Sixteen B2B buying statistics, traced back to the documents they came from.

The wording is real. The deck says “the average B2B buyer is 57% through the purchase decision before engaging a supplier sales rep.” The trouble is underneath it.

The sample description conflicts with itself. The deck states n=1,399. LinkedIn’s tracing of the same study describes 1,500 decision-makers across 22 major B2B companies. If the 22-company figure is right, the effective sample is 22 organisations — a materially weaker basis than the headline n implies. The conflict is unresolved.

And the 57% has no published derivation. The deck displays the number beside a pie chart of information sources, but never defines how “progress in purchase decision-making” was measured, or on what scale. It is a self-reported perception with no stated instrument.

The critique is not fringe. LinkedIn’s own B2B Sales blog ran it on 5 September 2016 under the title “This Popular Stat is Wasting Your Time: The 57% Engagement Myth.” CEB was acquired by Gartner in 2017 and the deck now survives only on third-party mirrors.

Heuer’s actual wording: “here’s the SiriusDecisions-approved version: 67 percent of the buyer’s journey is now done digitally.” Done digitally is a statement about channel, not about sequence. It says where buyers were, not how far through they were.

In the same post, under the section heading “Myth: Sales doesn’t get involved until more than halfway through the buying cycle”, she wrote: “The 67 percent statistic in no way says that no one talks to a salesperson before getting halfway through the buying cycle, but this is how some have interpreted it.”

So the firm that published the number rejected the pre-contact reading in the same year it published it, and reinforced that at its 2015 Summit by replacing the percentage with an interaction-count framework — 17 total interactions in a committee purchase, 9 of them human.

No sample size, fielding dates or methodology for the 67% has ever been published, by SiriusDecisions or by Forrester, which acquired it in 2019.

Wizdo’s sentence was hedged twice over. Buyers “might be anywhere from two-thirds to 90% of the way through their journey before they reach out for a sales person.” That is a 23-point range, stated as a possibility.

She was not quoting a study. LinkedIn’s tracing is blunt about it: Wizdo “wasn’t quoting from a specific piece of research — and she wasn’t claiming that a high proportion of buyers conduct virtually the entire purchase journey by themselves — just that some do.”

Forrester’s own 2021 work cuts against the popular reading: 27 buying interactions across a journey, with human contact distributed throughout rather than bunched at the end.

We could not retrieve the 2013 post. Forrester’s blog URLs from that era redirect, and the archive was rate-limiting. The wording above is from secondary tracing, not the source document, and we use it only as grounds for not citing the figure — never as evidence for anything.

The figure is real, and the gloss usually attached to it is right too: the 17% is divided across every supplier considered, leaving any one supplier roughly 5–6%.

But the primary is Win More B2B Sales Deals (© 2018 Gartner, Inc., edited by Adamson), footnoted to the Gartner 2017 Digital B2B Buyer Survey, n=750. The 2022 HBR article is a vehicle for a 2017 measurement. Cited as “Gartner (2022)” it reads as four years old. It is nine.

Gartner no longer publishes it. The 17% was on Gartner’s B2B buying journey page in April 2020 and in March 2022, and is absent from every capture from January 2024 onward, replaced by rep-free preference figures. There was no retraction and no explanation. That does not make a 2017 measurement false, but it does mean the figure’s own publisher has retired it.

And Gartner rejects the framing the figure is used to support: “B2B buying doesn’t play out in any kind of predictable, linear order. Instead, customers engage in what one might call ‘looping’ across a typical B2B purchase.” A journey with no linear progression cannot be 57% complete, or 67%, or 70%. Gartner files the 17% under seller access and share of influence, not stage.

The more useful number is on the same chart and almost nobody quotes it. Researching independently online 27%, meeting with the buying group 22%, researching independently offline 18%, meeting with potential suppliers 17%, other 16% — so roughly 45% of buying time is independent research.

Forrester’s own blog reports 27 buying interactions per individual buyer journey. Not per vendor.

The difference is not pedantic. Multiplying 27 by the number of vendors a buying group considers inflates the figure several-fold, and the inflated version is the one in circulation.

Our own source library carried the “per vendor” wording, and omitted the sample size.

This one is ours. It was published on our homepage, carried in our source library, and asserted a third time in our citation register as “direct LLM-usage evidence.”

What Forrester actually published: “Almost 95% of buyers anticipate using genAI to support their decision and purchase process in the next 12 months.” Forrester’s own blog describes the study as “more than 16,000 global business buyers.”

Four errors in one sentence. The number: ninety versus almost 95. The tense: we wrote “now rely on”, a measured present-tense behaviour, where Forrester measured stated intent for the coming year. The task: we wrote “to research and shortlist vendors”, a specific job, where Forrester asked about supporting the decision and purchase process generally. The sample: 11,352 does not match the publisher’s own description of its study, and we could not trace it to any Forrester document.

