How B2B Decision-Makers Buy in the LLM Era

Your pipeline reporting probably still shows a first touch that looks like a form fill or a cold email reply. The decision that mattered happened weeks before that, inside a chat window you cannot see, when a prospect asked a model to name the best options in your category and got back a list of three. If your name was absent from that list, the rest of your outbound sequence was already competing for a spot that had closed.

This is the practical problem facing anyone selling to B2B decision makers in 2026: the research phase moved somewhere you have no visibility and no direct influence, while the sales conversation stayed exactly where it was. At Martal Group, a B2B sales outsourcing agency with 16+ years of running outbound for our clients, we see the downstream effect of this every week in discovery calls that open with a prospect reciting a comparison they never asked us to make.

This guide covers what actually changed in how decision-makers research and choose, what stayed the same, and what a sales team should do differently on Monday morning. It is written for the seller’s side of the table. Most of the published work on this topic tells marketers how to earn citations; far less of it tells a VP of Sales what to say to a buyer who arrives already briefed by a machine.

The LLM Era of B2B Buying at a Glance

1. B2B decision-makers now use AI chatbots to compress vendor discovery, comparison, and shortlist building into a single research session, which moves the decisive stage of the journey earlier and out of a vendor’s view.
2. AI accelerates discovery but does not close the decision: buyers use models for speed and structure, then turn to peers, review sites, and sales reps for validation once the stakes rise.
3. The shortlist is the new battleground, because a vendor absent from the AI’s answer is rarely researched into contention later.
4. Buyers arrive at first calls with higher confidence and a meaningful chance of being wrong, since G2 found 83% feel more confident in their choice while 64% encounter inaccurate AI recommendations often.
5. The seller’s job has shifted from supplying information to validating it, correcting the model’s framing, and giving the buying committee the confidence to commit.
6. Reaching accounts earlier matters more than working them harder, because outreach that lands after the shortlist forms is arguing against a decision the buyer has already made.

What Changed in 2026

• G2’s The Answer Economy report found that 51% of B2B software buyers now start research in an AI chatbot rather than Google, up from 29% in April 2025, with only 3% saying AI has not meaningfully changed their research habits.
• Gartner’s buyer survey established that 69% of buyers use sales reps to validate AI-generated insights, reframing the seller’s role from primary information source to trusted verifier.
• Gartner’s sales survey found that 67% of buyers prefer a rep-free experience and 70% prefer a fully digital, self-service purchase, while buyers who reach value clarity are twice as likely to report a high-quality purchase.
• Forrester’s Buyers’ Journey Survey put AI usage among business buyers at 94%, up from 89% a year earlier, with 61% using private AI tools supplied by their own organization.
• Semrush’s clickstream analysis measured a 206% year-over-year increase in outbound referral traffic from ChatGPT to other websites between January 2025 and January 2026.

Terms Worth Knowing

• AI-generated shortlist is the set of vendors a language model returns in response to a buyer’s category or comparison prompt, which increasingly functions as the buyer’s initial consideration set.
• Zero-click research is buyer research that resolves inside a search or chat interface without the buyer visiting any vendor website.
• Answer engine is any system that returns a synthesized response rather than a list of links, including ChatGPT, Gemini, Perplexity, Claude, and Google’s AI Mode.
• Value clarity is Gartner’s term for a buyer’s confident understanding of how a solution improves outcomes in their specific role and business context.
• Buying committee is the group of stakeholders who influence or approve a B2B purchase, typically spanning several functions and seniority levels.
• Buyer intent signals are behavioral indicators, such as research activity, hiring patterns, or funding events, that suggest an account has entered an active evaluation.
• The 95-5 rule is John Dawes’s observation at the Ehrenberg-Bass Institute that roughly 5% of potential B2B buyers are in-market at any given time while 95% are not.
• Dark funnel is the portion of buyer activity that leaves no trackable signal, including AI chat sessions, private communities, peer conversations, and word of mouth.

What Actually Changed About How B2B Decision-Makers Buy

The research phase changed; the decision phase did not. B2B decision-makers now use language models to do the gathering, comparing, and narrowing that previously required a dozen browser tabs and several hours, but they still validate the result through humans before committing budget.

That distinction matters because it tells you which parts of your go-to-market are exposed and which are safe. The shift looks like this from the seller’s side of the table:

That distinction matters because it tells you which parts of your go-to-market are exposed and which are safe. It also applies well beyond software: Forrester’s analysis of its 2025 Buyers’ Journey Survey put AI usage among business buyers at 94%, up five points year over year, with 61% using private AI tools issued by their own employer. Adoption is not concentrated in one category or one buyer profile.

