The shortlist moved somewhere your funnel cannot see.
Nothing about B2B buying got faster or simpler. What changed is where the decisive part happens, and it is no longer on your website.
Forrester's 2026 Buyers' Journey Survey put the adoption figure at ninety-four percent, up from eighty-nine a year earlier. (Forrester, January 2026) That number on its own is easy to dismiss as another AI-usage statistic. The finding underneath it is not: twice as many buyers named generative AI or conversational search their most meaningful source of purchase information than named any other source, ahead of vendor websites, product experts and sales representatives.
Set that beside a finding from the year before. 6sense, surveying roughly four thousand buyers, found that ninety-five percent of the time the winning vendor was already on the buyer's Day One shortlist — the list assembled before formal evaluation began. (6sense, 2025) Nineteen deals in twenty are won by a vendor the buyer had identified before anyone in your company knew there was a deal.
Sixty-eight percent reported generative AI had no impact on their buying process at all.
Adoption arrives. Forrester calls it the arrival of AI-enabled buying.
Adoption saturates, and meaningfulness overtakes every other source.
Nineteen deals in twenty are won by a vendor the buyer had already identified before formal evaluation began. The twentieth is the only one a late entrant wins.
If the Day One list is the deal, and the Day One list is now built inside an answer engine, then the most consequential sales conversation of the cycle happens with a model, without you, and you never see the transcript.
That is the whole argument of this report. Everything below is either evidence for it or a response to it.
What buyers actually do inside the tool.
The usage figure is the least interesting part. The breakdown of what buyers use these systems for is where the strategic problem is.
Four of the five are vendors the group has worked with before. The single open slot is the one answer engines now fill.
All three of those happen before any vendor contact. (Forrester, 2026) The third is the one worth sitting with. Forty-seven percent of buyers are building the internal business case — the document that goes to the committee, the CFO, the security reviewer — inside a tool that assembles it from whatever it can find about you.
If your pricing model is only explained on a call, it is not in the business case. If your security posture lives in a PDF behind a form, it is not in the business case. If a competitor's positioning is more legible than yours, the document that decides your deal was written using their framing of the category.
Eighty-three percent of the buying cycle is now spent researching away from sales reps entirely, and the average buying group evaluates about five vendors while already having prior experience with four of them.
Read that last clause again. Four of the five are already known. The single slot available to a vendor the committee has not worked with before is the slot answer engines now fill.
Your analytics are not broken. They are measuring a smaller building.
The first symptom most teams notice is a traffic decline they cannot explain, arriving at the same time as pipeline that looks unattributed.
Forrester reports B2B companies seeing traffic declines of ten to forty percent as research migrates into answer engines. (Forrester, February 2026) The instinct is to treat this as an SEO problem and brief someone to fix it. It is not an SEO problem. The sessions did not go to a competitor's site. They stopped being sessions.
A buyer who reads a synthesized answer, forms a view, and later types your company name directly into the browser has been influenced entirely outside anything your analytics can attribute. The brand equity is real and the attribution path is gone. This is why so many teams are simultaneously reporting weaker top-of-funnel numbers and unchanged or better close rates: the funnel is not shrinking, the measurable part of it is.
Trackable. The part your dashboard reports on.
Dark. Answer engines, communities, peer conversation, review sites — no session, no source, no campaign.
- Direct traffic, rising without a campaign behind it
- Branded search volume, moving before demand-gen spend
- Inbound requests that already name your product correctly
- Sales calls that open at a later stage than they used to
- Which questions produced the answer you appeared in
- Which competitors appeared beside you, and in what order
- The queries where a category description appeared instead of any vendor
- The business case a committee assembled without ever contacting you
Four factors decide whether you are in the answer.
None of these are published by any platform. They are inferred from independent studies, and the four below are the ones with the strongest evidence and the clearest actions behind them.
The single largest effect size in any study we found, and the cheapest to act on. Publishing eight discrete, checkable facts about your product is a content task, not a platform migration.
Extractable facts, not adjectives
Models reward discrete attributes they can lift and reason over: pricing, integrations, deployment model, compliance posture, named use cases. Brands carrying eight or more structured attributes are cited 4.3 times more often than brands carrying fewer than three.
Erlin, 500+ brands, 2026
Publish the specification, not the promise. Every claim a buyer would need to defend internally should exist somewhere as a discrete, retrievable fact.
Explanation outranks persuasion
The same research found that superlatives work against citation. “Industry-leading” and “most trusted” are unverifiable and get discounted; content that explains how something works gets used.
Erlin, 2026
Rewrite the page a model is most likely to read as documentation rather than a pitch. It is the one rewrite with a measurable return.
Corroboration outside your own domain
Owned content accounts for around half of citations on factual product queries, which means the other half is not yours. Models synthesize across earned media, reviews, communities and third-party comparisons before deciding what to say about you.
Omniscient Digital, analysis of 23,000+ AI citations, 2026
Treat review platforms, analyst listings and comparison sites as part of the data layer. An unlinked mention on an authoritative domain can outweigh a link on a weak one.
