Every article about AI and B2B buying tells the same story. Buyers ask ChatGPT instead of Google. A shortlist forms before anyone visits your website. Your analytics can't see it. You are, the story concludes, invisible.
Most of the individual facts are right. The conclusion is wrong, because the story treats the shortlist as the new thing.
It isn't. Shortlists have been forming before seller contact for years, and we have unusually good data on exactly how that works. What changed is not that buyers decide early. It's what qualifies you to be on the list when they do. And if you optimize for the wrong one of those, you will spend the next year solving a problem you don't have.
Was the shortlist really forming before AI?
Yes, and the numbers are stark. 6sense's 2025 B2B Buyer Experience Report, based on just under 4,000 buyers across North America, EMEA, and APAC, found that buyers purchase from one of the four vendors on their Day One shortlist 95% of the time, up from 85% the year before. Buyers ordered that shortlist by preference before ever speaking to a seller in 94% of cases.
It gets more pointed. The vendor a buyer contacts first wins the deal roughly 80% of the time. 6sense tested whether that's persuasion or preference by isolating the rare buyers who hadn't ranked their shortlist in advance: in those cases the first vendor contacted won only 57% of the time. Being contacted first is a symptom of being preferred, not a cause of it.
So the "shortlists form before you know it" framing that dominates AI-search commentary describes a world that predates ChatGPT entirely.
Then what did AI actually change?
It changed who is eligible.
Before, the only reliable route onto a Day One shortlist was prior familiarity, and 6sense's data shows how closed that system was. Buyers say they already know 75% of the vendors on a typical five-vendor shortlist. 97% know at least one vendor on their shortlist. 85% had direct prior experience with the vendor that ultimately won, and 68% already personally knew a seller there.
Do the arithmetic. Roughly one seat out of five was ever available to a company the buying group didn't already know, and the currency for that seat was recognition, built over years of brand spend and category presence.
AI redistributed that seat. G2's Answer Economy report, a March 2026 survey of 1,076 B2B buyers, found that 51% of software buyers now start research with an AI chatbot more often than with Google, up from 29% in April 2025. The same survey found AI chatbots were the #1 source influencing which vendors make buyer shortlists; G2's June 2026 survey of 1,038 buyers put review sites a point ahead, 38% to 37%.
Then the number that matters most: a third of buyers said an AI chatbot led them to choose a vendor they hadn't considered, and 69% chose a different vendor than they originally planned based on chatbot guidance.
Read those two datasets together and the shift is clear. The pre-AI shortlist was a closed loop of the already-known. The AI-mediated shortlist admits vendors the buyer wasn't considering.
But doesn't 6sense say AI isn't replacing vendor research?
It does, and this is the part most commentary gets wrong in the other direction, including, until I read the full report, the draft of this article.
6sense found that 94% of buyers use LLMs during their journey, but that this has not reduced their interactions with vendors: buyers reported an average of 16 interactions with the winning vendor, statistically unchanged from prior years. LLM usage peaks in the middle of the buying journey, not at the start, because in established categories buyers already know who the vendors are. And in a genuine reversal, the point of first contact moved earlier, from 69% to 61% of the journey, with 58% of buyers saying they engaged sellers sooner to clarify AI-related capabilities or concerns.
So the honest synthesis is not "AI replaced the buying journey." It's narrower and more useful:
In established categories with experienced buyers, AI is a synthesis tool used mid-journey. Familiarity still dominates entry.
For buyers entering a category cold, the ones G2 measured, AI is increasingly the front door, and it will name vendors the buyer wasn't considering.
Both are true. Which one describes your buyer determines what you should actually do, and any article that tells you only one of them is selling something.
There's a second finding in there that should unsettle anyone writing off their website: buyers came to sellers early to clarify AI-related capabilities and concerns, from what an offering includes and costs to data security, and those are questions a vendor's own pages can answer. That's not a case for less published evidence. It's a case for more of the specific, checkable kind.
What does an AI system actually reward?
Not adjectives. Evidence with structure.
The clearest research remains the GEO study (Aggarwal et al., KDD 2024), which tested nine content tactics against GEO-bench, a benchmark of 10,000 queries. The paper reports that its methods can boost source visibility by up to 40% in generative engine responses, and identifies which tactics carry that gain: adding citations, quotations from credible sources, and statistics. Keyword density, the classical SEO signal, showed minimal influence on whether a source got cited.
The detail almost nobody quotes is the one that matters most for a challenger. The gains were not evenly distributed by rank. Lower-ranked sources gained substantially from adding cited sources and statistics; the already-dominant source did not, and in some conditions lost ground.
