Best Grok Bots for Research
Research is the job Grok Bots are most often asked to do and the one where you most need to check the answer. This page compares twelve research templates in the directory by what they are good at, what they cost you in setup, and where they should not be trusted on their own.
How we picked
We read each template's instructions, skills and routines as stored in the directory on 2026-10-31. We kept bots with a distinct research method, such as live fact-checking, primary records, claim verdicts, monitoring or cross-model comparison, and dropped ones that only repeated a method already covered. We did not run any of these bots. A good-looking method on a template page does not prove the output is accurate, so verify what comes back.
Pick by situation
- One claim to check, fast: Researchy, or the quick fact-check skill inside Deep Dive.
- A real investigation across many source types: Deep Dive.
- A brief someone else must trust: Ethan, which grades sources and gives claim verdicts.
- Court, spending, lobbying or archive records: Deep Research by Jonah.
- Tables of companies, enrichment or competitor alerts: Deep Researcher.
- A big question with a budget: Swarm.
- What people are saying right now: last30days for a 30-day pulse, Research for a daily sweep.
- A second opinion across models: Multi-model consensus.
- Academic, clinical or patent work: Researcher AI Desk, Curbside Consult or Prior Art Desk.
Want the how-to rather than the list? Read Run deep research with a Grok Bot for choosing a pattern, citing sources, checking claims and scheduling recurring research.
Watch-outs before you add one
- Paid services: Swarm and Deep Researcher use third-party search or research plugins. Check each provider's pricing and set a spending limit first.
- Third-party code and data: last30days installs a skill from GitHub and the social sweep bot uses a data API. Review what they install and what keys they ask for.
- Professional use: the clinical and patent templates are aids. Neither replaces a clinician or an attorney.
Side by side
| Bot | Best at | Plugins | Scheduling |
|---|---|---|---|
| Researchy | Quick live fact-check | None listed | One-off |
| Deep Dive | General deep research with self-check | None listed | Monthly learning report |
| Ethan | Cited brief and claim verdicts | None listed | None |
| Deep Research by Jonah | Primary records | None listed | None |
| Deep Researcher | Reports, enrichment, competitor alerts | Parallel | Monitors you set up |
| Swarm | Budgeted multi-angle report | Exa, Tavily, Firecrawl, Parallel, Hugging Face, X | One-off |
| Research | Social sweeps and source packs | X | Daily sweep, newsletter trigger |
| last30days | Last-30-days pulse | None listed (installs a GitHub skill) | One-off |
| Researcher AI Desk | Academic workflow desk | None listed | None |
| Multi-model consensus | Cross-model second opinion | None listed | One-off |
| Curbside Consult | Clinical evidence and dosing maths | None listed | None |
| Prior Art Desk | Patent overlap map | Tavily | Five weekly or monthly routines |
"None listed" means the template page lists no plugin and relies on the bot's built-in web access.
More on the directory
Browse the research and research digest label hubs for every template tagged that way.
This is a curated shortlist, not a how-to. For routines, see playbooks.
The picks
- 01
Researchy 
Researchy
by Farzad
A research and fact-check desk that runs every pass on the latest Grok model with live web search. For anyone who needs sourced, dated claims instead of a trai…
- Memories5
- Skills1
- Routines0
Why it is on the list
Smallest template here: one skill that launches the latest Grok model through the Grok CLI for a live, web-backed research or fact-check pass and returns dated, sourced claims once the process finishes. Suits a quick "is this true, and what is the current answer" check. Caveats: one skill and no routines, so there is no scheduling, no memory of your standards and no built-in verification step beyond what you ask for. Needs a computer where the Grok CLI can run, which we did not test.
- 02
Deep Dive 
Deep Dive
A deep-research assistant that investigates any question across studies, primary documents, news, and social media, reads sources in full, explains contradicti…
- Memories10
- Skills22
- Routines1
Why it is on the list
The broadest general-purpose researcher in the set. It interviews you once about your standards, searches across studies, primary documents, news and social posts, explains contradictions, states its confidence, and runs a self-check before delivering. Extra skills cover quick fact-checks, reading a study, redoing the maths on statistics, and checking whether a photo, video or quote is genuine. A monthly routine tests whether it is improving. Suits anyone who wants one bot for most questions. Caveats: Over 20 skills is a lot to learn, and answers take longer because of the self-check.
