Two-thirds of B2B software buyers open their very first AI research prompt with a category query (“best CRM for enterprise”) or a competitor name — not a request for basic education — meaning a vendor’s shortlist odds are largely decided before its own website ever gets a visit.
That’s the headline number from G2’s own 2026 AI Search Insight Report, a survey of 1,076 B2B software buyers fielded in March 2026: 33% of buyers say their first prompt is category-based, 31% is competitor-based, 22% is requirements/process-based, 9% is ecosystem-based, and 6% is budget-based. Two out of three buyers arrive at ChatGPT, Gemini, or Perplexity already naming names. This piece maps what buyers actually type at each stage of the buyer journey — discovery, consideration, decision, retention — building on our earlier look at how AI assistants decide which brands to recommend, and where the data from two different G2 surveys agrees, and where it doesn’t.
The four stages, and what changes at each one
G2’s Answer Economy report breaks buyer-journey source influence into four stages: discovery, consideration, decision, and retention. It also tracked which chatbot buyers lean on at each stage. ChatGPT’s share is highest at discovery (73%) and drops as buyers move deeper into the funnel — consideration (53%), decision (56%), retention (57%) — while Gemini’s share nearly doubles between discovery (14%) and consideration (22%). Claude “more than doubled its share” over the prior seven months, per the same report, concentrated among more technical roles: Engineering/R&D shows Claude at its highest share of any function (12%), versus 3-4% among senior management and finance.
That matters for AEO because it means the AI experience your brand gets judged on is not one model — it’s several, weighted differently by stage and by who inside the buying committee is doing the typing.
Each of those models also picks what to cite differently — see how ChatGPT, Perplexity, and Gemini choose citations differently — so the same underlying prompt can surface a different set of vendors depending on which engine the buyer happened to open.
Discovery: category and competitor prompts, not “what is” prompts
The report’s most consequential single stat may be the first-prompt breakdown: category-based (33%) and competitor-based (31%) prompts together account for 64% of first interactions — compared to just 22% for requirements-based prompts. That inverts the classic top-of-funnel assumption that buyers start broad and educational. They don’t. They start with a shortlist request.
| First AI prompt type | Share of buyers |
|---|---|
| Category-based (“best [category] software for enterprise”) | 33% |
| Competitor-based (names a specific vendor) | 31% |
| Requirements/process-based | 22% |
| Ecosystem-based (integrations, stack fit) | 9% |
| Budget-based | 6% |
Source: G2 2026 AI Search Insight Report, survey of 1,076 B2B buyers, fielded March 2026.
The consequence, per the same report: 54% of buyers cite AI chatbots as the top source influencing which vendors make it onto a shortlist — ahead of software review sites (43%), market research firms (36%), and vendor sites (36%). Being absent from the answer to a category-based prompt is functionally the same as not existing for a third of buyers before a human conversation starts.
Consideration: comparison and validation prompts dominate
Once a shortlist exists, prompts shift from discovery to evaluation. G2 ranked the primary use cases for AI in software research, and “comparing strengths and weaknesses across vendors” ranks first — ahead of basic category learning, initial vendor identification, and use-case validation. Lower on the list but still common: narrowing options to a shortlist, drafting RFP questions from stated business needs, and working through pricing and packaging options. This is a meaningfully more advanced set of prompts than “what is X software” — buyers are asking AI to do comparative analysis, not just retrieval.
Tool choice reflects that seriousness: 41% of buyers use Deep Research tools regularly and only 6% say they’ve never used one; 44% default to “Thinking” or “Reasoning” modes for software research specifically, choosing slower, more thorough output when the decision has weight. That is not casual browsing behavior — it’s closer to a multi-source evaluation report, generated inside a chat window in minutes instead of days.
Decision: where review-site citations start carrying more weight
The report’s citation-share finding, attributed to G2 growth advisor Kevin Indig’s analysis, is one of the more concrete numbers in the AEO literature: review-platform citations in AI answers rise 1.8x — from roughly 7% to 13% — as buyers move from discovery into evaluation. Review sites are described in the report as “the only source besides AI chatbots that gains influence deeper into the funnel.” Vendor sites, market research firms, and peer conversations don’t show the same late-funnel gain.
