehensive, but disorderly systems and orderly, but limited systems.” [7] That framing, from researchers who measured the three databases directly, is the single most useful sentence in this whole area.

The peer-reviewed evidence on Google Scholar’s reliability is severe and should change how you use it. Sauvayre’s study of errors in Scholar data concluded that virtually every reference it examined contained at least one error, which calls the reliability of Google Scholar data into question. It also warns that studies leaning on large-scale samples without verification and data cleaning are themselves affected, including by substantial biases and citation miscounts. [8] For calibration, that same paper puts Web of Science’s average error rate at 0.1% and Scopus’s at 1.0% — one to two orders of magnitude apart from Scholar.

Separately, a study of Scholar as a replacement for systematic literature searches found high relative recall but very low precision, and Haddaway’s work showed Scholar missing important literature in five of six case studies while overlapping inconsistently with Web of Science. [9] [10] There is counter-evidence too: Scholar’s inclusiveness exposes real gaps in the selective databases, especially in the social sciences and humanities. [10]

The practical rule: use Scholar to scope and to catch grey literature, use a curated database for the citations you will actually rely on, and never treat a Scholar citation count as a verified number.

Reference management: Zotero as the default, EndNote when your institution pays

For most students the decision collapses to three products plus an exit clause.

Zotero is the default recommendation. It is free and open source, runs on Windows, macOS and Linux, supports Word, LibreOffice and Google Docs, and claims 10,000+ citation styles — the largest count in Oxford’s comparison table, ahead of EndNote’s 7,500+. [4] Its free tier includes unlimited references with 300 MB of cloud file storage; paid storage is $20/year for 2 GB, $60/year for 6 GB and $120/year for unlimited, and those subscriptions fund the nonprofit that develops it. [11] Its documented weaknesses are the storage cap, that it “works best with one primary workstation” in Alberta’s phrasing, and the export/friction complaints above. [2] [6]

EndNote is the right answer when — and largely only when — your institution licenses it. Oxford and UCL both provide EndNote 2025 free to members. [4] [3] Bought personally it costs about $139.99–$249.95, its licence is tied to one device, and its learning curve is steeper. [2] It is the strongest option for very large libraries and for systematic reviews, where librarians prefer it for deduplicating large result sets, but it does not run on Linux and struggles with LaTeX workflows.

Mendeley should be chosen deliberately or not at all. It is free with 2 GB of storage and is genuinely good for Elsevier-ecosystem users, but its current paid tiers run $4.99–$14.99/month (or $55–$165/year), it has no Google Docs integration, and its history of retired features and reversed shutdowns is the clearest lock-in warning in the category. [12] [5]

There is direct evidence on which of these produces correct citations, which is the actual job. Kratochvíl compared machine-generated references against manually written ones across 15 biomedical journals and the NLM style. Zotero made the fewest mistakes for 11 of the 15 journals and Mendeley was second, while EndNote was worst for nine journals and RefWorks for the other four. Every tool failed on the same thing — URLs and access dates for online sources. The author’s conclusion is that “Zotero and Mendeley are the most suitable managers.” [13] That is 2017 evidence against 2016–2017 software versions, so treat the ordering as historical rather than current — but it is the only peer-reviewed accuracy comparison in this corpus, and it points the same way as the library guides.

Full-text access

Discovery is worthless if you cannot read the paper. Unpaywall is the workhorse here: a free database of tens of millions of legal open-access articles, queried automatically by browser extensions and by reference managers. [14] Combined with your library’s own link resolver and interlibrary loan, it closes most of the gap that a paywall creates — lawfully, which matters for a thesis you will have to defend.

