Blog · 24 September 2026 · By HeardOf

Do B2B buyers research software with AI? Four 2026 buyer surveys, every figure with sample and question

Correction, 25 September 2026: a fifth survey, Deloitte's 38% for agentic AI, was removed from this post with every count that included it, because the deloitte.wsj.com page it came from puts a bot check in front of the article and we do not read past one.

Do B2B buyers use ChatGPT to research software?

Yes, on the two of the four survey pages that name ChatGPT at all, and neither prints a ChatGPT share you can trace to its question. The nearest is G2's 2026 AI Search Insight Report, a survey of 1,076 B2B decision makers in March 2026: on its chart of which chatbot buyers use at each stage, ChatGPT's bar is 73% at discovery, 53% at consideration, 56% at decision and 57% at retention, with Gemini's at 14% to 23%, shares whose base the page does not print, and the page prints no question above that chart. Forrester's blog post of 22 January 2026, historical, says business buyers "are twice as likely to use ChatGPT and four times as likely to use Microsoft Copilot as consumers" and gives no share for either. The figure that circulates, 72% of B2B software buyers using ChatGPT during vendor evaluation, is listed by Crackle PR's statistics page (modified 20 September 2026) and printed in a MarketScale write-up of 20 July 2026, whose earlier version of 12 July says "Nearly three-quarters", each attributing it to a Forrester report and linking to Forrester's home page, and Crackle PR's page citing MarketScale's write-up in return; it is on none of the six Forrester pages we read. Asked plainly, do B2B buyers use ChatGPT to research software? Yes, with no share we could trace to its question on any page we could open on 24 September 2026.

What the surveys do print is whether buyers used AI at all, and that is the table.

How many B2B buyers use AI at all in a purchase?

From 45% (GenAI in a recent purchase, Gartner) to 94% (any AI, Forrester), and they are not the same measurement. Forrester's 94% is of "nearly 18,000 global business buyers" who "report using AI during their buying process", any purchase, any AI, including the private Copilot instance that 36% of them use behind a firewall; Forrester's post of 22 January 2026 says the figure was 89% a year earlier. TrustRadius's 63% is of 1,862 technology buyers who "used AI during their purchase journey" in the year before January 2026. G2's 71% is of 1,076 software decision makers, surveyed March 2026, historical, who "rely on AI chatbots for software research", a narrower tool and a narrower task, with 51% starting there "more often than Google". Gartner's 45%, as Demand Gen Report prints it, is buyers who "used Gen AI" during "a recent purchase", with no sample stated.

Three of the four print a trend. G2 has 71% from "roughly 60% just 7 months ago" and 86% saying they used chatbots more than the year before; Forrester has 94% from 89%; TrustRadius has 54% saying AI made research easier, from 40%, and 47% trusting online resources less than a year earlier, from 39%.

What do buyers use AI for, and do they trust what it says?

To research and compare, and then to check. Forrester's post of 12 February 2026 lists the tasks: researching product information (54%), product comparisons (55%), analysing RFP responses (48%), building a business case (47%). G2's first-ranked use is "Comparing strengths & weakness across vendors"; two-thirds of its respondents' first prompts are category or competitor questions, and four in five say chatbots made the decision faster.

The checking is in every survey that asked. TrustRadius: 94% of AI users fact-check "at least some of the time", the share who do so always or very often rose from 58% to 72% in a year, and 2% trust AI "always". Gartner, via Demand Gen Report: 69% "prefer to validate AI-generated insights with sales reps", and 51% say misleading information is likelier from GenAI against 49% who say a sales rep. G2: 64% meet inaccuracies "often (a few times per month) or very often (weekly or more)", and 45% say a review-site citation is the signal that most raises their confidence in an answer. Forrester's Barbara Winters writes that buyers "compensate by seeking validation from trusted sources: peers, product experts, industry analysts, and others within their buying networks".

One figure has a direction the others lack: G2's 69% who say a chatbot "led them to select a different software vendor than initially planned", and 33% of all respondents bought from a vendor "they'd never previously heard of before". No other page we read prints a figure on whether AI changed the outcome.

Which figures could we trace no further than a write-up?

Five that we could not trace past a home page, and three restatements. The 72% ChatGPT figure above, with 44% for Perplexity "during vendor shortlisting", Bain's "17 AI search queries per buyer per week", and Gartner's median buying committee of "8.2 people" with "17 touchpoints", are on Crackle PR's statistics page and in MarketScale's write-ups of 12 and 20 July 2026, attributed to Forrester, Bain and Gartner reports. On Crackle PR's page as rendered, the 72% links to forrester.com and the 17 queries to bain.com, both home pages, and the other three carry no link; MarketScale's source list links the same three home pages. Crackle PR's page says "Every third-party stat links to the primary source". MarketScale's 12 July write-up carries the 72 only in its address, and its 20 July version prints 72% in its headline.