- 03
Ethan 
Ethan
by JUMPERZ
Evidence-first specialist for deep research, claim verification, competitor intelligence, trend monitoring, and research quality audits. Turns a topic, company…
- Memories0
- Skills5
- Routines0
Why it is on the list
Built for decision-ready briefs. It keeps a claim ledger with dates and source grades, can give each claim a PASS, FIX or BLOCK verdict, compares competitors like for like, tracks a topic as a dated change log, and audits an existing research pack. It states that it will not publish, contact anyone, spend, authenticate or treat social posts as proof. Suits teams that need a paper trail and an independent claim check. Caveats: no routines, so recurring monitoring means adding one yourself, and it reports a gap rather than guessing, which can feel terse.
- 04
Deep Research by Jonah Deep Research by Jonah
by Jonah
A research bot that digs through primary sources like court records, laws, filings, spending, lobbying, papers, books and archives, checks them against each ot…
- Memories0
- Skills5
- Routines0
Why it is on the list
Aims at primary records rather than summaries: court filings, laws, spending and lobbying data, papers, books and archives, checked against each other with a citation per claim and a plain note on anything unconfirmed. Includes a source catalogue and a setup that offers ways to unlock more sources. Suits investigative and policy work where a news article is not enough. Caveats: some extra sources may need free access codes or paid services, so read the catalogue before you start, and it has no scheduled routine.
- 05
Deep Researcher 
Deep Researcher
by Lukas Levert
Deep Researcher researches the web for you with Parallel's APIs and brings back answers with sources: cited reports, filled-in spreadsheets, lists of companies…
- Memories3
- Skills6
- Routines0
Why it is on the list
Built around the Parallel plugin. It picks a Parallel research depth by question, writes cited reports, fills spreadsheet columns from a list of companies or people, finds every entity that matches a description, and sets up monitors that alert when a competitor ships or changes pricing. Suits market and competitor research that needs structured output. Caveats: it needs a connected Parallel account, and we did not check whether Parallel charges for its API, so look at their terms first. The monitor and enrichment features run on schedules or batches that use credits.
- 06
Swarm 
A deep research bot. Give it one topic and it fans out across many angles, audits every source, and delivers a cited, conclusion-first report with a spend line…
- Memories12
- Skills2
- Routines0
Why it is on the list
Uses paid tools: set the budget. Swarm splits one topic into many angles, audits every source and delivers one conclusion-first, cited report with a spend line. Its plugin list covers Exa, Tavily, Firecrawl, Parallel, Hugging Face and X, and its setup asks for your budget and tools. Suits a single high-stakes question where you want breadth. Caveats: it is slow by design, paid plugins may bill you, and you should set a firm limit before the first run.
- 07
Research 
A read-only research analyst that turns public conversation on X, Reddit and the web into cited briefs, builds source packs for stories and videos, and fact-ch…
- Memories8
- Skills4
- Routines2
Why it is on the list
Read-only analyst for people who publish. A daily sweep of recent X and Reddit posts on your topics returns ranked, cited picks, a newsletter-issue scan pulls story candidates, and other skills build a sources file that separates confirmed facts from rumour and review a finished video before it goes out. Uses the X plugin and a Scrape Creators API. Suits creators and newsletter writers. Caveats: it relies on a third-party data service, social posts are leads rather than proof, and the daily sweep uses usage every morning.
- 08
last30days 
last30days
by Matt
Research what people actually say about any topic in the last 30 days. Installs the latest last30days skill from GitHub, walks first-run setup (ScrapeCreators…
- Memories8
- Skills1
- Routines0
Why it is on the list
Installs third-party code: read before you run. Researches what people have said about a topic over the last 30 days across Reddit, X, YouTube, TikTok, Hacker News, Polymarket, GitHub and the web. On first use it installs a skill from GitHub, walks through source setup (including a data service with free credits) and checks for a new version every 30 days. Suits trend and sentiment pulses. Caveats: the result is what people said, not what is true, and installing code from outside the directory means you should review it first.
