In March 2026, Gartner published a survey of 646 B2B buyers with a headline that got quoted everywhere: 67% prefer a rep-free buying experience — up from 61% the year before.

Two months later, the same survey wave produced a second headline that got quoted almost nowhere: 69% of those buyers turn to a sales rep to validate AI-generated insights.

Read side by side, those look like a contradiction. They aren’t. They’re the most accurate description available of how people actually buy now, and if you run a site where buyers arrive with requirements — a vertical directory, a buyer’s guide, a spec-driven manufacturer’s technical library — they describe your problem precisely.

The paradox isn’t a contradiction

A buyer who says “I’d rather not talk to a rep” is not saying “I don’t want help.”

They’re saying they don’t want your help on your schedule, filtered through someone whose commission depends on the answer. Same survey: 70% prefer a completely digital, self-service buying experience, and buyers consulted an average of seven information sources on a recent purchase.

But self-directed research doesn’t resolve a decision. It produces a pile of plausible, mutually contradictory claims. Gartner has been measuring the consequence of that for years: in research across more than 1,000 B2B customers, nine out of ten buyers said the information they encountered was high quality — the problem was never bad information, it was too much good information pointing in different directions. Sellers who helped customers make sense of that pile, rather than adding to it, closed high-quality, low-regret deals 80% of the time.

So the 69% figure isn’t buyers rejecting self-service. It’s buyers hitting the wall at the end of it and reaching for something authoritative to check their conclusion against.

Right now, the only thing on the other side of that wall is a human being and a calendar invite.

Meanwhile, the decision is happening before you know about it

6sense surveyed roughly 4,000 buyers for its 2025 Buyer Experience Report, median deal size $200–300K. The findings that matter here:

The shortlist is the whole game, and it gets built while nobody is watching.

Add the layer that’s arrived since: Forrester’s 2026 Buyers’ Journey Survey of 18,000 global business buyers found 94% used AI during their most recent purchase — 55% comparing vendors inside AI tools, 54% researching products, 47% building the internal business case, all before contacting a vendor. Forrester’s analysts report companies seeing 10–40% traffic declines as research migrates into answer engines, and advise that a B2B site’s job is shifting “from a traffic generation and lead harvesting engine to an information syndication engine.”

So: the shortlist forms before contact, and increasingly it forms inside a tool that has never seen your catalog.

Why this lands hardest on directories and spec-heavy manufacturers

If you operate a vertical directory or a buyer’s guide, your entire business model is the shortlist. Listing fees, membership fees, routed inquiries, placement — all of it is priced off your ability to be the place where the shortlist gets built. That is the exact moment the research above says is moving away from you.

If you manufacture something specified rather than merely purchased, the equivalent moment is the spec. An engineer or architect decides whether your product clears a code requirement, a dimension, a compatibility constraint — and they decide it from documentation, at 11pm, without submitting an RFI. If your data sheet doesn’t answer it and a competitor’s does, you lost the job before anyone at your company knew it existed.

Both businesses have the same asset and the same problem. The asset is a genuinely authoritative, structured body of information — vendor profiles, feature matrices, spec sheets, compliance certifications — that a general-purpose AI does not have and cannot reliably reconstruct. The problem is that the asset is sitting behind a filter panel and a contact form.

A filter panel was never guidance

Here is the thing worth being blunt about: faceted search is not the self-service version of your advisor. It never was.

A filter panel requires the buyer to already know which attributes matter, what the acceptable range is on each, and how to trade them off. That’s the expertise they came to you to borrow. Handing them checkboxes and calling it self-service is like answering “which of these is right for me?” with “what are your filtering criteria?”

That’s why almost every serious directory eventually grows a human advisory layer — a “talk to an advisor” line, a recommendation form somebody fills in by hand, a person who quietly fields “which one should I pick” emails all week. It exists because the filters didn’t work. It’s also capacity-capped, business-hours-only, and the single most expensive thing on the site.

The manufacturer version is an applications engineer answering, for the fourth time this month, a compatibility question already documented on page 11 of a PDF.

Both are the same failure: the guidance exists, it’s correct, and it’s only available synchronously, from a person, during business hours.

What the research actually recommends

Gartner’s own advice to sellers in the March release is worth reading literally:

Deploy buyer- and seller-facing AI agents supporting both self-directed validation and value articulation. Structure content into modular, agent-ready blocks for dynamic assembly.

“Self-directed validation” is the 69% statistic turned into an instruction. The bet is that the validation step — currently a phone call — can be served by something grounded in your own data, available at the moment the buyer hits the wall.

And the payoff metric from the same survey is the one to keep: buyers who achieve value clarity — who understand how a product improves their specific role and business context — are twice as likely to report a high-quality purchase. For a directory whose revenue is routed inquiries, buyer confidence isn’t a soft metric. It’s the conversion event.

Five things that follow from all of this:

  1. Treat the shortlist moment as the product. Not traffic, not sessions, not time on site. Whether a buyer leaves your site with a defensible shortlist and the confidence to act on it.
  2. Answer requirement-shaped questions, not keyword-shaped ones. “Cloud MRP for a 40-person job shop with QuickBooks integration” is how buyers describe what they want. Returning a keyword list for that is a lost buyer.
  3. Cite everything. The reason buyers go to a human for validation is that generic AI output has no provenance. An answer that links back to the specific vendor profile or spec sheet it came from is doing the validation job, not adding to the pile.
  4. Instrument what you can’t answer. The questions your site fails to answer are the most valuable content-gap research you will ever get, and currently most of them are being asked and discarded in silence.
  5. Structure the corpus. Both Gartner and Forrester land on the same unglamorous conclusion: the work is making your information machine-assemblable. That’s a data project, not a marketing project.

Where the numbers stop and the argument begins

Being honest about what this research does and doesn’t establish, because the people reading it in your market will be:

What would settle it isn’t another survey. It’s a before-and-after on a real catalog: inquiries routed, percentage of sessions that end in contact with anyone, measured for 30 days before and 30 days after. That’s a small number of directories away from existing.

The short version

Your buyers want to finish the research themselves and then check their answer with someone who knows. They’re doing the first half inside tools that have never seen your catalog, and the second half is a phone call you can’t scale.

The gap between those two is where the shortlist gets decided — and right now, for most sites in this position, nothing lives there.

Sources