Freshness, because the source set moves
The domains an engine draws on shift by roughly 65 percent every two weeks. Visibility is not a state you reach; it is a position you hold.
Wellows, 2026
Measure continuously against the questions your buyers actually ask, not once at the end of a project.
The case against everything above.
A report that only argues one side is marketing. Here is the strongest evidence that this shift is smaller than it sounds, and what we think it actually means.
Gartner: 45 percent used GenAI, primarily to gather vendor and product information.
Forrester: 94 percent used generative AI during their most recent purchase process.
of B2B buyers still turn to a sales representative to validate AI-generated insights
Gartner, 645 buyers, late 2025
information sources used on average across a single buying process
Gartner, 2025
GenAI usage in Gartner's sample, against Forrester's 94 percent — the studies disagree, and by a lot
Gartner, 2025
Gartner's forty-five percent and Forrester's ninety-four are not reconcilable, and anyone presenting either as settled fact is overselling. Sample composition, question wording and what counts as “using AI” differ enough to produce that spread. Treat the direction as established and the magnitude as contested.
The sixty-nine percent finding is the more important one, and it does not weaken the argument. It says AI does not replace the sales conversation — it decides who gets to have one. A buyer who validates with a rep is validating a shortlist that was already drawn.
The honest version of this report's claim is narrow: answer engines are not closing your deals. They are increasingly deciding which vendors are in the room when the deal is closed by people.
Twelve weeks, in the order that matters.
This is the sequence we run. It is ordered by what unblocks the next step, not by what is easiest to sell.
Read the model back
Ask ChatGPT, Claude, Perplexity, Gemini and Google AI Overviews the twenty questions your buyers actually ask. Record who is named, who is cited, and what the engines think you do. Most teams discover a category error before they discover a ranking problem.
Fix the entity, then the facts
Make the company legible first: what you are, who you serve, what you are not. Then publish the discrete attributes a buying committee needs to build its case — pricing model, integrations, security posture, deployment, named outcomes.
Build the comparison surface
The comparison and business-case queries are where the shortlist is set. Publish the comparisons yourself, honestly, including where you are the wrong answer. Models cite documents that resolve a question, and buyers trust vendors who name their own limits.
Instrument it, then hold it
Route AI-referred sessions into the CRM with the original question attached where it can be captured. Set a recurring measurement cadence. Without it you cannot tell an improvement from a model update.
Every step above is measurable except the one that matters most, which is whether a committee you never met put you on a list. That is the honest limit of this work, and the reason to measure the leading indicators weekly rather than the lagging one quarterly.
Every figure in this report, and where it came from.
No figure in this report is Pyxl's own survey data. Where studies disagree, both are shown. Sample sizes are given where the publisher disclosed them.
Approximately 18,000 global business buyers. Source of the 94 percent adoption figure, the 2× meaningfulness finding, the 55/54/47 use-case split and the 10–40 percent traffic decline.
Approximately 4,000 B2B buyers across North America, EMEA and APAC. Source of the 95 percent Day One shortlist finding and the 83 percent self-directed research figure.
645 B2B buyers. Source of the 69 percent who validate AI output with a sales rep, the average of seven information sources, and the 45 percent GenAI usage figure.
Source of the 4.3× structured-attribute citation finding and the explanation-over-persuasion result.
Source of the owned-versus-third-party citation split.
Source of the AI-as-primary-discovery-source finding.
Source of the roughly 65 percent fortnightly change in cited source sets.
Published August 2026 by the Pyxl B2B practice as Volume 02 of the Pyxl AI Visibility Reports. Volume 01, The Cited Catalog Report, covers the same question for eCommerce.Both volumes are here.
Questions we getbefore the first call.
Is this argument only relevant to large enterprises?
No, and the mechanism is worse for mid-market companies. A large brand has enough third-party coverage that an answer engine can describe it without your help. If nobody writes about you, the model builds its answer from whatever it can parse on your own site — which is exactly the layer this volume is about.
We already rank well in search. Does that carry over?
Not reliably. Ranking decides which links appear beneath an answer; being named decides whether you are in the answer. The two are produced by different signals, which is why a page can be first on Google and absent from the recommendation a buyer actually reads.
How do B2B buyers use AI when choosing a vendor?
Ninety-four percent of B2B buyers used generative AI in their last purchase, and forty-seven percent build the internal business case inside an AI tool before contacting any vendor. The practical effect is that most of the evaluation happens before you know the deal exists, with no form fill and no session to attribute it to.
What is the Day One shortlist?
The set of vendors an AI assistant names when a buyer first asks about a category. Ninety-five percent of the time, the vendor that eventually wins was already on it. Getting added later in the process is possible but rare, which is why the list is effectively the deal.
Is B2B website traffic actually declining because of AI?
Forrester measured a ten to forty percent decline in website traffic as vendor research migrates into answer engines. The traffic that does arrive is later-stage and higher-intent, which means the same lead volume now represents a different, shorter funnel.