Key findings
51% of people buying B2B software start research with an AI chatbot more often than with Google. G2, n=1,076.
69% chose a different vendor than planned because a chatbot recommended it. G2.
94% of those who use AI fact-check what it tells them. TrustRadius.
How many software evaluations start with AI?
About half, and rising. G2 surveyed 1,076 people who buy B2B software in March 2026. 51% said they start their research with an AI chatbot more often than with Google. More striking, 69% said they chose a different vendor than they had planned because it was part of the chatbot’s recommendation.
Other studies point the same way. Forrester reported in January 2026 that 94% of business customers use AI somewhere in their buying process, and that 61% use private AI tools supplied by their employer. TrustRadius found in 2026 that 63% used AI to research a software purchase. Responsive reported in October 2025 that generative AI had overtaken traditional search for a quarter of B2B customers.
Do people trust what the AI says?
They check it. In the same TrustRadius study, 94% of those who used AI said they fact-check what it tells them. That matters for what you publish. An answer that names your product sends the reader to your pricing page, your documentation and your reviews to confirm it. If those pages disagree with the answer, the shortlist place is lost at the second step.
6sense reported in 2026, as covered by Demand Gen Report, that 81% use an LLM during evaluations but only 26% use one to discover vendors. Being known before the question is asked still matters. The AI answer decides who stays on the list.
Which sources do the engines read for software questions?
Your product, pricing and security pages. The facts a team checks first.
Documentation and changelog. What the product does today, including integrations and limits.
Review sites. G2 reported, using Profound data, that about a third of the review-site citations in ChatGPT and Google AI Overviews came from G2, and closer to three quarters in Perplexity. That is G2 describing G2, but it shows how much weight review profiles carry.
Reddit and community forums. Semrush tracked 230,000 prompts over 13 weeks in 2025. ChatGPT cited Reddit in close to 60% of responses in early August and around 10% by mid-September. Forum citations are real and volatile, so do not build on one source.
Comparison and alternatives pages. The pages retrieved for versus questions.
What to do first, and sources
Ask the questions your customers ask at each stage: how teams like theirs solve the problem, the best tool for their size and stack, you against a named competitor, and whether you are secure enough and what you cost. Record which vendors are named and which pages are cited. Correct the pages that carry old pricing or missing integrations before you publish anything new. How SourceNode does this for B2B software.
The researcher
Odin Lund is co-founder and COO of SourceNode and leads its data analytics and research, including the daily tracking behind this work across ChatGPT, Gemini and Google AI Overviews. Odin founded First Query, the AI search practice that became SourceNode, and co-founded The Pixel Paper.
Meet the research team
About this research
Published by SourceNode
SourceNode is a data and research-first AI search firm. We record what AI models say about a business every day, trace each claim to its source, and use that data to get businesses named in AI answers. Every figure we publish carries its method and its sample.
Share Blog
Related research
AI search
AI search statistics for 2026: the numbers we trust, with sources
Read the research
AI search
What McKinsey’s AI search research says, and what to do about it
Read the research