Research moved from gathering to inference

Buyers used to assemble evidence and draw their own conclusions. They now ask a model for the conclusion and spot-check the evidence behind it. G2’s Chief Innovation Officer Tim Sanders described this as a move from reference to inference, the third compression of the buyer journey after the phone book and the search results page.

The behavioral data supports the framing. According to G2’s March 2026 survey of 1,076 B2B software buyers, 71% now rely on AI chatbots for software research, up from 60% seven months earlier, and comparing vendor strengths and weaknesses is the single most common use case at 41%. Buyers are not asking models what a category is. They are asking which vendor to pick.

Research compressed, but the rest of the cycle did not

A reasonable assumption is that faster research produces faster deals. Published benchmarks disagree, and the disagreement matters if you are about to rebuild a forecast around it.

Some datasets covering 2024 to 2025 show average B2B cycles shortening by roughly a month year over year. Others covering a similar period show cycles lengthening by 20% to 30% since 2021, driven by larger buying committees, CFO sign-off on mid-size purchases, and security review. Both can be accurate, because they measure from different starting points. A benchmark anchored to first touch captures the research phase that AI compressed. A benchmark anchored to opportunity creation starts after that phase has already ended.

For a sales team, the useful split is this. AI compressed the part of the cycle you never had access to anyway, which is the buyer’s independent research. It did nothing to committee alignment, procurement, security review, or budget timing, which is where deals actually stall. The Gartner attention figure above explains why the net effect on your forecast can look like nothing at all: the compressed portion was already happening outside your view.

The practical consequence is that shorter research time means less time for an unknown vendor to become a known one, not faster revenue. Treat cycle-length benchmarks as a measurement question before treating them as a market signal, and check what your own CRM uses as the start date before comparing against any published figure.

The shortlist forms before you know the deal exists

This is the part that hurts pipeline. G2 found that 69% of buyers chose a different vendor than they initially planned based on chatbot guidance, and 33% purchased from a vendor they had never previously heard of.

Read those two numbers together, and you get a genuinely two-sided outcome. Incumbency is worth less than it was, since a preferred vendor can be displaced by an answer the buyer did not go looking for. Obscurity is also worth less, because a vendor with no brand recognition can enter a deal purely by being well represented in the model’s sources.

The consequence for sales is timing. Research from Bain and Google, cited in Google’s B2B marketing guidance, indicates that around 92% of buyers already have a shortlist of preferred vendors before the formal buying process starts. AI has not created that dynamic. It has accelerated it and made the shortlist harder to influence once set.

That is a capacity problem before it is a messaging problem. Covering more accounts earlier in the cycle takes either more headcount or an outsourced sales team already running the motion, and that tradeoff is what most sales leaders are actually weighing when they see this data.

How little of the journey you actually see is worth stating precisely. Gartner’s research on the B2B buying journey finds that buyers spend just 17% of the total purchase journey meeting with potential suppliers, and when they are comparing several vendors at once, any single rep may get only 5% or 6% of that time. That is the origin of the commonly quoted claim that 70% to 80% of the journey is complete before a vendor is contacted. The figure predates AI. What the LLM era changed is not the share, but how much a buyer can accomplish inside it.

What did not change

Three things held steady, and each is a place where sellers still have leverage.

High-stakes decisions still route through people. Buyers do not defend a six-figure purchase to their executive team by citing a chatbot, so peer input, references, and vendor conversations remain the final gate.

Committees still buy. A model can brief one stakeholder efficiently and does nothing to align the other six, which means the internal consensus problem is unchanged and arguably worse now that different stakeholders arrive with different AI-generated framings of the same category.

Fit still decides. Buyers reward vendors who are specific about who they serve well, and language models are poor at judging whether a solution matches a particular tech stack, compliance regime, or team structure.

If you want the underlying sequence in detail, that belongs to a separate discussion of the stages in the B2B buying process. The focus here is the layer AI added on top of it, and what it demands from the people selling.

The Validation Gap: Why Buyers Want Fewer Reps and More Reps at Once

Buyers want to avoid sales reps during research and want sales reps during decision-making, and both preferences are now measurable. Gartner’s data captures the split precisely: 69% of buyers turn to reps to validate AI-generated insights, while a separate survey found 67% prefer a rep-free buying experience.

Sellers who read only one of those numbers draw the wrong conclusion. The first suggests sales is more important than ever; the second suggests sales is being designed out. Both are true at different moments in the same deal.