Machine Relations Research's page of 24 March 2026 (historical; modified 4 July), which its own footer classifies as "Research Synthesis", restates Forrester's 94% with the 18,000 sample and adds that "AI answer engines are now the #1 vendor research source"; Forrester's wording is "twice as many buyers … named generative AI or conversational search as a more meaningful or important source of information than any other source". Apollo's insights pages, undated as served, carry the prior year's 89% "According to Forrester research" and an "80% of B2B tech buyers now use genAI as much as traditional search for vendor research" with no source named beside it; the same figure, for buyers in tech, is in Responsive's release of 15 October 2025 (historical), a survey of 350 B2B buyers. MarketScale's three write-ups of the TrustRadius report (25 July, 31 July and 6 August 2026) add no TrustRadius figure that is not on the two hginsights.com pages.

What do the surveys not tell you?

The question, in three of four; the category, in four of four. Only G2 prints wording beside its charts; Forrester's 94% and TrustRadius's 63% are not one measurement, and without the questions a reader cannot see whether the difference is the word "AI", the word "process", or who was asked. By buyer industry, G2 does; by software category, none of the pages we read gives a figure, so none says whether buyers of fleet software or applicant tracking systems ask a chatbot more or less than the average. None links raw data: Forrester's and Gartner's full reports are for clients, TrustRadius's is behind a form we did not fill in, and G2's is a web page with charts. Three of the four are historical: Forrester's and Gartner's pages are older than 90 days on the day we read them, and so is G2's survey month, on a page that carries no date but was public by 4 May 2026. It is the same test we ran on the category's own studies, in which AI-visibility studies you could check. Of the four, three state a sample and two name a month; of the 33 vendor posts that report their own measurement, 28 and 20 did. None of the four links raw data, and neither did any of the 61 vendor posts.

Vendor-run is a label, not a verdict. TrustRadius and G2 run the review sites their surveys rank among the most influential resources; Forrester and Gartner sell the reports. The test is whether the page shows its sample and its question, and by that test G2's page, run by a review site, is the one that prints both. The same test applies to agencies' price pages: what GEO agencies publish about their prices sets each figure beside its page, its date and whether the page names a source.

What does HeardOf do with this?

Nothing to the surveys, and the other half of the measurement. These four count how many buyers ask an AI; we count what the AI says back when they do: which brands it names and which sites it cites, for buyer prompts in your category, on the four engines we ask, as updated 23 September 2026, our own settings from that post and from our own systems, not from any survey here. The second of the three questions in GEO agency vs SEO agency, whether your buyers use assistants before Google, is the one these surveys answer with a number; the first thing to check after reading them is what those assistants say about you. Whether Google itself shrank while buyers turned to assistants is a separate question, taken source by source in did search volume fall 25% by 2026.

How was this read?

Every page was fetched on 24 September 2026 between 22:24 and 23:09 UTC with a browser user-agent and read in full, and each figure was found in the text of the page it is attributed to, served or, for the two rendered pages, rendered. Demand Gen Report answers 403 to a script, so its Gartner brief was rendered in a headless browser at 22:32 UTC; Crackle PR's statistics table is drawn by JavaScript, so that page was rendered at 22:37 UTC. Every URL came from the sites' own sitemaps, on-site search or links on pages we read; the two search engines we tried returned a picture puzzle or no result links. Not reached: G2's release on PR Newswire (no response), Forrester's on BusinessWire (403), Gartner's newsroom (403) and 6sense's buyer report (403). A sixth 2026 survey reached the same way, Idea Grove's as reported by Demand Gen Report on 18 May 2026, asks 1,000 U.S. consumers rather than business buyers and is not in the table. Six Forrester pages and both Apollo pages link to ChatGPT, Perplexity and, on Apollo, Claude with a prefilled prompt about the company itself ("Ask AI About Working With Forrester", "Ask AI About Apollo"); we did not follow them, and no page carried text addressed to a reader of this kind that bore on the figures. Everything this post says about the surveys carries a link and the day it was read, and you can open any of it; what it says about our own system comes from our own posts and cannot be checked from here.

What GEO agencies publish about their prices: seven agencies' own pages, read 25 September 2026