Search & Research

ChatGPT (OpenAI, San Francisco; launched Nov 2022, 800M+ WAU / ~1.2B MAU est. 2026) = the generalist AI surface​ — chat + GPT-5.6 Sol/Terra/Luna routing + Codex agent + ChatGPT Work + Deep Research + Images 2.0 + Voice (GPT-Live-1) + Memory/Projects/Custom GPTs + 60+ connectors. Not Claude (writing/coding-first), not Gemini (Workspace-native), not Perplexity (cite-first). 

Consensus (consensus.app, Montreal/remote; founded 2021 by Eric Olson + Josh Kapla) = the citation-grounded academic answer engine​ — 250M+ peer-reviewed papers (Semantic Scholar + OpenAlex + PubMed + Crossref + publisher-licensed full text), natural-language yes/no questions → synthesized answer + Consensus Meter​ (agree/mixed/contradict ratio) + Study Snapshots + Deep Search agentic lit review. Not a general chatbot (ChatGPT/Claude), not a citation-context tool (Scite), not a systematic-review workbench (Elicit). 

Not shut down.​ Elicit launched 2023, spun out of Oakland nonprofit Ought (CEO Andreas Stuhlmüller), raised $22M Series A Feb 2025 ($33M total), 2M+ researchers, NASA/NIH/CDC/Harvard/Yale/J&J on logo wall. Active 2026-08, independent, no acquisition/rebrand/shutdown. Corpus = 138M+ papers + 545K ClinicalTrials.gov. Category ai-productivity-data; pipeline_stage none. Killer differentiator = sentence-level cited structured extraction into custom columns (sample size / effect size / adverse events) + guided 4-step PRISMA-style Systematic Review workflow (search→screen 5,000→extract→report) + Research Agent that screens and tables without per-paper human reading; not Consensus (yes/no evidence meter, no extraction tables), not Scite (citation support/contradict context only), not Semantic Scholar (discovery graph, no extraction), not ChatGPT Deep Research (web-general, no PRISMA screen, hallucinates DOIs). 2026 trap: free = 2 reports/mo + 2 columns only (real SLR needs Pro $49), recall ~39.5% sensitivity in 2025 PMC study vs ~94.5% human (must supplement PubMed/Scopus/Web of Science), extraction needs human verification on critical arms, biomedical/social-science weighted, humanities/law thin, no web/news source, Plus $12 tier quietly sits between Free and Pro on some pages.

Feedly — AI-Augmented RSS Reader + CTI/CI Intelligence Platform: Consumer News Reader (100→2,500 sources, Leo AI prioritize/summarize/dedupe) + Feedly Threat Intelligence (Real-Time Threat Graph: CVE/malware/TTP/IoC/threat-actor, STIX 2.1, MCP server mcp.feedly.com) + Market Intelligence (AI Feeds, Ask AI, Newsletter Templates); Free 0/Pro6 / Pro+ 8.25/MarketTIStd1,600·mo / Advanced 2,400⋅mo/ThreatTI1,600-3,200·mo (2007 Khodabakchian, Redwood City private) Feedly, Inc., Redwood City CA; founded 2007​ by Edwin Khodabakchian; private independent, no acquisition, no public VC round disclosed​ (Google Reader migration 2013 built its 14M-account base); two surfaces — Consumer News Reader (RSS/Atom + Leo AI) and B2B Feedly Threat Intelligence / Market Intelligence (AI Feeds + Real-Time Threat Graph). Independent private, not acquired-kept​ (reverse Clipdash/Slack/MonkeyLearn), not VC-scale (reverse Supabase/Framer/Miro), not bootstrap (reverse Bannerbear/Todoist), not open-source (reverse FFmpeg).

Not shut down.​ Frase (founded ~2018, Boston-based) is active​ as of 2026-08, repositioned from "AI SEO brief generator" into an agentic SEO+GEO content platform (Frase Agent, 80+ skills, Content Guard, AI Visibility across ChatGPT/Perplexity/Claude/Gemini/Google AI). Old Solo/Basic/Team tiers retired; 2026 sells Starter/Professional/Scale + custom Enterprise. 7-day no-card trial, no perpetual free plan. Category ai-productivity-data; pipeline_stage none. Killer differentiator = folds SERP research + brief + AI draft + dual SEO/GEO score + AI-answer citation tracking + decay-watch auto-revision into one workspace; not Surfer (deeper on-page scoring, weaker AI writing), not Clearscope (enterprise editorial, no agent/auto-publish), not NeuronWriter (cheaper LTD, shallower GEO), not Jasper/Copy.ai (pure AI copy, no SERP optimization). 2026 trap: all tiers meter articles + audit pages + AI-visibility prompts; Starter hard-stops at 10 articles/mo; Rank-Ready one-offs $3.50 each; AI drafts still need human fact-check.

