Where Software Research Now Starts

A forecast about search volume has been widely repeated for two years, and the change that has actually been measured is a different one. Surveys of B2B software buyers over the past eighteen months record a shift not in how much searching happens but in where a purchase enquiry begins. The opening move has moved to a surface that produces no impression data, no query record and no ranking, and that introduces vendors the buyer had not previously heard of. This report sets out what the surveys measured and what follows from it for an organisation being described on that surface.

The forecast and the measurement

Gartner predicted in February 2024, in a press release attributed to VP Analyst Alan Antin and sourced to a client note, that by 2026 traditional search engine volume will drop 25%, with search marketing losing market share. This is a forecast about the year now in progress, not an observation, and it rests on no published survey or measurement. It is quoted here only because it is the claim most often mistaken for evidence in this area.

What is evidenced is narrower and more useful. G2, which operates a software review marketplace and therefore has an interest in how buyers research, surveyed 1,076 B2B software buyers and decision-makers in March 2026 and reported that 51% now begin their software research with an AI chatbot more often than with Google, up from 29% in April 2025. The same survey found that 71% rely on AI chatbots for software research, compared with 60% seven months earlier. The measured change is in the first step and in overall use, not in the volume of searching, which these surveys do not address.

The commercially consequential findings

Two results from that survey matter more than the adoption figures. G2 reports that 69% of buyers chose a different software vendor than initially planned based on AI chatbot guidance, and that one in three, 33%, purchased from a vendor they weren’t familiar with. The second is the one with structural implications. An assistant is not only reordering a set the buyer already held; in a third of cases it is adding a name to it. Entry into consideration is being decided by whatever the assistant has to say about a vendor, and a vendor with nothing substantive said about it anywhere cannot be the beneficiary of that mechanism.

Buyers also report the process improving. The same survey found that four out of five say chatbots accelerated their purchasing decision and that 83% report feeling more confident in their final choice. Confidence is self-reported and is not evidence that the decisions are better ones, but it does indicate that the behaviour is unlikely to reverse.

A second, independent survey supports the direction while placing the level lower. Gartner surveyed 645 B2B buyers between August and September 2025 and found that 45% said they used generative AI, primarily to gather information on vendors and products, within an average of seven information sources used during a recent purchase. The gap between 45% and G2’s 71% is partly a difference in field dates and partly a difference in population, and neither figure should be read as correcting the other.

A third survey should not be read as a continuation of the first. G2’s later Buyer Behaviour Report, covering 1,038 decision-makers surveyed in June 2026, found that 82% of buyers sourced software recommendations from an AI chatbot in the last 24 months. This is a different question over a different window and does not extend the 51% figure into a trend line. The same report found review sites at 38% against AI chatbots at 37% as the top influence on vendor shortlists, which places the assistant alongside the existing intermediaries rather than above them.

The assistant is not the last step

The surveys are equally clear that buyers do not treat the output as settled. Gartner found that 69% of B2B buyers prefer to validate AI-generated insights with sales representatives, and that 51% say they are more likely to encounter misleading information from generative AI, against 49% who say the same of a sales representative — a near-tie that is itself the finding.

Checking is close to universal among users. TrustRadius and HG Insights, surveying 1,862 technology buyers and 444 vendors globally in January 2026, found that 63% used AI during the purchase journey and that 94% of buyers who used AI said they fact-check its responses at least some of the time.

What this means for an organisation

The two halves of this evidence are usually reported separately and belong together. Research now opens on a surface where the vendor has no visibility, and it then moves to a corroborating step where the buyer checks what the assistant said. That combination makes the independent record about a vendor more consequential than it was, not less. The assistant’s description is the first impression, and the material a buyer finds when verifying it is what determines whether that impression survives.

An organisation cannot inspect the surface where this begins. There is no query log, no impression count and no position to report. What it can inspect is the corroborating layer: the review listings, third-party coverage, documentation and directory entries a buyer reaches at the verification step, and against which the assistant’s own account is checked. Where those sources are absent, thin or inconsistent with each other, the verification step resolves against the vendor whether or not the assistant named it favourably.

The one finding to take literally is the 33%. Unfamiliar vendors are entering purchases they would not previously have entered, on the strength of a description assembled from public material. That is an opportunity only for organisations about which enough correct public material exists to assemble one.