Evidence optimization is a challenger's tool. It compresses the gap for companies that aren't the default answer, which is precisely the population that was locked out of the Day One shortlist under the familiarity regime.
G2's research points at the same mechanism from the buyer's side: review-site citations were the strongest signal making buyers trust a chatbot's recommendation. Verifiable third-party proof, not brand volume.
There is a cruder prerequisite underneath all of this. Vercel and MERJ's crawler analysis found that ChatGPT's crawler fetched JavaScript files in 11.5% of requests and Claude's in 23.84%, and neither executed them. Their conclusion was that none of the major AI crawlers render JavaScript, with Gemini and AppleBot the exceptions: Gemini inherits Google's infrastructure, and AppleBot uses a browser-based crawler. If your pricing or comparison content is injected client-side, the engines forming the shortlist see an empty shell. You can have the best evidence in your category and still be unreadable.
Does this mean traffic and websites stop mattering?
No, and the pieces claiming otherwise are overcorrecting.
6sense's own read is worth setting against the doomers: where B2B sites are losing traffic to LLMs, the visitors going missing are likely the professionally curious and future buyers, not in-market buyers, who still interact with vendors at the same rate. Declining sessions are a real problem for pipeline creation, but they are not evidence that active buyers stopped engaging.
The measurement consequence is the part worth internalizing. When a buyer forms a view inside a chatbot and never clicks, no analytics platform records it. There is no referral log for the deal that never started. Teams read flat traffic as flat demand, when the consideration decision has moved somewhere their instrumentation cannot reach.
That blind spot is why "we're not seeing AI traffic" is a dangerous conclusion. Most AI influence produces no session at all.
What should a marketing team actually do differently?
Stop asking whether you appear. Start asking whether you're explicable.
The practical version:
Can a machine determine what your company does, who it's for, and what it costs, without inference?
Are your comparisons to obvious alternatives published, honest, and specific, or absent, leaving a model to assemble the comparison from competitors' framing?
Does verifiable third-party evidence exist about your outcomes, or only first-party adjectives?
Do your pages answer the questions that pulled buyers into early sales calls: pricing logic, security, implementation, what your AI actually does? Or do they force a conversation?
None of that is a content-volume problem. Publishing forty more articles into a category where a model can't confidently state what you do will not put you on a shortlist. Making four things unambiguously true, current, and checkable might.
Why we built Trovance
Because I kept hitting the same wall, and the tools built for it stopped exactly where the work started.
I run marketing as a team of one. When AI visibility became a real question, I did what everyone does: I checked. And the checking tools all told me the same thing in slightly different dashboards: here's your score, here's where you're absent, here are the prompts you're losing. Then nothing. No indication of why the answer broke, or whether the fix was a page, a pricing update, a comparison I'd never published, or nothing at all because the buyer was never ours.
That gap is the whole reason Trovance exists. A visibility score is a symptom. It isn't a brief.
Trovance is the evidence and production layer for the agentic web.
It preserves the actual answers and citations across engines, prompts, and repeated runs, rather than a single number that flattens the variance. It diagnoses which specific gap produced the result: identity, relevance, missing proof, weak comparisons, stale facts, thin third-party validation, or a technical retrieval problem your CMS created. It turns the approved gaps into source-backed assets your team can publish. Then it reruns the question and tells you what actually changed.
What it won't do is decide for you.
Trovance inspects, recommends, drafts, and verifies. You own what's true, which buyers matter, what can be public, and what ships. We're not building the autopilot that makes marketers redundant. We're building the thing that makes a small team's judgment faster and better-armed. That boundary is the product, not a disclaimer on it.
And a note on incentives, since they shape what a tool tries to make you do: we're not venture-funded. Nobody is asking us to inflate a metric or ship a feature to tell a fundraising story. If a gap in your visibility isn't worth acting on, Trovance is built to say so. That's a hard thing to build when your growth model depends on every gap looking urgent.
Where this goes next
The funnel that mattered for twenty years was: get found, get clicked, get converted. The one taking shape runs earlier and in a different order (get parsed, get cited, get shortlisted, get chosen) and a growing share of it completes before a human sees your homepage.
The uncomfortable implication is that the asset that matters most is no longer the page that persuades. It's the evidence that survives being read by something that cannot be persuaded.
The comfortable implication, which almost nobody is writing about: this is the first time in a decade that the Day One shortlist has had a seat open on it.