Gemini = the Google-shaped AI surface: open gemini.google.com / iOS / Android / Chrome side-panel → chat in 40+ langs with Gemini 3.6 Flash​ (free default) or bump to 3.1 Pro / 3.7 Flash (via Spark)​ → paste 1M-token PDF/codebase → Deep Research writes cited report → NotebookLM (500 notebooks, 300 sources each, AI Pro+) for source-bound Q&A + Audio Overview → Gemini Live voice interrupt → Gems (custom mini-assistants) + Canvas (prompt→app/quiz/infographic) → Nano Banana 2 image gen/edit in-thread → Veo 3.1 text/image→8s clip with synced dialogue/audio (Lite on Pro, 1080p on Ultra) → Flow filmmaking suite → Workspace: draft in Gmail, summarize Drive, build Slides, analyze Sheets, Meet recaps → Android default assistant (replaces Assistant, reads screen/SMS w/ consent) → Chrome Auto Browse (AI Pro+ multi-step form-fill, US-only Aug 2026) → Spark agent (AI Pro+ 160+ countries, Mariner deeper on $199.99 Ultra) → API via ai.google.dev (3.7 Flash $0.75/$3.75, 3.1 Pro $2/$12 per 1M, >200K doubles). Fits ai-productivity-data; pipeline_stage none (chat→deep research→notebooklm→workspace→spark→api). For Google-stack workers, long-doc/codebase ingestion, video-with-audio gen, Android users — default; for privacy-strict client work, Claude; for agentic coding, Claude/ChatGPT.

NotebookLM = "research microscope, not Swiss Army knife": drop PDF/DOCX/EPUB/Google Drive/Sheets/YouTube URL/text → AI ingests with 1M-token window → ask questions (each sentence inline-linked to source span) → generate Audio Overview (two-host 15-min podcast), Video Overview, Mind Map, Study Guide, Flashcards, Quizzes, Slide Deck (PPTX export), Data Table, Infographic → Deep Research agent blends corpus + web into cited report → save chat history, build source repo from a chat. Fits ai-productivity-data; pipeline_stage none (upload→ground→query→synthesize→export). For students turning lecture PDFs into walk podcasts, analysts grinding 40-source competitive teardowns, lawyers/consultants grounding memos in client docs, and researchers pre-reading before Elicit/Consensus, it's the cleanest grounded layer; for open-ended writing/coding/web-no-corpus, ChatGPT/Claude win.

Iris.ai = the RAG foundation for regulated R&D: connect internal repos + PubMed/Scopus/patents → Axion normalizes to AI-ready chunks → Neuralith builds the retrieval engine → RSpace runs agentic literature reviews (problem→RMap→Focus 10k→100→Analyze extract tables→draft with citations) → agents monitor competitor patents / flag new filings / answer "what do our 200 papers agree on" with page-level traceability → deploy in private cloud/on-prem (Clean Room, no training on customer data) → evaluate + monitor via custom frameworks. Fits ai-productivity-data; pipeline_stage none (ingest→map→filter→extract→synthesize→deploy). For pharma/chemical/materials/agri/energy R&D, USDA/Mercedes/ArcelorMittal/L'Oréal tier procurement, it's the defensible EU-sovereign seat; for a grad student doing one systematic review, Elicit+Semantic Scholar cost 1/200th.