3. Writing, Citation, and Research-Integrity Tools

Writing and formatting

The word-processor decision is less aesthetic than institutional. Aalto University’s guidance is that Microsoft Word is the pragmatic default — user-friendly, with templates, citation features, spell and grammar check and collaboration — and that Aalto students get Office 365 free. [15] LaTeX, usually via Overleaf, is the better tool for mathematically dense work and for anyone who wants citations and cross-references handled by code rather than by hand; Aalto notes that Google Docs “may not be ideal for longer documents (i.e. theses)” though it is fine for short collaborative pieces. [15]

The load-bearing advice is about templates, and it is a hard rule rather than a preference: use your own faculty’s current template. Formatting requirements are enforced by examination committees and similarity-check systems, not by taste. Romania’s own requirements are concrete and worth stating because they are typical of enforced templates. At the West University of Timișoara’s Faculty of Letters, History, Philosophy and Theology, a licență thesis must be a minimum of 35 pages (Applied Modern Languages), 40 pages (Languages and Literatures, History, Orthodox Pastoral Theology) or 30 pages (Philosophy). Length aside, the format is fixed: Times New Roman 12, 1.5 spacing, justified, with 3 cm margins left and bottom and 2.5 cm right and top, and the cover page must follow the faculty’s Annex 1 model exactly. [16]

Similarity checking: what the software does and does not do

The most important operational fact about plagiarism detection is on the vendors’ side of the fence: Turnitin, iThenticate and Crossref Similarity Check “are the same detection engine, offered through three different channels to three different markets,” and a similarity score “is not a plagiarism verdict.” [17] Turnitin is the education-market product universities license to screen coursework and thesis deposits; iThenticate serves publishers and pre-submission drafts and does not index submissions into a repository. The score is the percentage of your text matching somewhere in a corpus — high scores can come from properly cited quotations and shared methodological boilerplate, and low scores can miss paraphrased or translated borrowing. A 2024 comparison found a statistically significant difference in mean similarity indices across the three engines and, usefully, no significant difference in which sources they matched. The tools agree about what is similar and disagree about how much to flag. [18]

Treat similarity checking as a diagnostic you run before submission, not as a verdict you receive after it. In Romania the similarity report is a formal submission artefact: UVT requires the similarity report to be deposited together with the licență thesis as certification of authenticity. [16]

The new integrity layer: AI use must be declared

For theses, the governing question in 2026 is no longer whether AI tools may be used but how their use is disclosed — and disclosure is becoming mandatory. Université Laval’s Faculty of Graduate and Postdoctoral Studies states that “starting in the winter 2026 semester, Master’s and doctoral students who make their initial submission must submit a mandatory disclosure on the use of generative AI,” and failure to do so delays submission. [19] McGill’s thesis guidance sets a five-part disclosure framework — tools used, purpose, extent, human oversight and verification, and impact on the research — and relates that its attestation “may become mandatory” after Fall 2026 consultations. [20]

Munster Technological University’s booklet states the boundary in one sentence: “GenAI tools may support, but must not replace, the student’s intellectual and analytical work.” [21] MTU also lists permitted uses — language editing, translation of your own text, readability, formatting, and workflow support such as coding or brainstorming — and prohibits generating substantive academic text, fabricating data, and misrepresenting authorship. Both MTU and McGill place responsibility for verifying AI-assisted outputs, citations included, squarely on the student. [21]

Two practical consequences follow. Keep a log of your AI interactions as you go — retrofitting a disclosure at submission time is both harder and less honest. And cite AI tools properly if you use them; library guides now publish APA, Chicago and MLA formats for doing so. [22]

4. Data Analysis and Knowledge-Management Tools

Qualitative data analysis

The free/open-source tier is now genuinely usable, which changes the answer for undergraduates. Taguette is free and open source, handles text coding, and can be self-hosted — meaning “your data is as secure as your computer.” [23] QualCoder has the widest feature set of the free options, importing text, survey, image, audio and video, though Illinois warns its “installation may be a challenge.” [24] Both are legitimate choices for a licență thesis. [25]

Above them sit the commercial packages with student licences — NVivo, ATLAS.ti, MAXQDA, Dedoose, Delve and Quirkos — whose real differences are collaboration model, learning curve and data handling rather than coding capability. [26] [27] Two selection criteria deserve more weight than they usually get. First, where your data lives: Illinois warns explicitly that cloud QDA tools store data on a vendor’s server. Researchers must confirm this is permitted by their ethics approval and participant agreements. [1] Second, who can work simultaneously: real-time collaboration exists only in a subset of the tools, while NVivo and MAXQDA desktop are merge-only. [1]