Where the number probably came from makes it worse rather than better. Forrester’s Buyers’ Journey Survey — a different instrument, 17,500–18,000 buyers annually — reports 89% using AI in their buying process in the 2024 wave and 94% in the 2025 wave. Those are real measurements. But they are a different survey with a different sample, and they measure AI use in buying broadly, including organisation-provided internal tools, which 61% use. It is not a shortlisting figure.

So the sentence was a conflation rather than an invention: a roughly correct number welded to the wrong study, the wrong sample size, and a far narrower task than was ever measured. That is harder to spot than a fabrication, and easier to defend badly.

This is the single most quoted sentence supporting AI shortlisting, and it comes from a firm whose research is normally excellent — which is exactly what makes it travel.

There is no B2B survey behind it. No sample, no fieldwork dates, no geography. The only survey Bain cites in the piece is its Generative AI US Consumer Survey of September 2025, n=1,500 — a consumer study — from which the B2B statement is extrapolated.

It reads like Bain data. It is Bain opinion. Those are different things, and the sentence does not say which it is.

The column this is read off is a share-of-influence allocation across ten sources, and it sums to about 100%.

17.1% is generative AI’s share of influence within that allocation. It is not the proportion of buyers doing anything. The two are not convertible.

The Edelman–LinkedIn B2B Thought Leadership report contains no AI statistics at all.

The figure belongs to Responsive, an RFP-software vendor, n=350+, method undisclosed. It has been attached to a source it never appeared in — and to one that a great many people, ourselves included, already cite for other things.

The Microsoft Canada report that popularised the figure sourced it to an outfit called Statistic Brain. The BBC’s More or Less went looking for the underlying study and could not trace it.

We include it because it is the clearest case in the set and because we retired it in our own library before this audit, with the reasoning attached. That appendix is the model for everything else on this page.

The 67% is sound as a single data point: Gartner press release of 9 March 2026, 646 B2B buyers, fieldwork August–September 2025. Cite it that way and it holds.

The “up from” does not hold. Gartner’s rep-free figures run 33% (September 2020), 75% (on Gartner’s own page through 2024 and into mid-2025), 61% (June 2025, n=632, fielded August–September 2024), then 67%. A number that goes 33 → 75 → 61 → 67 is not a trend line. The question and the instrument are moving underneath it.

Gartner’s own data also complicates the preference. In the 61% release, buyers completed on average 2.3 activities with supplier reps against 1.8 through digital self-service. They say rep-free, and then involve reps in the majority of activities.

We drafted the “up from 61%” version ourselves while preparing this audit, and removed it before publication. The pull towards a tidy direction of travel is strong, and it is exactly what produces an over-claim.

G2’s March 2026 report put AI chatbots first, at 54%. G2’s June 2026 report puts review sites (38%) ahead of AI chatbots (37%).

Same publisher, same construct, four months apart, opposite conclusion. The question design changed. Building a trend across the two waves would manufacture a finding rather than report one.

G2’s page does not disclose fieldwork dates or geography for either wave.

The two ends of that line come from different publishers, different populations and different instruments. Splicing them produces a trend that neither measured.

TrustRadius’s own series settles it. Its January 2025 edition reported 7% using LLMs in the buying process. Its January 2026 edition reports 63%. A 7-to-63 jump in twelve months from one publisher on one instrument is question wording, not behaviour.

If a single publisher’s wave-on-wave numbers can move that far, a cross-publisher trend line is worthless. An earlier draft of our own audit argued for this doubling. We retracted it internally before it reached a page.

Three publishers have published a 94% in this territory, for three different things. Forrester: buyers using AI in their buying process. 6sense: buyers using LLMs. TrustRadius: the share of AI users who fact-check what it tells them.

That last one is not an adoption figure at all — it is a scepticism figure, and it points the opposite way.

One number, three constructs, already being conflated in secondary coverage. If anyone writes 94%, the construct has to be named in the same sentence.

Both are forecasts. Neither is a measurement of anything that has happened.

They are worth listing together because they are routinely quoted against each other by people arguing opposite cases, and the same firm published both. A forecast quoted as a finding is a category error regardless of which side it flatters.

This one is ours as well. Our source library’s entry for the 2022 HBR article carried this sentence alongside the Gartner 17% figure.

It has the shape of the SiriusDecisions 67% — a statement about digital channel share, restated as a statement about sequence — wearing a Gartner citation. We could not locate it in the HBR article.

The 17% is the figure that article carries. It is the one that entry should rely on.