Research
Where researchers are moving: traffic, discovery, users and revenue of research tools in 2026
In our post on communities, researchers and students were the group with the thinnest public numbers. This post fills that in for the tools they live in: QuillBot and Grammarly for writing, Google Scholar, Elicit and Consensus for finding papers, Zotero, Mendeley and EndNote for references, Overleaf for LaTeX. And it asks where they are going.
The short answer: general AI assistants are taking the searching and the writing, while the tools that hold a researcher’s own work, their library and their manuscript, keep their users and grow plugins around them. Searches for Claude rose nine times in a year and NotebookLM’s more than doubled. QuillBot’s fell by 45%. Overleaf’s held flat while its users doubled in four years.
9×
Google searches for Claude, October 2025 to September 2026
−45%
searches for QuillBot over the same year
20M+
Overleaf users by its own count, up from 10 million in 2022
Where research attention is moving
Brand searches are the clearest public signal of who people reach for. We compare the same month a year apart, so the thesis season is the same on both sides. A change of about a fifth is one step in Google’s rounding and should be read as flat.
Writing tools that rewrite text are losing to assistants that write it. QuillBot fell 45%, Grammarly 33%, Scholarcy 63%. Paperpal, which works inside manuscript submission, held steady. Finding papers is splitting: Consensus and Undermind each rose 83%, while SciSpace and Connected Papers fell a third. Reference managers are steady to down by search, EndNote most of all. Perplexity’s searches fell 70%, which more likely means people open it directly than that they left.
- October 2025
- September 2026
How people find them
Two front doors matter: Google, and the browser. On Google, the writing tools win by far: QuillBot and Grammarly each draw over five million organic visits a month in the US alone. In the browser, the reference managers do: the Zotero Connector has 9 million Chrome users and EndNote Click 7 million, against search traffic in the tens of thousands. A reference manager is found once, then lives in the toolbar.
| Tool | Organic visits, US | Chrome users |
|---|---|---|
| QuillBot | 6.94M | 6M |
| Grammarly | 5.12M | 41M |
| Google Scholar | 3.19M | 3M (Scholar Button) |
| ResearchGate | 2.50M | — |
| Scribbr | 2.34M | — |
| Semantic Scholar | 301k | — |
| Overleaf | 243k | — |
| Paperpal | 140k | 20k |
| SciSpace | 96k | 300k |
| Consensus | 96k | — |
| Elicit | 36k | 30k |
| Zotero | 35k | 9M (Connector) |
| Mendeley | 24k | 3M (Web Importer) |
| EndNote | 15k | 7M (EndNote Click) |
Who uses them, by their own count
Only some companies say how many people use them, and the numbers are their own. Overleaf’s is the one with a history: 10 million users in June 2022, 20 million now.
| Tool | Stated users | Where |
|---|---|---|
| Overleaf | 20 million+ users, 100 million+ documents, 189 countries | overleaf.com/about; 10 million in June 2022 |
| Jenni AI | Over 6 million academics | jenni.ai |
| Elicit | Over 5 million researchers | elicit.com |
| Paperpal | 5 million+ researchers, 1,500+ journals | paperpal.com |
| ResearchRabbit | 1 million+ researchers | researchrabbit.ai |
| Litmaps | 350,000+ researchers | litmaps.com |
Who pays
The large research tools publish no revenue, and most are owned by publishers or information companies: Overleaf and Writefull by Digital Science, Mendeley by Elsevier, EndNote by Clarivate. What is public is small and telling. When scite was bought, its software brought in $3.6 million a year. And the makers whose revenue is verified from their payment provider are mostly selling to students: AI academic writing earns; verifying citations, so far, barely does.
| Tool | Revenue | When | Basis |
|---|---|---|---|
| scite | $3.6M a year in software subscriptions when Research Solutions bought it | Nov 2023 | Research Solutions press release |
| fastwrite.io | $183,900 in 12 months, $13,213 in the last 30 days (AI academic writing) | Oct 2026 | TrustMRR, via Stripe |
| Blainy | $21,511 in 12 months (AI academic writing) | Oct 2026 | TrustMRR, via Stripe |
| ThesisDraft | $1,707 since August 2026 (thesis checking) | Oct 2026 | TrustMRR, via Stripe |
| TrustCite | $130 in 8 months (citation verification) | Oct 2026 | TrustMRR, via Stripe |
Plugins, extensions and marketplaces: where researchers find new tools
Researchers rarely go looking for a new app. New tools reach them inside the ones they already use, and Zotero is the clearest case: its plugin ecosystem is large enough to have its own marketplace plugin. The most-downloaded plugins translate PDFs, take notes on papers and export clean BibTeX for LaTeX, the jobs Zotero itself leaves to others.
