Last updated: October 2025
This evaluation is grounded in a computational biology research environment — days split between running experiments, analyzing data, and writing papers. Over the past year, AI tools have been integrated into nearly every stage of the research workflow. Some of them materially improved how the work gets done. Others were impressive demos that didn’t survive contact with real academic work.
Here’s what actually gets used, what’s been abandoned, and what would be recommended to fellow researchers.
Literature Discovery and Review
Finding relevant papers is the foundation of research, and it’s also where most of us waste enormous amounts of time. Google Scholar works, but it’s a blunt instrument. These tools are sharper.
Semantic Scholar
Semantic Scholar (by the Allen Institute for AI) has become the primary literature search tool in this workflow, replacing Google Scholar for most tasks. The AI features that matter:
TLDR summaries: Every paper gets a one-sentence AI-generated summary. When you’re scanning 50 search results, these summaries let you identify relevant papers in seconds instead of clicking through to read each abstract.
Semantic Reader: Open a paper in their reader and you get inline definitions of technical terms, citation context (what the cited paper actually found, not just that it exists), and a structured summary of the paper’s contributions. For papers outside one’s immediate subfield, this is invaluable.
Research feeds: Tell it your research interests and it surfaces new papers daily. The recommendations are better than Google Scholar alerts because they understand semantic similarity, not just keyword matching. It found a relevant paper in a robotics journal that would never have been discovered through usual channels. The methods were applicable to protein folding work despite being in a completely different field.
What works:
- Search quality is excellent for finding conceptually related work, even when the terminology differs across fields
- The citation graph visualization helps you trace the lineage of an idea quickly
- It’s fast. Noticeably faster than Google Scholar for complex queries
- Completely free
What doesn’t:
- Coverage gaps in humanities and social sciences. It’s strongest in CS, biomedicine, and engineering
- The TLDR summaries occasionally miss the point of a paper, especially for theoretical work
- No full-text search. You’re searching titles, abstracts, and metadata
Pricing: Free. Entirely free. No premium tier.
Elicit
Elicit is purpose-built for literature review. You ask a research question in natural language (“What are the effects of sleep deprivation on memory consolidation in adults?”) and it returns relevant papers with extracted data points, methodology summaries, and key findings organized in a table.
The evaluation used Elicit for a systematic review last year. The traditional approach would have been: search multiple databases, screen hundreds of abstracts, read dozens of full papers, and extract data into a spreadsheet. Elicit compressed the initial screening phase from two weeks to two days.
The killer feature: You can define columns for data extraction (sample size, methodology, key findings, limitations) and Elicit fills them in automatically from the papers. It’s not perfect — I’d estimate 80-85% accuracy — but it gives you a structured starting point that you then verify and correct. That’s dramatically faster than building the table from scratch.
What works:
- Natural language search is genuinely better than keyword search for exploratory questions
- The data extraction tables save enormous time on systematic reviews
- It identifies methodological details (sample size, study design, statistical methods) reliably
- The “find similar papers” feature is excellent for expanding your search
What doesn’t:
- It sometimes surfaces tangentially related papers with high confidence. Always verify relevance yourself
- The free tier is limited to 5,000 credits/month, which runs out fast during intensive literature review
- Full-text analysis is only available for open-access papers. Paywalled papers get abstract-only treatment
- Extraction accuracy drops for complex or unusual paper formats
Pricing: Free tier (5,000 credits/month). Plus is $12/month (12,000 credits). Pro is $49/month (unlimited credits + priority processing). For a serious literature review, you’ll want at least Plus.
Connected Papers
Connected Papers creates visual graphs of related papers. Input one seed paper and it generates a network showing the most similar papers based on co-citation and bibliographic coupling. It’s not AI in the deep learning sense, but the algorithm is smart and the visualization is immediately useful.
This tool proves useful at the start of every new project to map the intellectual neighborhood of a research question. In 10 minutes, it’s possible to identify the seminal papers, the recent developments, and the key research groups in any subfield.
Pricing: Free for 5 graphs/month. Academic plan is $3/month for unlimited graphs. Absurdly good value.
Reading and Analyzing Papers
SciSpace (formerly Typeset)
SciSpace’s Copilot sits alongside a paper as you read it. Highlight any passage and it explains it in simpler terms, defines technical jargon, or relates it to other concepts. For reading papers outside your specialty, this is like having a knowledgeable colleague looking over your shoulder.
The evaluation used it extensively when reading immunology papers for a cross-disciplinary collaboration. Instead of spending 20 minutes Googling every unfamiliar term and concept, I’d highlight and get an instant, context-aware explanation.