One caution about the evidence itself: the comparison matrices published by several university libraries descend from a common ancestor chart, so three library guides are not three independent sources. The genuinely independent second axis is the CAQDAS Networking Project’s selection framework, which — after reviewing the whole field — declines to name a best package and suggests that local support and the availability of free or cheap software is a sensible starting point. [28]

Quantitative analysis

SPSS is the incumbent and the most expensive habit a student can acquire. The free alternatives are now strong enough that the burden of proof has shifted: jamovi is free and open source, runs offline, and shows the exact R syntax behind every analysis, so point-and-click work stays reproducible. [29] JASP is likewise free and open source and covers frequentist and Bayesian versions of the same analyses. [30]

The honest caveat is in jamovi’s own documentation: it “currently has fewer features than SPSS,” covering roughly “90% of the most frequently used analyses in psychology,” with gaps in procedures such as OLAP cubes and ratio statistics and with advanced methods requiring additional modules. [31] It also beats SPSS in at least one place — its Bayesian ANOVA “can handle several factors while SPSS is limited to one factor.” The practical implication: for a typical licență or dissertation, jamovi or JASP is sufficient and free; check the specific procedure your method requires before committing, and step up to R (which both tools sit on) if you hit a gap.

Knowledge management

For doctoral work the useful arrangement is a division of labour: the reference manager stores sources and the note tool stores thinking. The established bridge is Zotero plus Obsidian, connected through Better BibTeX and one of two community plugins — one reading a static exported file, the other connecting live and pulling in PDF annotations. [32] The workflow’s fragility is the part rarely mentioned: unpinned citation keys regenerating is described as “the single most common cause of ‘my literature notes stopped matching my library'”. A recent Zotero major version then changed how citation keys work, superseding the pin-your-keys advice that most existing guides give. [32] [33] If you adopt this stack, adopt it early, and prefer the live-connection plugin over the static export.

5. Best Tools for Bachelor’s Theses

A licență thesis is a bounded project with an enforced format and, usually, no budget. The right stack is the cheap one, chosen so nothing has to be relearned later.

Minimum viable toolkit:

• References: Zotero [11]
• Discovery: Google Scholar for scoping plus your library’s curated database for the citations you actually rely on [8]
• Writing: Microsoft Word with your faculty’s template [15]
• Similarity check: the institutional similarity check before submission [17]
• Statistics: jamovi or JASP rather than buying SPSS [29]
• Qualitative: Taguette or QualCoder rather than buying NVivo [25]

Romanian specifics to plan around:

• Page minimums by specialisation (30–40 pages depending on faculty)
• Formatting: Times New Roman 12 at 1.5 spacing with 3 cm/2.5 cm margins
• Cover page: the faculty’s exact cover template
• Bibliography: at least 15 titles
• Theory chapter: capped at roughly 30% of the work
• Submission: with the similarity report through the faculty’s platform [16]

Budget working days, not hours, for the submission window.

The single highest-value habit at this level is starting the reference library on day one. The second is reading the faculty’s template annex before writing a single page.

6. Best Tools for Master’s Dissertations

A dissertation adds an empirical component and a longer literature base, so the marginal tool is the one that handles your method, not your citations.

