By Ethora Research
Ethora Research has published The State of Conversational AI, Q3 2026, a free, vendor-neutral report that maps the whole conversational-AI stack using primary data rather than analyst opinion. The report, its charts, and its underlying datasets are free to read and reuse under a Creative Commons licence.
Built entirely from measured sources (GitHub, Hugging Face, package registries, search demand, sampled AI-assistant answers, open-job counts, and talent data) the report documents a field reorganising quickly around a few clear shifts.
Key findings
Agents became the story. US search for “ai agents” grew from roughly 60 searches a month in early 2023 to about 50,000 a month in 2026, close to a 700-fold rise, and now leads the category’s vocabulary. The word shows up in the search data, in the most-active repositories, and directly in hiring.
A new layer standardised in a single year. The Model Context Protocol (“mcp server”) had effectively no search volume before late 2024 and reached tens of thousands of monthly searches within about eighteen months, with its developer libraries already among the most-downloaded packages in the field. It is rare to watch an interoperability layer go from nonexistent to assumed infrastructure this fast.
Self-hosting is gaining ground, at least among developers. The most-starred open-source projects in the space are the ones that let teams run models and chat on their own infrastructure: Ollama, llama.cpp, vLLM, Open WebUI. The report is careful to note this is developer and early-adopter momentum; most enterprises are still adopting cloud-hosted AI, and how far real commercial self-hosting follows is a trend to track rather than a settled outcome.
The category is now answered by AI first. Of 28 buying-intent search terms tracked, 24 trigger a Google AI Overview, and the assistants build their recommendations largely from a handful of third-party “best of” lists rather than from vendors directly, changing how products get discovered.
A clear talent geography. The United States leads on absolute AI-role demand and supply, India is the largest net exporter of practitioners, and by concentration the small, dense hubs lead, with Switzerland and Singapore topping the per-capita tables, a finding the report corroborates against the Stanford AI Index.
A map, not a ranking
The report also introduces a “Periodic Table of Conversational AI”, a capability matrix of eighteen notable systems verified against their current documentation, and a five-layer model of how any conversational-AI product is assembled, from the interface a person meets to the infrastructure it runs on.
Unlike the paid analyst evaluations that rank enterprise vendors for a single use case, this report is deliberately not a vendor ranking. It is a measured map of the whole ecosystem, open source included, aimed at the people building conversational-AI products and anyone tracking where the field is heading.
Every figure traces to a dated source, every table links to its full underlying data, and the datasets and collection scripts are published so any number can be independently reproduced and re-run in future editions.
“The raw model is commoditising, and the value is moving to the edges, up to memory and agents, and down to who controls the infrastructure and the data,” said Taras Filatov, Founder and CEO of Dappros Ltd (Ethora). “We wanted to put real, reproducible numbers behind that shift, and give the whole industry a free, neutral map they can build on and argue with.”
Read the report
- Report: https://ethora.com/research/state-of-conversational-ai-2026-q3/
- Data and scripts (CC BY 4.0): https://github.com/dappros/state-of-conversational-ai
- DOI: https://doi.org/10.5281/zenodo.22131590
