Method note: the buyer-behavior figures come from G2's 2026 research (1,076 B2B software buyers) and 6sense's buyer survey (about 4,000 buyers). Both platforms sell to software companies, so we label them vendor research. The citation-structure figures come from Parse's 5.17M-link study. Interpretations that go beyond what a study says are labeled as ours.

Three numbers from G2's 2026 research explain a deal loss every software vendor knows. 69% of buyers said AI answers changed which vendor they chose. One in three said they bought a product they had never heard of before they started researching. And 51% now start software research with an AI chatbot more often than with Google, up from 29% a year earlier.

Put together: the market leader coasts on brand memory while a challenger the buyer had never heard of enters the shortlist through an answer, and wins. The loss happened before any sales conversation took place.

Why AI names vendors nobody has heard of

An answer engine does not rank the market. It composes an answer from citable evidence. The Parse study of 5.17 million cited links puts B2B software citations at 46-52% general web, 25-29% the vendor's own site, and 10-13% community forums. Meanwhile 6sense's survey (about 4,000 B2B buyers) found 94% use LLMs somewhere in their purchase process, up from 89% in 2024.

Composition rewards concentrated evidence over market share. A challenger that has published honest comparison pages ("X vs Y", "alternatives to Z"), a series of explainers for one narrow use case, and visible answers in the communities where its category is discussed, hands the engine a specific and quotable record for that slice of the market. The incumbent's generic category page often has less to quote. This is our reading of the citation data, not a claim the studies themselves make.

The pattern in the published record

Zapier's "best apps" articles came to rank for more than 400,000 keywords, and the site roughly quadrupled its organic traffic in three years, from 1.19 million to about 5 million monthly visits (salt.agency's 2024 analysis). Auth0 had engineers write its deep technical content, which took backlinks from 200,000 to 4.4 million in two years and built a readership of 700,000 developers a month (gracker.ai's third-party analysis of Auth0's program).

Both playbooks predate AI answers. Both produced exactly the kind of specific, task-level written record that current citation studies find in AI answers: lists aimed at one persona, comparisons that name competitors, technical depth authored by people who do the work.

What this means for incumbents

The same mechanics cut the other way. Absence from answers is not neutral: with 69% of buyers reporting that AI changed their vendor choice, absence means your brand was weighed and passed over, or never weighed at all. Answers also persist. They repeat whatever evidence exists until someone publishes something better, which means the incumbent that never wrote the "us versus them" pages has left its own comparison layer to whoever shows up there.

Buyer questions that reveal this

Illustrative examples, phrased the way buyers phrase software searches:

  • "Vanta vs Drata for a 50-person startup"
  • "Cheapest CRM for a 10-person sales team"
  • "Alternatives to [market leader] for SMB budgets"

Run your category's real buyer questions against ChatGPT, Perplexity, and Gemini and record who gets named, including the vendors you have never heard of. Or start with CiteGap's free scan (5 questions, 1 engine). For the full SaaS playbook, see AI visibility for SaaS companies. For why a mention inside the answer beats a cited link, read why being named by AI beats ranking on Google.

Get the written diagnosis ($490)

Frequently asked questions

Do buyers really buy software they have never heard of?

One in three respondents in G2's 2026 research (1,076 B2B software buyers) said they purchased a product they had never heard of before researching it, and 69% said AI answers changed their vendor choice. G2 sells to software companies, so the figures carry vendor-survey caveats, but the sample is large and specific to software buying.

Why would AI recommend a smaller vendor over the market leader?

Because the answer is composed from citable evidence, not from market share. A smaller vendor with specific comparison pages, task-level explainers, and community presence offers more to quote than an incumbent whose public record is a generic category page. This is our reading of the Parse citation data; no study measures the effect directly.

Does this apply in categories with two dominant players?

The mechanics are the same, but the contested ground shifts to the alternatives layer: questions like "alternatives to [leader]" or "[leader] vs [challenger]". Whichever vendor publishes the most honest, specific material for those questions tends to control what engines say. Our incident management audit found a category where most vendors had published no comparison pages at all.

How do I know whether AI names my product?

Run your real buyer questions across ChatGPT, Perplexity, and Gemini and record every answer: who is named, in what order, with what description. CiteGap's free scan runs 5 questions on 1 engine as a first pass, and the written diagnosis tests 30 buyer questions across all three engines.

Sources