Keep Zotero as the reference spine and add storage if you will accumulate PDFs — $20/year for 2 GB is the cheapest upgrade in this report. [11] If your dissertation includes a systematic or structured review, add a screening tool: a feature analysis of fifteen such tools recommended Covidence and Rayyan, and the authors of that evaluation concluded, “we would recommend Covidence and Rayyan to systematic reviewers looking for suitable and easy to use tools to support T&Ab screening within healthcare research.” [34] Rayyan’s free tier is real — free forever with three active reviews and two reviewers — which makes it the sensible entry point. [35]

For the empirical chapter, the choice should follow the method rather than the marketing: a student licence to a commercial QDA package if your supervisor’s workflow depends on it, Taguette or QualCoder if not [26]; jamovi or JASP if quantitative, with R as the escape hatch. [31] If the work is mixed-methods, check collaboration and data-storage constraints before committing, not after. [1]

Disertație formatting is longer but the same logic applies — UVT’s requirements set a minimum of 50 pages for a disertație — and the similarity report requirement is unchanged. [36]

7. Best Tools for Doctoral Research

Doctoral work changes the problem rather than the tools: the literature base runs to thousands of references over four or more years, reproducibility starts to matter, and the failure modes are cumulative. Three requirements dominate.

Bibliographic scale and durability. Choose a reference manager you can still export from in year five. Zotero with paid storage handles this at $60–$120/year. [11] EndNote is a defensible choice if your institution licenses it and your discipline is large-library, biomedicine-heavy and Word-based. That is precisely where the peer-reviewed accuracy comparison put it alongside Zotero and Mendeley at the top, and where librarians prefer it for deduplicating systematic-review result sets. [13] Whichever you choose, export a BibTeX or RIS backup on a schedule. Tools get discontinued; libraries must not. [5]

Review methodology. For any systematic or scoping component, the screening tool is not optional at this scale — Rayyan for the free route [35] or Covidence where a full Cochrane-aligned workflow with data extraction and PRISMA flow diagrams is needed. [34] Build the search strategy across multiple databases rather than trusting one: the evidence that Google Scholar alone is insufficient for evidence reviews is direct and repeated. [10] [9]

AI tools, used with discipline. AI literature tools are legitimately useful at doctoral scale, and they are also the newest way to damage a thesis. Their citation behavior has improved, but the peer-reviewed record is cautionary. An evaluation of Elicit found repeatability failures — identical searches returning 246, 169 and 172 results on successive runs — and concluded the tool cannot be used on its own; a 2026 umbrella-review evaluation found retrieval overlap with a Boolean search of only about 8% between the two approaches. [37] [38] Earlier work on general-purpose assistants documented substantial hallucinated-reference rates for literature-search tasks. [39]

Elicit and Consensus are the two most-recommended tools in this category and both are transparent about their corpora — 138 million papers and 545,000 clinical trials for Elicit, 250 million papers for Consensus, the latter with a student discount worth checking. [40] [41] Note the commercial framing: Elicit’s own accuracy claims come from an evaluation it ran and published itself, and it concedes the differences against competitors are “small and generally not statistically significant.” [42]

The doctoral rule is straightforward: use AI search to find candidates, never to certify them. Verify every citation against the source before it enters your library, and declare the use in the form your institution requires — at Laval that is mandatory from winter 2026, and the direction is clearly that way everywhere. [19] [21]

8. Recommended Tool Stacks by Degree Level

Job	Bachelor’s (licență)	Master’s (disertație)	Doctoral (doctorat)
Reference management	Zotero (free, 300 MB) [11]	Zotero + 2 GB storage ($20/yr) [11]	Zotero + 6 GB–unlimited, or institutional EndNote [11] [4]
Discovery	Google Scholar to scope; library database to cite [8]	Scholar + Scopus/Web of Science, citation chaining [7]	Multiple curated databases; Scholar for grey literature only [10]
Full-text access	Unpaywall + library resolver [14]	same	same + interlibrary loan
Writing	Word + faculty template [15]	Word or LaTeX/Overleaf	LaTeX/Overleaf where discipline warrants
Similarity check	Institutional Turnitin before submission [17]	same + similarity report deposited [16]	Same, plus pre-submission iThenticate for manuscripts
Qualitative analysis	Taguette or QualCoder (free) [25]	Student licence commercial package if method requires [26]	NVivo/ATLAS.ti/MAXQDA with data-storage constraints checked [1]
Statistics	jamovi or JASP (free) [29]	jamovi/JASP, R for gaps [31]	R/Python for reproducibility
Review screening	—	Rayyan (free tier) [35]	Rayyan or Covidence [34]
Knowledge management	—	Optional: Zotero + Obsidian [32]	Zotero + Obsidian, adopted early [33]
AI tools	Use sparingly; check your faculty’s rule	Use with disclosure [21]	Elicit/Consensus for discovery only; verify every citation [37]