The two Gartner findings that look contradictory

The reconciliation is about sequence rather than sentiment. Buyers want autonomy while they are learning and support while they are deciding.

Gartner’s survey of 645 B2B buyers, conducted from August through September 2025 and presented in May 2026, found buyers using an average of seven information sources per purchase, with 45% using generative AI among them. Gartner’s Robert Blaisdell framed the implication for sales teams directly: comfort with digital channels does not eliminate the seller, because buyers still rely on reps for reassurance and decision support at specific moments.

The moments Gartner identifies as most dependent on human contact are researching the business problem, identifying a preferred supplier, securing internal support, and finalizing the purchase. Three of those four are late-stage. One is early. The middle of the journey, where most sales teams concentrate their follow-up, is where buyers least want you.

What buyers actually want from a rep now

The value a rep adds has shifted from information to judgment. A buyer who has read a synthesized comparison does not need your feature list recited back at them.

What they cannot get from a model is a specific answer to a specific situation: whether your product works with the version of the ERP they actually run, what happens when their compliance team asks about data residency, how long implementation takes for a team of their size with their constraints. Gartner’s March 2026 sales survey found that buyers who reach value clarity, meaning a confident understanding of how a solution improves outcomes in their particular context, are twice as likely to report a high-quality purchase.

That gives sales a clear target. Your job on the call is to convert a general understanding into a specific one.

What this means for your first call

Open by finding out what the buyer already believes and where it came from. A useful opening sequence looks like this:

“Before I walk through anything, it would help to know what you have already looked at. What did your research turn up about us, and about the two or three others you are considering?”

The answer tells you three things at once: who else is in the deal, what claims you are being measured against, and whether the buyer’s mental model of your category is accurate. Reps who skip this step spend the call answering questions that were already settled and never address the one objection that will actually kill the deal.

The tradeoff worth naming: this approach surfaces competitors you might otherwise never hear about, which some reps find uncomfortable. It is better to hear the comparison in minute three than to lose to it silently in week six.

For a deeper treatment of the conversation mechanics with senior stakeholders, see our guidance on how to sell to decision makers.

The Confidence Problem: AI-Informed Buyers Arrive Certain and Sometimes Mis-Briefed

Buyers now arrive at sales conversations with more confidence and no more accuracy than before, which is a harder combination to sell into than plain ignorance. G2 found that 83% of buyers report feeling more confident in their final choice after AI-assisted research, while 64% encounter inaccurate AI chatbot recommendations often or very often.

A confident buyer working from a flawed premise will argue with you rather than listen to you. Recognizing this pattern early is now a core selling skill.

Where the errors cluster

The inaccuracies are not random. They concentrate in four predictable areas, which makes them easy to check for.

The pattern behind all four is the same: models describe you using the corpus written about you, which is mostly not written by you. Where third-party coverage is thin or stale, the description drifts.

How to correct the framing without telling the buyer they are wrong

Direct contradiction fails here, because the buyer is defending their own research rather than the model’s output. Correcting a formed impression is a well-studied problem in how decision-makers evaluate options, and the practical technique is to give the buyer a reason to revise rather than a reason to defend.

Four questions that surface a bad premise without confrontation:

1. “What did you see on pricing? I want to make sure the number you are working from matches what we would actually quote for a team your size.”
2. “When you compared us against [competitor], what did the comparison say we do differently? I want to check it is current.”
3. “Which use case did the research put us in? We get placed in two different categories and they carry very different price points.”
4. “Was there anything the research flagged as a limitation on our side? I would rather address it now than have it surface in procurement.”

The fourth question does the most work. Buyers told researchers repeatedly that they value vendors who are candid about where they do not fit, and asking for your own weaknesses signals that candor before you have to claim it.

The nuance: sometimes the model is right

Not every AI-surfaced objection is a hallucination. If three different buyers in a quarter tell you the research said the same unflattering thing, that is market feedback rather than a technical glitch. Track what buyers report hearing, because it is the closest thing you have to a live read on how your category positions you.

Where AI Enters Each Stage, and the Seller’s Counter-Move

AI now touches every stage of a B2B purchase, and each stage calls for a different response from the sales team. The table below maps the buyer’s AI-assisted behavior against the counter-move that actually moves a deal, based on what our team sees across client campaigns.

Two rows deserve emphasis.

The shortlist row is the pivot point of the whole table. Everything above it is winnable through visibility and outbound timing. Everything below it is a defensive game played against a preference the buyer has already formed.