Zotero plugins by release downloads (these include automatic update checks, so they rank plugins rather than count people)
Overleaf brings its add-ons in-house: Writefull’s language tools are part of its AI features. The browser is the other marketplace, from Grammarly’s 41 million users to SciSpace’s 300,000. ChatGPT’s app directory lists 18 research apps, among them Consensus, Elicit, SciSpace, scite and Wiley’s Scholar Gateway. Claude reaches research through Anthropic’s life-sciences plugin, which connects PubMed, bioRxiv, ClinicalTrials.gov and Consensus.
Agents are the newest arrivals, and they install their own tools. Zotero’s Python library is downloaded 35,735 times a week, the arXiv MCP server 23,493 times, and searches for an Overleaf MCP server rose fivefold since January.
| Tool | Counted by | Count |
|---|---|---|
| PDF (reference MCP server) | npm downloads a week | 438,200 |
| pyzotero (Zotero's API, used by agents) | PyPI downloads a week | 35,735 |
| arXiv MCP server | PyPI downloads a week | 23,493 |
| Zotero MCP server (54yyyu) | PyPI downloads a week; GitHub stars | 7,099; 5,390 |
| Semantic Scholar MCP | PyPI downloads a week | 4,127 |
| anthropics/skills: pdf, docx | skills.sh installs | 210,542; 201,136 |
| firecrawl research-papers skill | skills.sh installs | 33,337 |
| nature-skills: polishing, citation | skills.sh installs | 16,413; 14,696 |
Where hedwigai plugs in. We built for the same front doors: a Zotero plugin that checks a collection’s references, a free reference checker for a pasted list or .bib file, and a reference-checker skill for agents on skills.sh. Each looks every reference up in Crossref and flags retractions; signed in, a whole draft’s sentences are read against the works they cite.
Who is arriving
Students, in season. Searches for a citation generator rose half again to 1.5 million a month; thesis-writing tools peak every April. Knowledge workers who take research notes: Obsidian’s searches rose half again, and its MCP server draws 12,100 searches a month. Agents, through MCP servers and skills for papers, PDFs and reference libraries.
Where the evidence is thin
No large research tool publishes revenue, and few publish users with a date. Search volumes are rounded, so small moves are noise, and falling searches for a brand people bookmark can mean loyalty rather than loss. Organic visits are an estimate of Google traffic in the US only. We left out user counts reported only second-hand, and aggregator revenue estimates.
Check your references free
Paste a reference list or .bib file and each entry is looked up in Crossref, with retractions flagged. Free, no account.
More head-to-head evals
| Measure | SubQ | jev | Decider | Nimble |
|---|---|---|---|---|
| Found, mean of five columns at section length | 64% | 65% | 66% | 68% |
| Mechanism of action, found at section length | 57% | 51% | 61% | 54% |
| Measure | ask | Apollo | treg |
|---|---|---|---|
| Usable leads | 13 of 40 | 7 of 40 | 6 of 28 |
| Cost per usable lead | $0.037 | $0.27 | $0.23 |
| Measure | ask with jev | ask with Decider |
|---|---|---|
| Questions right, of 92 | 89 | 89 |
| Wrong answers | 1 | 2 |
| Measure | jev | ask |
|---|---|---|
| Our 370 clinical decisions, right | 94.6% | 93.0% |
| PubMedQA, right (best EdgeEvals scored: 62.2%) | 77.0% | 74.4% |
| Measure | ask | Claude Opus 5.5 |
|---|---|---|
| Questions right, of 95 | 59 | 63 |
| Small companies right, of 20 | 13 | 9 |
| Measure | realestate-us | Best of four assistants |
|---|---|---|
| Floor-plan rents right, of 26 | 26 | 23 |
| Amenities claimed the listing doesn't state | 12 | 129 |
| Measure | ask | jev |
|---|---|---|
| Right, of the questions answered | 98% | 79% |
| Wrong answers, of 92 | 2 | 19 |
| Measure | Trained localiser | Best of four frontier models |
|---|---|---|
| Mean overlap with the defect (IoU, 0 to 1) | 0.421 | 0.109 |