Three commitments explain most of the difference between a thesis that goes smoothly and one that does not:

• Start the reference library on day one and keep an exportable backup — every tool in this report can be replaced except the time you spent re-entering references.
• Choose the free tool first and pay only for a specific capability you can name — Zotero over EndNote, jamovi over SPSS, Taguette over NVivo, unless your institution licenses the paid one for free, in which case take the licence and keep the export habit.
• Treat AI as a search assistant with a disclosure obligation attached — the tools have improved markedly on citation accuracy, but the peer-reviewed evaluations still describe tools that cannot stand alone, and every institution in this corpus is moving toward requiring you to state what you used and how you verified it. [19]

Underneath all of it sits the finding the libraries keep repeating: the tool matters far less than the discipline of using it. A verified citation in Zotero is worth more than an unverified one in a $250 package, and a similarity report you ran yourself two weeks before submission is worth more than any tool’s score. [1]

Sources

• Choosing QDA Software – Qualitative Data Analysis – LibGuides at University of Illinois at Urbana-Champaign. https://guides.library.illinois.edu/qualitative/qdas [1]
• Reference Management – Citation and Reference Management – Subject Guides at University of Alberta Libraries. https://guides.library.ualberta.ca/citing/reference-management [2]
• Which software should I use? – Reference management software – Library guides and databases at University College London, Global. https://library-guides.ucl.ac.uk/reference-management-software/which-software [3]
• Choosing a Reference Manager – Managing your references – Oxford LibGuides at Oxford University. https://libguides.bodleian.ox.ac.uk/reference-management/comparison-tables [4]
• Mendeley – Wikipedia. https://en.wikipedia.org/wiki/Mendeley [5]
• so-long-zotero.html. https://cosroe.com/2025/02/so-long-zotero.html [6]
• Google Scholar, Web of Science, and Scopus: Which is best for me? – LSE Impact. https://blogs.lse.ac.uk/impactofsocialsciences/2019/12/03/google-scholar-web-of-science-and-scopus-which-is-best-for-me/ [7]
• Romy Sauvayre (2022). Types of Errors Hiding in Google Scholar Data. Journal of Medical Internet Research. https://doi.org/10.2196/28354 [8]
• Martin Boeker et al. (2013). Google Scholar as replacement for systematic literature searches: good relative recall and precision are not enough. BMC Medical Research Methodology. https://doi.org/10.1186/1471-2288-13-131 [9]
• Neal Haddaway et al. (2015). The Role of Google Scholar in Evidence Reviews and Its Applicability to Grey Literature Searching. PLoS ONE. https://doi.org/10.1371/journal.pone.0138237 [10]
• Zotero | Storage. https://www.zotero.org/storage [11]
• Mendeley | Pricing. https://www.mendeley.com/reference-management/premium [12]
• Jiří Kratochvíl (2016). Comparison of the Accuracy of Bibliographical References Generated for Medical Citation Styles by EndNote, Mendeley, RefWorks and Zotero. The Journal of Academic Librarianship. https://doi.org/10.1016/j.acalib.2016.09.001 [13]
• Unpaywall: An open database of 20 million free scholarly articles. https://unpaywall.org/ [14]
• Tools for academic writing | Aalto University. https://www.aalto.fi/en/services/tools-for-academic-writing [15]
• Regulament privind organizarea și. https://lift.uvt.ro/wp-content/uploads/2024_2025_Anexa-2-Structura-generala-a-lucrarii-de-licenta_disertatie.pdf [16]
