Half of B2B software buyers now start their research with an AI chatbot more often than with Google, and seven in ten say a chatbot led them to pick a different vendor than they had planned. Those are G2's figures from a March 2026 survey of 1,076 B2B software buyers and decision-makers, published on 15 April 2026. The shortlist that used to be assembled by the buyer over a few weeks is now returned by an assistant in a paragraph, with three or four names and a reason for each. For a software company, the question is no longer whether buyers will find you, but whether the pages an assistant reads for your category put you on that list.
This page is for founders and marketing leads at B2B software and developer-tools companies in the US and UK. It sets out what the survey found, what a shortlist is made of now, and what the 45 answers to software-buyer questions in our own October panel cited, so you can check the same places. If the terms are new, start with what generative engine optimisation is.
What G2 Found
G2 is a software review site, so it has an interest in the answer, and it publishes its method: 1,076 B2B software buyers and decision-makers surveyed in March 2026 across North America, EMEA and APAC. With that stated, the findings that matter for a vendor are these:
- 51% of buyers start their software research with an AI chatbot more often than with Google. In G2's 2025 buyer behaviour report the figure was 29%.
- 71% rely on an AI chatbot at some point in the research process.
- 69% said a chatbot led them to select a different vendor than they initially planned, and a third bought from a vendor they had not been familiar with.
- 85% think more highly of a vendor when an AI chatbot includes it in a recommendation.
- 41% use deep-research tools regularly for structured software evaluations.
- ChatGPT was the assistant most used for software research, at 63%.
- Asked what most increases confidence in a chatbot's answer, buyers put a citation from a software review site first. Demand Gen Report's coverage adds that buyers often meet wrong or invented answers, which is why they verify.
Half of B2B software buyers (51%) now start their research with an AI chatbot more often than with Google.
Two cautions before using any of this in a deck. A survey measures what people say they do, and a review site's survey will find that review-site citations matter. Neither makes the figures wrong, but they are a direction, not a rate for your buyers.
What a Shortlist Is Now
A buyer's AI shortlist is the three to six vendor names an assistant returns to a category question, with a sentence of reasoning and, on most assistants, a set of cited pages. The assistant does not test the products. It reads what is already published about the category, weighs it in ways nobody outside the vendor can fully see, and names whoever appears on the pages it trusts for that question. That is why the G2 finding that a third of buyers bought from a vendor they had not heard of is the one to take seriously: the assistant can put an unknown company in front of a buyer, and it can leave a known one out.
It also explains the 69% figure. A buyer who arrives with a vendor in mind asks a comparison question, and the answer reframes the choice around whatever the cited pages say. The vendor who was in the buyer's head but not on the cited pages loses the slot.
What Our Software Answers Cited
We keep a locked panel of buyer questions and ask them every Monday on ChatGPT, Claude, Gemini and Perplexity through their APIs with web search on, and on Google AI Overviews through a search-results data provider. Nine of the questions are the ones a B2B software or developer-tools buyer in the US or UK would ask about AI search visibility and go-to-market help. In the run of 26 October 2026 those nine questions returned 45 answers. ThriveFinity was named in none of them, which we report because the point of the exercise is the list, not us. The sites cited most often across those 45 answers were:
| developers.google.com (Google Search documentation) | 13 answers |
|---|---|
| beomniscient.com (content agency) | 6 answers |
| help.openai.com (OpenAI help pages) | 5 answers |
| developers.openai.com (OpenAI developer docs) | 5 answers |
| derivatex.agency (AI search agency) | 5 answers |
| linkedin.com | 5 answers |
| openpr.com (press release site) | 5 answers |
| siftly.ai (AI visibility tool) | 5 answers |
| learn.g2.com and g2.com | 3 answers |
| techradar.com | 3 answers |
| semrush.com | 3 answers |
Source: ThriveFinity AI Visibility panel v3, weekly run of 26 October 2026, nine software-segment prompts on five assistants, each asked once, 2026. An answer can cite several sites. Gemini's links could not be traced and are not counted. The g2.com and learn.g2.com counts are shown together, each was cited in 3 answers. Chart by ThriveFinity.