- 09
Multi-model consensus 
Runs Claude Code, Codex, and Grok on the same hard question — one-shot sanity checks or multi-round debate until they agree. Built for forecasts, critiques, PR…
- Memories2
- Skills1
- Routines0
Why it is on the list
Asks Claude Code, Codex and Grok the same question and either takes one pass as a sanity check or runs rounds of debate until they agree. Suits forecasts, critiques, product requirement drafts and strategy decisions where you want a second and third opinion. Caveats: it relies on the other tools being available on the bot's computer, which we did not test, agreement between models is not proof, and it uses usage from more than one product.
- 10
Researcher AI Desk 
Researcher AI Desk
Research Boss ops on Grok: live to-do, morning surface, Lit+Draft seats, grok.com Project paste, optional calendar or grok.com schedule, optional mail read (dr…
- Memories3
- Skills15
- Routines0
Why it is on the list
A research-assistant ops desk for academics: literature triage that is told never to invent citations, paper and chapter outlines, a plain-language critique, a citation checklist, a to-do list, a decision ledger and an approval queue. Optional calendar and mail reading is draft-first and sending needs an exact yes. Suits a researcher who wants workflow rather than a deep-research engine. Caveats: more of a task manager than a source-finder, it mentions routing work to other model tools only when you name them, and the mail option touches your inbox, so leave it off unless you need it.
- 11
Curbside Consult A curbside consult for clinicians. Answers clinical questions with live, cited evidence (PubMed, guidelines, FDA labels), does tested dosing math, and reviews…
- Memories0
- Skills9
- Routines0
Why it is on the list
Clinical use: not a substitute for judgement and never for patient identifiers. Looks up evidence from PubMed, guidelines, FDA labels and trial registries, does dosing arithmetic with a built-in self-test, checks reference lists against Crossref and PubMed, and screens clinical documents for error-prone wording. Suits clinicians and pharmacists. Caveats: the template forbids patient identifiers, you remain responsible for every dose, and a locally loaded protocol overrides general references.
- 12
Prior Art Desk 
Prior Art Desk
named write-up only · overlap ≠ FTO · I do not file. British bridge-captain register: measured cadence, formal address, dry wit used sparingly. Light seasoning…
- Memories8
- Skills1
- Routines5
Why it is on the list
Compares a named invention write-up that you paste in against US patents and dated non-patent or product disclosures, keeping the two evidence lanes separate. Weekly routines report only what changed. Suits an inventor preparing a packet for a patent attorney. Caveats: the template insists the result is an overlap map, never a clearance or freedom-to-operate opinion; it refuses to run without a write-up, and Tavily search needs to be connected. It writes its answers with a light theatrical voice.
Exa
Firecrawl
Hugging Face
XSide by side
| Bot | Labels | Integrations | Routines | Author | Updated |
|---|---|---|---|---|---|
| Researchy | Research-Digest | — | — | Farzad | 30 Aug 2026 |
| Deep Dive | Research-Digest, Research | — | 1 routine
| Warren Wilson | 22 Oct 2026 |
| Ethan | Research-Digest, Research | — | — | JUMPERZ | 28 Aug 2026 |
| Deep Research by Jonah | Research-Digest | — | — | Jonah | 25 Oct 2026 |
| Deep Researcher | Research-Digest, Research | Parallel | — | Lukas Levert | 11 Oct 2026 |
| Swarm | Coding, Finance, Legal, Social-Publishing | Exa, Tavily, Firecrawl, Parallel, Hugging Face, X | — | Josh Birdwell | 27 Oct 2026 |
| Research | News-Briefing, Transcript-Recap, Research-Digest | X | 2 routines
| Trond Eirik Haavarstein | 18 Oct 2026 |
| last30days | News-Briefing, Research-Digest | — | — | Matt | 28 Aug 2026 |
| Multi-model consensus | Research, Product | — | — | Austen | 5 Sept 2026 |
| Researcher AI Desk | Research, Email, Calendar, Chief-of-Staff | — | — | David Anderson | 26 Sept 2026 |
| Curbside Consult | Research, Education, Health | — | — | Cyber_Holler | 28 Sept 2026 |
| Prior Art Desk | Legal, Research-Digest | Tavily | 5 routines
| Stephen Michael Schmitt | 15 Sept 2026 |