The Invisible Shortlist

One number that changes everything

In 2025, 29 percent of B2B buyers started their vendor research with an AI tool. One year later, that number is 51 percent — and still rising.

G2’s Answer Economy study (March 2026) surveyed 1,076 B2B software buyers and decision-makers: 51 percent now start their research with an AI chatbot more often than with Google. G2’s 2026 Buyer Behavior Report (June 2026, n=1,038) adds the outcome side: buyers who use AI for recommendations are markedly more likely to buy from their initial shortlist — 80 percent did so in at least three of their last five purchases, against 65 percent of non-AI users — and AI chatbots (37 percent) now sit level with review sites (38 percent) as the top source shaping that shortlist. Bain & Company’s survey of 1,500 US online buyers (fielded September 2025, published April 2026) corroborates the starting point: 44 percent mostly start in an LLM or split their search between AI tools and search engines.

This is not a prediction. It is not a trend to prepare for. It has already happened to your pipeline — you just don’t have a metric for it yet.

What takes place before the first visit

The conventional model of the B2B buying journey assumed that discovery happened through search, outreach, or referral — and that the website was where evaluation began. Awareness led to intent, intent led to a visit, a visit led to a lead.

That model has a new first chapter. Before your website enters the picture, a buyer is now likely to have already:

• Named the category of problem they need to solve
• Asked an AI for a list of vendors who address that problem
• Received a synthesized answer framing two to five companies as the relevant options
• Formed an initial preference based on how those companies were described

Only then, if your name appeared, does the website visit happen. If it didn’t appear, the journey continues without you.

The G2 data captures what this means at the outcome level. 69 percent of buyers who used an AI tool during vendor research ended up switching from the vendor they had originally intended to contact. And 33 percent of completed purchases involved a brand the buyer had no awareness of before the AI recommended them.

The clicks that stopped coming

It is tempting to look at traffic data and conclude that nothing fundamental has changed. Sessions are holding steady. Search still drives the majority of referrals. The dashboards look familiar.

But the dashboards are measuring arrival. They cannot measure absence.

Three years of longitudinal data from 69 German-speaking websites tells the structural story: between 2024 and 2026, Google’s share of total sessions fell from 40.8 percent to 21.9 percent — cut in half. That is not a rounding error. That is a structural reallocation of attention.

The traffic is not merely shifting. For many queries, it is not arriving at all. SISTRIX’s analysis of the German market quantifies one mechanism: AI Overviews now suppress the click-through rate at position one by 60 percent. Ranking first no longer guarantees the visit it used to.

What the data points at, collectively, is a buyer population that is researching more thoroughly and clicking less — because they are getting more of what they need before they leave the search environment. The research is happening. You just aren’t seeing it.

The measurement problem

B2B marketing teams are sophisticated about the signals they can measure. They instrument conversion paths, build attribution models, study session recordings, and optimize toward MQL quality.

None of those instruments measure what AI systems say about a company in the millions of conversations that happen before a buyer considers visiting a website.

There is no Search Console for AI responses — the two first-party reports that appeared in 2026 show impressions in Google’s AI features and citations in Copilot and Bing, and nothing from ChatGPT, Claude or Perplexity. There is no analytics segment that captures the buyer who asked three AI tools about your category, received an answer that didn’t mention you, and updated their shortlist accordingly. The pipeline entry that never happened leaves no trace in any system you own.

The size of that absence has now been measured from outside. In a 2026 test of 175 brands across eight AI platforms, 96 percent were described accurately when the AI was asked about them by name — and 89 percent never appeared when the AI was asked about their category. The systems know the companies. They just do not bring them up.

AI visibility is not search visibility

The instinct, when confronted with a new channel, is to apply the playbook that worked for the last one. For search, the playbook was SEO: technical structure, keywords, backlinks, authority. It took years to mature. It is reasonably well understood. The instinct for AI visibility is to ask the same question: how do we rank?

When a buyer asks an AI tool to recommend vendors for a complex, high-consideration product, the tool does not return a list ordered by domain authority. It synthesizes an answer from the sources it has retrieved and the knowledge it was trained on — weighting for credibility, specificity, and fit to the question. The companies that appear in that answer are the ones the system has enough structured, credible, relevant information about to include with confidence.

Getting into an AI-synthesized answer requires that the right content exists, in a form that AI systems can process and trust, in sources that AI systems retrieve and weight. That is a different discipline from search optimization — though not a separate one: Google’s AI features draw on its core ranking systems, so a page that does not rank is not retrieved either. Search is the entry ticket; it is no longer the game. The discipline is still being defined, and the gap between early movers and late movers is widening every quarter. This is the work of Generative Engine Optimization.

When they do click

The largest study of what AI-referred visitors do on arrival is more sober than the hype. Twelve months of first-party data from 973 e-commerce sites (Kaiser & Schulze, Marketing Science, 2026) show ChatGPT referrals making up under 0.2 percent of traffic and converting worse than every established channel — organic and paid search, email, affiliate — except paid social. Two findings point the other way: conversion rose steadily over the year, and for information-intensive categories, where buyers need guidance before they commit, the authors already see ChatGPT as a useful complement.

For B2B, that second finding is the relevant one. A buyer who arrives after an AI has explained the category and named your company may well be further along than one who arrived from a keyword click — evaluating, not browsing. That is a plausible reading, not yet a measured B2B result.

What is measured is the step before: whether you are on the shortlist at all. That is where AI visibility decides.

The diagnostic question

Most B2B companies have not yet asked what AI systems say about them — not in any systematic way. This is not negligence. The tooling to measure it at scale is new. The awareness that it matters is spreading but uneven.

What comes next

AI’s share of the research journey will not retreat. The buyers using these tools are not going back to ten blue links for complex decisions. The behavior shift captured in the G2 and Bain data reflects a preference that has already been formed and is being reinforced with every useful answer a buyer receives.

The companies that build systematic AI visibility now — that understand how they are represented, in which contexts, against which competitors, with what accuracy — will hold a compounding advantage that grows harder to close as the channel matures.

The companies that wait for the channel to become measurable in the same way that search is measurable may find, when the numbers finally appear, that the shortlist was written without them.