Read the list as a buyer would. The assistant answering a software company's question about AI search went first to the platforms' own documentation, then to agencies and tools that publish about the topic, then to LinkedIn, a press-release wire, a review site and the trade press. No vendor homepage is near the top. The shortlist for this category is assembled from documentation, directories and other people's articles. The pattern matched the 19 October run we published in why ChatGPT recommends your competitors, with different names in some slots.
What It Suggests
What we observed. A survey of 1,076 buyers reporting that half start in a chatbot and that the answer changes their choice. Forty-five answers to our software-buyer questions, cited mostly from platform documentation, agencies and tools, LinkedIn, press releases, G2 and trade press.
What it suggests. If the survey is even roughly right about where buyers start, the shortlist is formed before a buyer sees a vendor's website. And if our run is representative of the pages assistants read for software questions, the shortlist is formed from pages most vendors do not control. The G2 trust finding closes the loop: a buyer who sees a review-site citation believes the answer more, which is the review site's own interest but also matches how a careful buyer behaves. These are two sources, one of them ours and small, so treat the conclusion as a working hypothesis.
What to do next. Four checks a software company can run this week, in the order that gave us the most information.
Four Checks for a Software Company
- Ask the category question your buyer asks, on the assistant your buyer uses. G2's respondents used ChatGPT most, so start there, then Perplexity and Google AI Overviews. "Which tools help a developer-tools company get recommended by ChatGPT" is a category question. Record who is named and which pages are cited. The walkthrough is in how to run an AI visibility check.
- Check your review-site presence first. G2 and its peers appeared in our run and were the top trust signal in the survey. A profile with the same description as your site, current pricing and recent reviews is cited more usefully than a profile with a 2023 description. Our AI Presence service starts here, and most of the work is yours to do without us.
- Read the documentation and comparison pages that were cited, and see whether you are in them. For software questions the assistants cited platform docs and third-party comparison articles. If a comparison article for your category names five tools and not yours, that article is the shortlist. Getting into it honestly is slower than publishing your own page, and it is the lever.
- Read what the answer says about you, not only whether you appear. A buyer who verifies, as G2's respondents say they do, will find the page the assistant cited. If that page is stale, the verification fails. The order in which to fix your own pages is in the pages AI assistants cite about your company.
Then repeat the same questions a week later. One answer is one draw. The count over several weeks, with its range, is the measure.
Limits of This Page
- G2's survey is self-reported behaviour from 1,076 respondents recruited by a review site in March 2026. We cite the figures as G2 published them on 15 April 2026 and in Demand Gen Report's coverage of 4 May 2026, checked on 6 November 2026.
- Our panel data is nine software-segment questions asked once each on five assistants in one week, 45 answers. Four assistants were API calls with web search on, not the consumer apps. Gemini's links could not be traced. Rates at this sample size carry wide intervals and are not reported as percentages.
- Our questions are about AI search visibility and go-to-market help for software companies, which is our category. A buyer researching, say, observability tools would see different cited pages, and probably more review sites.
- Errors we find are logged on the errata page.
Spotted something out of date or wrong on this page? Email om@thrivefinity.uk and we respond within 5 working days.
❓ Common Questions
Is the G2 survey representative of my buyers?
Which AI assistant do software buyers use most?
What makes a buyer trust an AI recommendation?
We are a small developer-tools company. Can we get onto an AI shortlist?
Sources
- G2. In the Answer Economy, Don't Win the Click, Win the Answer. 15 April 2026. Survey of 1,076 B2B software buyers, March 2026. Checked 6 November 2026.
- G2. The Answer Economy: How AI Search Is Rewiring B2B Software Buying. Report page, 2026. Checked 6 November 2026.
- Demand Gen Report. Half of B2B Software Buyers Now Start Their Research with AI Chatbots: G2. 4 May 2026. Checked 6 November 2026.
- ThriveFinity. AI Visibility panel v3, weekly run of 26 October 2026, software and developer-tools segment (nine prompts, five assistants, 45 answers). Method on the methodology page.