Perplexity = the research stack that starts where search ends: type a question → Sonar model fires dozens of real-time searches → reads full source docs → synthesizes with numbered inline citations you click-verify​ (reasoning chain visible) → follow-up threads keep context → Free 5 Pro/day, Pro $20 unlocks Deep Research (40-source reports ~4 min, 90%+ accuracy), model picker (GPT-5.6 / Claude Sonnet 5 / Sonar 2), file/image upload, image gen, Labs (code/data/asset generation), Pages (shareable formatted reports), Spaces (team curated research desks with custom instructions) → Brain (Max, 15 langs, self-improving wiki memory, +10pp memory benchmark, -6% dissatisfaction) → Perplexity Computer​ (Feb 25 2026, Firecracker microVM, orchestrates 20 models, Slack @computer + web/mobile, enterprise Snowflake/Salesforce/SharePoint/HubSpot/Datadog + SSO/SCIM/audit/SOC2/zero-retention) → Comet browser​ (Chromium, address bar = answer engine, sidecar reads current tab, agent fills forms/compares pricing/cross-tab summary, voice, ad/tracker blocker, Finance Hub, Model Council on Max; free Win/Mac/iOS/Android since Mar 18 2026, Comet Plus $5 publisher content). Fits ai-productivity-data; pipeline_stage none (query→search→synthesize→verify→report→agent-act). For journalists/investors/analysts/policy researchers needing auditable real-time intelligence, it's unmatched; for long-horizon academic PDF synthesis, Gemini 2.5 Pro deeper; for creative/open chat, ChatGPT/Claude.

Qualtrics — Enterprise Experience Management (XM) Platform: CustomerXM + EmployeeXM (People Engage/Lifecycle) + Strategy & Research (Strategic Research/Strategic Brand) + XM Platform Base, Survey Engine + Text iQ (Claude/LLM-grounded NLP) + Stats iQ + Experience Agents + Synthetic Audiences + Press Ganey Forsta Healthcare, Priced per Interaction/Employee/Response, No MCP Server, REST v3 + Webhooks (2002 Smiths, Provo, Silver Lake private 2023, $12.5B) Qualtrics International Inc., Provo UT; founded 2002​ by Scott M. Smith + Ryan Smith + Jared Smith + Stuart Orgill; SAP 8Bacquisition2018,Nasdaq:XMIPOJan2021(SAPretained 7112.5B take-private Jun 28 2023, delisted, Jason Maynard CEO Feb 2026, acquired Press Ganey Forsta May 2026 ($6.75B); ~6,000 staff, 13,500+ orgs, 99% Fortune 100 XM footprint; SOC 2 Type II / ISO 27001 / HIPAA / FedRAMP / GDPR. PE-backed private independent, previously SAP-acquired-kept 2018-2023, now NOT acquired-kept​ (reverse Clipdrop/Slack current-state), not VC-scale (reverse Supabase/Framer/Miro), not bootstrap (reverse Bannerbear/Todoist), not open-source (reverse FFmpeg).

ResearchRabbit is a citation-network literature discovery and mapping engine, not an AI research assistant that reads papers for you. You start with one or more "seed" papers, and it expands outward through six discovery modes — Similar Work, Earlier Work, Later Work, Citations, References, Authors — building an interactive graph of papers, co-authors, citations, and topic clusters. As your collection grows, its recommender learns your direction and surfaces Suggested Papers; you drag useful nodes back into the collection and repeat. The result is a visual "rabbit hole" for backward/forward citation chasing that replaces hours of manual Scopus/Web of Science clicking. It exports to Zotero / BibTeX / RIS / CSV and supports shared collections for collaboration. It does not​ summarise papers, answer questions, or synthesize evidence — that's Elicit's job; and it does not​ classify citation intent (support/contrast) — that's Scite's job.

Scite.ai is a citation-intent verification layer for scientific literature, not a general-purpose research chatbot. Its core unit is the Smart Citation: for any paper it shows not just how many times it was cited, but whether each citing paper supports, contrasts, or merely mentions the original claim, with the exact sentence context pulled from full text. Layered on top are Scite Assistant (grounded Q&A over the corpus), Reference Check (retraction/editorial-notice scan on a manuscript's reference list), citation alerts, dashboards, Zotero + browser-extension badges, and a 2025 MCP server so external AI agents can query citation sentiment programmatically. It replaces "cited-by count" with "cited-how", which is why systematic reviewers, grant writers, and fact-checkers reach for it before trusting a number from Google Scholar.