The internal consensus row is the one most teams underestimate. When each member of a buying committee runs their own research, they arrive with divergent views assembled from different prompts. Your champion is not just selling your product internally; they are reconciling several different machine-generated summaries of your category. Giving them a single clear document that addresses the finance objection, the security objection, and the operational objection separately is more useful than any deck.

What Outbound Has to Change

Outbound still works in the LLM era, but the timing requirement has tightened considerably. Reaching an account while the shortlist is still forming produces a fundamentally different conversation than reaching it afterward, and the window between those two states has narrowed.

Timing: reach accounts before the shortlist hardens

The practical answer to a compressed research phase is earlier detection. Working from a static list built two quarters ago means most of your outreach lands after the decisive research session has already happened.

This is where buyer intent signals earn their place in a sales motion. Research activity, hiring patterns, technology changes, and funding events indicate that an account has entered an evaluation, which lets you reach them during the window when the shortlist is still open.

Martal’s AI sales platform monitors more than 10 million intent signals continuously, which is the layer that tells our Sales Executives which accounts to work this week rather than which accounts fit the profile in general.

The nuance: intent data tells you an account is active, not that it is winnable. Treat it as a prioritization input rather than a qualification decision.

Multithreading: the committee reads different answers

Single-threaded outbound was always fragile. It is now worse, because your one contact is reconciling their own AI research against several colleagues’ independent research.

Engaging four or five stakeholders across email, LinkedIn, and phone within one coordinated omnichannel motion gives you a chance to shape more than one person’s understanding. In a three-month pilot with Complete EDI, an EDI solutions provider, our team engaged roughly 6,781 prospects per month with a single fractional rep and produced 14 sales-qualified leads within the pilot window, with the first SQLs landing early in the engagement. [verify against the live case study page]

Volume is not the point of that example. The point is that a disciplined qualification standard applied across a committee produces meetings your closers can actually use.

Targeting: the ideal customer profile does more work now

A sharper ideal customer profile matters more when the cost of a wasted touch has risen. If the shortlist window is shorter, spending it on poor-fit accounts is expensive in a way it was not when research phases ran for months.

Tighten the negative profile as much as the positive one. Knowing which accounts to skip is what creates the capacity to reach the right ones early.

What to stop doing

Stop sending sequences that explain your category. Buyers have already had it explained, usually more clearly than a cold email can manage, and a message that starts from first principles signals you are behind their process.

Stop treating a form fill as the start of the journey. By the time someone completes a form, they have usually finished the comparison stage, which means your follow-up should assume knowledge rather than build it.

Stop measuring outbound purely on reply rate. A sequence that generates fewer replies from better-timed accounts will outperform a higher-reply sequence aimed at accounts that already picked someone else.

Teams that lack the headcount to run this motion internally often bring in a sales outsourcing partner to operate it, which is the model we run for clients through our managed programs: onshore Sales Executives handling research, multithreaded engagement, and qualification, with appointment setting delivering the qualified meetings to the client’s closers.

How to Become the Vendor the Model Recommends

Vendors appear in AI answers when credible third-party sources describe them clearly and consistently, not when their own marketing is most polished. This is the uncomfortable part for most marketing teams, because the highest-leverage material is the material they control least.

Third-party validation carries the weight

Review site citations are the single strongest trust signal inside an AI answer. G2 found that 45% of buyers named citations from software review sites as the most confidence-inspiring element of an AI-generated response, and review sites are the only source besides chatbots that gains influence as buyers move deeper into the funnel, rising from 40% at discovery to 47% at retention.

Peer opinion sits alongside it. TrustRadius’s 2026 B2B buyer research found buyers leaning further toward sources they perceive as unbiased, with younger buyers notably more likely to consult Reddit than older cohorts.

The practical implication is that a review program and a community presence are now discovery infrastructure rather than reputation management.

What makes your own content usable

Your website still matters, because it is where buyers verify what the model told them. Google’s research with the National Research Group, surveying 2,063 senior leaders involved in B2B purchasing, found buyers using AI to build an initial vendor list and then running searches to double-check the output.

Content that survives that check shares a few properties. It states specifics rather than adjectives, so pricing, implementation timelines, and integration details are on the page rather than gated behind a form. It says who the product is wrong for, which buyers consistently reward. And it stays consistent across every surface, because contradictions between your site, your review profiles, and your sales conversations are exactly what erodes a buyer’s confidence.

The audit worth running this week

Write the eight prompts your buyers would actually use, then run them across ChatGPT, Gemini, and Perplexity and record what comes back.