What works:
- Explanations are grounded in the paper’s context, not generic definitions
- The math explanation feature breaks down equations step by step
- You can ask follow-up questions about specific sections
- It handles tables and figures reasonably well
What doesn’t:
- Occasionally oversimplifies complex concepts. The explanations are good for orientation but shouldn’t replace deep understanding
- Processing time for long papers (30+ pages) can be slow
- The Chrome extension sometimes conflicts with journal website layouts
Pricing: Free tier with limited queries. Premium is $20/month or $199/year. Student discount available.
NotebookLM for Research
Google’s NotebookLM deserves special mention for research use. Upload your collection of papers (up to 50 sources) and it creates an interactive research assistant grounded entirely in those documents. Every answer includes citations to specific passages in your uploaded papers.
It excels at synthesizing findings across multiple papers. “What do these 15 papers say about the relationship between X and Y?” gets a structured answer with citations. It’s like having a research assistant who’s actually read everything.
Recent updates have made it even more useful: it now processes up to 8x more data per notebook, has 6x longer conversation memory for complex multi-step analysis, and can generate slide decks and structured data tables from your research. The mobile app also supports editing slides on the go.
The audio overview feature generates a podcast-style discussion of your uploaded papers. Listening to these during a commute is a great way to identify connections missed during reading.
Pricing: Free.
Writing and Editing
Writefull
Writefull is trained specifically on published academic papers. Unlike Grammarly (which is trained on general English), Writefull understands academic conventions — hedging language, citation integration, section-specific writing styles, and field-specific terminology.
What it catches that Grammarly doesn’t:
- “This proves that…” → suggests “This suggests that…” (appropriate hedging)
- Inconsistent terminology within a paper
- Sentences that are grammatically correct but stylistically unusual for academic writing
- Overuse of passive voice in sections where active voice is conventional (and vice versa)
The paraphraser is useful for revising awkward sentences while maintaining academic tone. The “academizer” takes informal writing and adjusts it to academic register, which is useful for converting rough notes into draft prose.
Pricing: Free tier with basic features. Premium is $15.37/month or $5.41/month billed annually. Institutional licenses available.
Paperpal
Paperpal (by Cactus Communications, a major academic editing company) offers AI-powered language editing for manuscripts. It goes beyond grammar to check for clarity, conciseness, and academic tone. The “Submission Readiness” check evaluates whether your paper meets journal formatting and language standards.
Paperpal was used for the last two journal submissions. It caught several issues that co-authors had missed: inconsistent abbreviation usage, a dangling modifier in the abstract, and a results section paragraph that was technically accurate but confusingly structured.
Pricing: Free basic checks. Prime is $11.99/month or $95.88/year. Institutional pricing available.
Citation Management
Zotero + AI Plugins
Zotero itself isn’t AI-powered, but the plugin ecosystem has exploded with AI tools. Top picks:
zotero-gpt: Adds a chat interface to your Zotero library. Ask questions about your collected papers, get summaries, find connections between papers. It’s like NotebookLM but integrated into your citation manager.
Better BibTeX: Not AI, but essential. Automatically generates and manages citation keys, handles export to LaTeX, and keeps your .bib file clean.
Zotero’s built-in reader now includes annotation tools that work well with AI summarization plugins. Highlight key passages, tag them, and export structured notes.
Pricing: Zotero is free (with 300MB cloud storage). Additional storage is $20/year for 2GB or $120/year for unlimited. Plugins are free.
The Actual Workflow
Here’s how these tools fit together in practice:
- Discovery: Semantic Scholar for broad search + Connected Papers for mapping the field + Elicit for structured literature review
- Reading: SciSpace Copilot for papers outside the immediate field + NotebookLM for synthesizing across papers
- Writing: Draft in Overleaf → Writefull for academic language editing → Paperpal for final submission check
- Citations: Zotero with Better BibTeX throughout the entire process
Total monthly cost: about $40-60 depending on which premium tiers currently using at any given time. That’s less than a single hour of professional editing, and The workflow uses these tools daily.
What I’d Recommend
If you’re a grad student just starting: Semantic Scholar (free) + Elicit free tier + NotebookLM (free) + Zotero (free). This gives you a complete research toolkit for $0. Add Writefull when you start writing your first paper.
If you’re doing a systematic review: Elicit Pro ($49/month) is worth it for the duration of the review. The data extraction tables alone save weeks of work. Cancel when you’re done.
If English isn’t your first language: Writefull + Paperpal together cost about $17/month and will significantly improve your manuscript quality. Cheaper and faster than professional editing for routine language issues.
If you read across disciplines: SciSpace Copilot is the single most useful tool for understanding papers outside your expertise. The $20/month is well spent if you do any cross-disciplinary work.
One important caveat: none of these tools replace critical thinking. They speed up the mechanical parts of research (finding papers, extracting data, polishing prose), but the intellectual work of formulating questions, designing experiments, interpreting results, and constructing arguments is still entirely on you. Use AI to handle the tedium so you can spend more time on the thinking that actually matters.
Current researcher guide: AI tools for researchers in 2026.