Headquartered in Seattle, Washington, the Allen Institute for AI (AI2) is an independent 501(c)(3) nonprofit research institute founded on the endowment of late Microsoft co‑founder Paul G. Allen. First launched in 2015 under the AI2 umbrella, Semantic Scholar operates as a public-good AI primitive dedicated to academic literature intelligence. The institute’s funding primarily derives from the Allen estate and major public research grants, including a $2M NSF award in 2022 and a joint $152M NSF/Nvidia research grant secured in August 2025. Institutionally, AI2 is strictly non-VC-backed, non-bootstrapped, non-publicly-traded, and unacquired, and it is not classified as a Google/META platform primitive. As of current benchmarking, Semantic Scholar maintains no official first-party MCP server. All agent integrations rely entirely on community-built MCP bridges — such as the open-source smaniches/semantic-scholar-mcp — that wrap its public REST APIs, with zero vendor-hosted or officially supported MCP infrastructure. This positioning places Semantic Scholar firmly within the no-first-party MCP cluster alongside Motion, Todoist, Qualtrics and Documenso, clearly differentiating it from the first-party MCP cohort including Klaviyo, n8n, Databox, Typeform and Gmail.

Tavily is an Alioth-adjacent AI research tool operating with a NYC/remote-first model, founded in 2023 by Rotem Weiss and Assaf Elovic, spun out of the open-source GPT Researcher project. The company completed successive financing rounds including a mid-2024 seed round and a $20M Series A led by Insight Partners and AlphaWave in August 2025, bringing total capital raised to $25M. In February 2026, Tavily was fully acquired by Nebius Group for a base value of $275M, with potential milestone-based uplifts reaching up to $400M. Post-acquisition, the product maintains independent operation under the original Tavily brand within Nebius AI Cloud; it retains full business continuity and is not absorbed or sunset, distinguishing it from sunset-acquired tools such as MonkeyLearn. Tavily offers an official, stable first-party MCP server hosted at mcp.tavily.com/mcp, complemented by an open-source MIT-licensed repository tavily-ai/mcp-server-tavily with over 1.2K GitHub stars. Its MCP tool suite covers core web intelligence capabilities including search, extract, map, crawl, and end-to-end research generation. Ecosystem-wise, Tavily belongs to the stable first-party GA server cluster alongside Klaviyo, Feedly, Slack and Databox. It is clearly differentiated from n8n dual-GA architecture, Google Sheets/Gmail first-party preview stack, Typeform beta-grade MCP, Nanonets Composio-exclusive integration model, and the zero-first-party MCP cohort consisting of Documenso, QuickChart and Semantic Scholar.

Wolfram|Alpha computes answers rather than retrieving prose. You type natural language ("integrate x^2 dx", "population of France / GDP per capita", "Mars opposition 2026"), it disambiguates via its NLU system, executes against the Wolfram Language engine + curated Wolfram Knowledgebase (20+ domains, real-time feeds for weather/stocks/astronomy), and returns pods of exact results, derivations, plots, and source-attributed data. In the LLM era its role is the deterministic compute+fact layer behind probabilistic models: the model formulates a query, Wolfram executes it exactly, and the result is woven back into fluent text. Stephen Wolfram called the combined system "the first truly large-scale statistical + symbolic AI system" — the neural net handles language, Wolfram handles anything that must be exact, computed, looked up, or visualised.

Zotero project of Corporation for Digital Scholarship (CDS), Vienna VA nonprofit; originated 2006 Roy Rosenzweig Center for History & New Media, George Mason Univ, Mellon/Sloan/IMLS seed; nonprofit-independent public-good​ (CDS 501(c)(3)-style nonprofit, past grant support from Mellon/CZI but CZI not owner — unlike Semantic Scholar/AI2 which is AI2-owned; Zotero owned by CDS, CZI was grantee-funder not parent) — joins Semantic Scholar nonprofit-independent bucket but distinct: CDS is single-purpose reference-manager nonprofit, AI2 is multi-mission AI institute; no first-party MCP server​ — community MCP bridges (54yyyu/zotero-mcp ~stars, kujenga/zotero-mcp MIT, danielostrow/zotero-mcp-server, jooyoungseo/zotero-translator-mcp wrapping translation-server, PapersGPT plugin) wrap Web API / local API / translation-server → aligns Documenso/QuickChart/Semantic Scholar none-cluster, distinct from Klaviyo/n8n/Databox/Tavily first-party.