• Anti-Plagiarism Software: Turnitin vs iThenticate — CASRAI. https://www.casrai.org/guides/anti-plagiarism-software [17]
• (2024). Journal of Physical Education and Sport ® (JPES), Vol. 24 (issue 2), Art 51, pp. 419 – 424, February 2024. Journal of Physical Education and Sport. https://doi.org/10.7752/jpes.2024.02051 [18]
• Responsible use of generative AI in Master’s and Doctoral theses | Études supérieures et stages postdoctoraux | Université Laval. https://www.fesp.ulaval.ca/en/masters-and-doctoral-theses/responsible-use-of-generative-ai-in-masters-and-doctoral-theses [19]
• GenAI Research Statement and Disclosure Guidelines for Theses (McGill GPS, May 1 2026). https://www.mcgill.ca/gps/files/gps/generative_ai_statement_for_thesis_students_may_1_2026.pdf [20]
• Guidance on the Use of Generative AI Tools in Theses mtu.ie | 01. https://www.mtu.ie/media/mtu-website/download/MTU—AI-Digital-Booklet-v3.pdf [21]
• Citing AI – Artificial Intelligence – Guides at McGill Library. https://libraryguides.mcgill.ca/ai/citation [22]
• Taguette, the free and open-source qualitative data analysis tool. https://www.taguette.org/ [23]
• QualCoder | Computer aided qualitative data analysis software. https://qualcoder.wordpress.com/ [24]
• Free Tools for QDA – Qualitative Data Analysis – LibGuides at University of Illinois at Urbana-Champaign. https://guides.library.illinois.edu/qualitative/free [25]
• Tools for Analysis – Qualitative Data Analysis – Research Guides at University of Delaware. https://guides.lib.udel.edu/qda/tools [26]
• Qualitative Tools – Software for Data Analysis – Research Guides at Northwestern University. https://libguides.northwestern.edu/c.php?g=114906&p=9075213 [27]
• Choosing a CAQDAS Package. https://www.surrey.ac.uk/sites/default/files/2009ChoosingaCAQDASPackage.pdf [28]
• jamovi — open statistical software. https://www.jamovi.org/ [29]
• JASP – A Fresh Way to Do Statistics. https://jasp-stats.org/ [30]
• Comparison of Which Analyses Are Available in SPSS and jamovi — jamovi Documentation. https://docs.jamovi.org/spss2jamovi/s2j_Comparison_of_analyses.html [31]
• Zotero to Obsidian: Literature Notes Setup — CASRAI. https://www.casrai.org/guides/zotero-obsidian-literature-notes [32]
• Better BibTeX for Zotero. https://retorque.re/zotero-better-bibtex/ [33]
• Hannah B. Harrison et al. (2020). Software tools to support title and abstract screening for systematic reviews in healthcare: an evaluation. BMC Medical Research Methodology. https://doi.org/10.1186/s12874-020-0897-3 [34]
• Rayyan Pricing | Choose the Best Plan for Your Research Team. https://www.rayyan.ai/pricing [35]
• MINISTERUL EDUCAȚIEI ȘI CERCETĂRII. https://www.uvt.ro/wp-content/uploads/sites/3/2024/03/Regulament-examene-finalizare-studii-licenta-si-masterat-UVT.pdf [36]
• Using artificial intelligence for systematic review: the example of elicit – PMC. https://doi.org/10.1186/s12874-025-02528-y [37]
• Evaluating Elicit’s systematic reviews workflow in an umbrella review on air pollution and acute lower respiratory infections: a methodological study for quality appraisal – PubMed. https://doi.org/10.1186/s12874-026-02892-3 [38]
• Journal of Medical Internet Research – Hallucination Rates and Reference Accuracy of ChatGPT and Bard for Systematic Reviews: Comparative Analysis. https://doi.org/10.2196/53164 [39]
• Elicit: AI for research & decision-making. https://elicit.com/?utm_source=direct [40]
• Consensus: AI for Research. https://consensus.app/?utm_source=direct [41]
• How we evaluated Elicit Reports – Elicit. https://blog.elicit.com/elicit-reports-eval/ [42]