1. “What are the best [category] providers for a [company size] [industry] company?”
2. “Compare [your company] and [main competitor].”
3. “What are the drawbacks of [your company]?”
4. “How much does [your company] cost?”
5. “Is [your company] a good fit for [your ICP description]?”
6. “What do customers say about [your company]?”
7. “Who are the alternatives to [main competitor]?”
8. “What should I look for when choosing a [category] provider?”

Log whether you appear, what is said, and which sources are cited. Run it quarterly. The citation list is the more useful output, because it tells you which third-party properties are shaping your category’s answers and therefore where earned coverage is worth pursuing.

How to Measure Something You Cannot See

This is the newest layer of what B2B teams call the dark funnel: every interaction a buyer has with your brand or your category that your analytics and CRM cannot attribute. The older layers are peer conversations, private communities, podcasts, and word of mouth. AI research belongs in the same category and is now probably the largest single contributor, because it happens in a private session that produces no trackable signal and frequently no click.

LLM influence is largely invisible in standard analytics, which is why most teams underestimate it. A buyer who researches in ChatGPT and then types your brand name into Google arrives as branded organic or direct traffic, with no record of what prompted the visit.

Why the influence hides

The referral data understates the effect by design. Only a minority of AI-influenced sessions carry a referrer that identifies their origin, and much of the influence produces no click at all, since the buyer takes the answer and moves on.

Volume is genuinely small and growing quickly. Semrush’s clickstream analysis measured a 206% year-over-year increase in ChatGPT’s outbound referral traffic between January 2025 and January 2026. Treat the current traffic number as a leading indicator rather than a measure of impact.

Three questions worth adding to your process

You will learn more from asking buyers than from your analytics stack. Add these to discovery calls and to inbound qualification, and record the answers as structured fields rather than call notes.

1. “How did you first come across us?”
2. “Did you use any AI tools while researching this? Which ones?”
3. “What did they tell you about us?”

The third question is the valuable one. Answers accumulate into a picture of how your category is being described to buyers, which no visibility tool currently reports as well as your own prospects will. It also feeds directly back into the correction techniques covered earlier.

The reporting nuance worth flagging to your leadership: if AI-influenced deals are landing in your CRM as direct or branded search, your attribution model is crediting the wrong channel, and you will underfund the work that actually created the demand. This is a measurement problem rather than a performance problem, and it is worth naming as such before budget season.

What the 95-5 Rule Means When Models Have Memory

The 95-5 rule states that only about 5% of potential B2B buyers are in-market at any given time, and it becomes more important in the LLM era rather than less. The rule comes from Professor John Dawes at the Ehrenberg-Bass Institute, whose observation is that corporations change service providers roughly once every five years, leaving the large majority of any addressable market out-of-market at any moment.

The Ehrenberg-Bass Institute’s work on the 95:5 rule argues that marketing works by increasing the probability that a brand comes to mind when a buyer eventually enters the market. Language models add a second, parallel mechanism: when the buyer enters the market and asks a model instead of consulting their memory, what surfaces is whatever the corpus contains about your category.

That corpus was written long before the buying window opened. A vendor with three years of substantive third-party coverage, review volume, and credible mentions is well represented in the answer. A vendor that only publishes campaign material during active quarters is not.

The out-of-market 95% is therefore doing double duty now. It is building human memory in the traditional sense, and it is building the source material that will be retrieved on the buyer’s behalf later. Both take time, neither responds to a quarterly push, and both explain why the vendors winning AI-generated shortlists in 2026 are frequently the ones that invested in credibility during 2023 and 2024.

The tradeoff is real and worth stating plainly. This work does not produce pipeline this quarter, which makes it difficult to defend under short-term revenue pressure. Pairing it with outbound that engages the in-market 5% directly is how most teams fund the patience it requires.

Conclusion

The change in how B2B decision-makers buy is narrower than the discourse suggests, and more consequential. Buyers did not stop needing sales. They stopped needing sales for information, moved their research into a system you cannot observe, and arrive at your first conversation with a view already formed and a confidence level that outpaces its accuracy.

The teams handling this well are doing three things. They reach accounts earlier, using intent signals rather than static lists, so they participate in the shortlist rather than contest it. They open sales conversations by diagnosing what the buyer already believes before presenting anything. And they invest in third-party credibility, because that is the material language models actually draw on when describing a category.

None of that requires abandoning the B2B sales motion you already run. It requires moving it earlier and making the first conversation diagnostic rather than informational.

If you want help building the outbound layer that reaches decision-makers while their shortlist is still open, book a consultation.