Separately, 45% of buyers say a citation from a review site is the single most confidence-inspiring signal in an AI answer, rising to 50% among self-identified daily power users of AI chatbots. That trust gap is also a consistency problem: buyers who see ChatGPT, Gemini, and Claude describe the same vendor differently treat the mismatch as a red flag worth digging into further, which is the same failure mode covered in our piece on entity consistency and how AI engines resolve who a brand is.
Where two G2 surveys disagree — and why that’s worth flagging
Two G2 studies published in 2026 measured overlapping questions with different results, and the discrepancy is itself informative. The AI Search Insight Report (1,076 buyers, fielded March 2026) puts AI chatbots as the top shortlist-influencing source at 54%, with review sites second at 43%. The 2026 Buyer Behavior Report (1,038 buyers, fielded June 2026) instead has review sites narrowly ahead at 38% versus AI chatbots at 37%. Both are G2 first-party surveys three months apart with similar sample sizes; the ordering of the top two sources flips. That’s consistent with a fast-moving, still-volatile buyer population rather than a methodology error in either study — but it means any single-survey stat on “the #1 influence source” should be read as a snapshot, not a fixed ranking.
The two reports do agree on directional trend: AI-chatbot usage in software research is climbing quickly (71% of buyers rely on it somewhere in the process, up from roughly 60% seven months earlier), and buyers who source recommendations from AI chatbots are considerably more likely to buy from their resulting shortlist — 80% did so in at least three of their last five purchases, versus 65% of buyers who didn’t use AI chatbots for sourcing, per the Buyer Behavior Report.
Retention: the prompt doesn’t stop after the purchase
Both reports treat retention as a distinct, ongoing prompt-generating stage rather than a one-time event. The Buyer Behavior Report notes that “a better alternative is always one AI query away” and that buyers actively re-run vendor comparisons on tools they’ve already purchased. Chatbot share at the retention stage (ChatGPT 57%, Gemini 19%, per the Answer Economy report) looks close to decision-stage share, suggesting the same comparison-style prompts buyers used pre-purchase continue post-purchase — a renewal risk that AEO strategy has to account for, not just a discovery-stage concern.
What this means for measuring AI-driven prompts
Two consequences follow directly from the prompt data. First, since two-thirds of first prompts already name a category or a competitor, a brand’s AEO priority for discovery is winning comparison and alternatives content, not glossary-style educational pages — a different content set than classic top-of-funnel SEO. Second, since 64% of buyers say they hit AI inaccuracies weekly or more — a rate consistent with what we found tracking why AI chatbots hallucinate about brands — and their default response is to cross-check against peer reviews or another AI tool rather than defaulting to a familiar brand, a thin review footprint costs a vendor confidence precisely at the moment (consideration through decision) when review citations are gaining the most influence. Vendors trying to quantify any of this downstream should pair prompt-level tracking with the referral-traffic methodology in how to measure AI referral traffic, since prompt volume and actual referred sessions are not the same signal.
FAQ
What percentage of B2B buyers start their software research with an AI chatbot instead of Google?
51% of B2B software buyers say they now start research with an AI chatbot more often than Google, according to G2’s 2026 AI Search Insight Report (March 2026 survey, n=1,076) — up from roughly 36% just seven months earlier by the same report’s tracking, and G2’s separate March 2026 press release cites a comparison point of 29% a year earlier.
Do buyers ask AI different questions at different funnel stages?
Yes. First prompts skew category- and competitor-based (64% combined). Once a shortlist exists, prompts shift to comparison (“compare strengths and weaknesses”), use-case validation, RFP drafting, and pricing/packaging questions — a use-case ranking G2 reports directly from its 2026 survey.
Which AI chatbot do buyers use most at each stage?
ChatGPT leads at every stage but its share is highest at discovery (73%) and lower at consideration (53%), decision (56%), and retention (57%), per G2’s Answer Economy report. Gemini’s share roughly doubles from discovery to consideration (14% to 22%), and Claude’s share is highest among technical roles like Engineering/R&D.
Last updated August 31, 2026. This page is refreshed as benchmarks and scores move.