Research tools in 2025 summarised in one word: DEPTH.

These tools (unlike design tools) have been around and getting better linearly with every LLM improvement since GPT-3 dropped.

They stopped being just better search engines and became actual research systems.

The tedious parts are solved. Finding papers, extracting data, verifying citations are all faster and accessible now.

You don’t need a university library card anymore.

I tested 12 research tools this year. Here's what won.

We put together an AI Survival Hackbook for 2026: a practical guide on how AI changed work between 2019–2025, what will matter next year, and how professionals are quietly using AI for leverage (not hype).

It includes:

It’s free for the next two weeks.

Launched in Feb 2025.

The time it takes to research a topic now is down to 3 minutes now. Used to be 8-10 hours before.

That’s a 200x improvement.

The technical edge: 93.9% accuracy on SimpleQA benchmark. That's fact-checking across thousands of questions, verified against credible sources.

Perplexity beats Gemini, ChatGPT and Deepseek on most metrics.

In October 2025: they made it FOC for everyone.

When Zapier tested both on 10 citation tasks, Perplexity finished 22% faster overall.

Runner-up: Google Gemini (instant results but sacrifices depth)

They have been around for a while but completely overhauled their design in October 2025.

Discovery at the crux of it is about finding papers you'd never think to search for.

They partnered up with Litmaps and that changed everything.

The new visualization maps citation networks with customizable X/Y axes, publication year, citation count, and anything that matters to your field.

UX wise: start with three papers. It shows you the research landscape: foundational work, current developments, emerging trends.

PhD students say it cuts literature review time by 50%. "I had to deal with dozens of different articles, references, and authors... ResearchRabbit made a lot of difference."

Free tier allows 50 input papers, 5 authors, 1 project. That’s enough for most graduate students.

Runner-up: Litmaps (better visualizations, steeper learning curve)

This isn't even close.

Most tools count citations. Scite interprets them.

From billions of in‑text citation statements across 200M sources, Scite shows whether a paper is cited as supporting, contrasting, or simply mentioning another work.

Their deep learning model classifies each citation by rhetorical function and not as sentiment.

For eg: A supporting citation can have negative language, or a contrasting citation can be p

Their team keeps shipping.

In November 2025, they added patent search where you can search through 200M+ patent families with filters and PDFs.

Limitations: Their coverage is deepest in medical/life sciences, followed by computer science and physics, the rest (eg: humanities) still lighter but improving.

Runner-up: Consensus

Elicit has indexed and extracted over 138 million papers and is the best at attaching the exact sentence from the source paper so readers can verify the claim in context.

In a May 2025 study: they found that Elicit has 39.5% sensitivity and 41.8% precision.

Translation: recalls 4 out of 10 studies, but over 40% of what it finds is actually relevant.

That doesn’t sound great. but it’s margins are much better than traditional broad searches.

A case study: Formation Bio (pharma company) extracted 40 technical variables from 300 papers.

"Hundreds of hours" of work reduced to 10.

They also have integrated my favourite model in Claude Opus 4.5.

Key features you actually use:

The limitation: Lower search sensitivity means you shouldn't use it as your only search tool. Combine with Google Scholar for comprehensive coverage.

Runner-up: Consensus

Contains over 220M+ peer-reviewed sources.

Here's the problem Consensus solves for what matters.

You find 50 papers on a topic. Some say yes, some say no, some say maybe. Which ones matter?

Consensus Meter 2.0 (launched in Feb) counts these votes AND weighs them.

An N=1 case report doesn't count the same as a Cochrane Systematic Review. A 2010 study with 50 citations doesn't weigh equal to a 2024 meta-analysis with 500.

The Meter shows methodology quality, recency, journal impact, study design and visualises it.

That's how science actually works. Not all evidence is equal.

It is also feature rich.

Like a medical mode for high quality medical sources only. And a deep search mode that runs multi-step literature reviews.

Universities like Oregon State, ETH Zurich and Singapore Management University gives students unlimited Pro searches + 50 Deep Searches monthly.

That’s solid validation.

Runner-up: Scite

The feature velocity is what stands out.

Started the year searching ~150M sources. Sits at 220M+ now.

Added tons of features like Deep Search and Medical Mode.

And most recently a LibKey integration for institutional access

Universities trust it enough to pay for campus-wide licenses now.

Positioning itself as "evidence-first" search worked in its favor and is now the standard for verifying scientific claims.

Hot take, and I don’t like saying it but gotta call spade a spade.

NotebookLM is great for general research.

Not so much for academic research.

It has no in-built citation verification, nor has any academic database access.

It reads your uploaded PDFs and chats about them. That's not research.

Compare to:

50 source limit per notebook is very low for any serious research.

The summary of reviews by the researchers who tried it: "Where are the citations?" "How do I verify this?" "Wait, it only knows what I uploaded?"

Every other tool on this list does maybe one thing well.

Perplexity is the king of the entire research workflow.

Deep Research hits 95% accuracy on SimpleQA in 2-4 minutes. That's the speed and verification covered.

To top it off: Export to PDF, Convert to Perplexity Pages, Share with your team, Connect to Gmail, Notion, Linear, GitHub.

It's Research → Synthesis → Production without switching tools sorted.

Most research tools live in isolation.

You search in Elicit, export citations, paste into your doc, share via email, discuss in Slack.

Perplexity sits inside that workflow.

2025 solved accessibility and speed.

But 2026 will see bigger leaps in my opinion.

Agentic capabilities are getting better.

SciSpace already launched their AI Agent with Deep Review: an agentic system that iteratively searches, synthesises findings, refines understanding, and performs deeper searches based on what it learns.

The question to think about:

Can LLMs generate novel research?

We've consolidated human knowledge into these systems.

Processing capabilities keep scaling.

What happens when you give these agents enough autonomy?

Do they just retrieve existing knowledge faster?

Or do they start connecting dots humans haven't connected yet?

Identifying patterns across disciplines that no single researcher would see?

Proposing hypotheses worth testing?

I don't expect this solved in 2026. It's a decade-scale challenge.

But the building blocks are dropping into place.

Until next time,
Vaibhav 🤝🏻

If you read till here, you might find this interesting

#AD 1

Most AI tools chat. Gladly actually resolves. Returns processed. Tickets routed. Orders tracked. FAQs answered. All while freeing up your team to focus on what matters most — building relationships. See the difference.

Find out how

#AD 2

Udacity helps you build the AI and tech skills employers actually need—fast. Learn from industry experts through hands-on projects designed to mirror real-world work, not just theory.

Whether you’re advancing in your current role or preparing for what’s next, Udacity’s flexible, fully online courses let you learn on your schedule and apply new skills immediately. From AI and machine learning to data, programming, and cloud technologies, you’ll gain practical experience you can show, not just list on a résumé.

Build confidence, stay competitive, and move your career forward with AI and tech skills that are in demand.

Learn More