The statistic in your deck cites a domain that no longer resolves

The statistic in your deck cites a domain that no longer resolves

Run this:

dig marketingdonut.co.uk

On 14 August 2026 it comes back status: NXDOMAIN, ANSWER: 0. Nothing there. To prove the resolver is working, dig bbc.co.uk on the same machine returns four A records.

That domain is the cited origin for two of the most repeated numbers in B2B sales: that eighty percent of sales require five follow-ups, and that forty-four percent of salespeople give up after one. Those numbers are in pitch decks, in enablement training, on the walls of sales floors, and in the opening slide of a great many vendor webinars. The source they point at stopped existing sometime in the last five months.

That is the version of this story that gets told. The version that came out of actually checking is worse.

The citation evaporated. The claim did not.

An internal audit named three pages that carry the claim. All three were fetched on 14 August 2026 and all three returned HTTP 200. All three still assert it: the string "80%" appears 2, 8 and 30 times across them. And all three contain zero occurrences of the word "marketingdonut".

The claim has outlived its own footnote. It now circulates with no attribution at all, phrased as established fact. One of the pages says "80% of sales require at least 5 follow-up contacts, yet 44% of reps quit after just one attempt" and immediately adds that these "are the most important numbers in this article". No source is offered because there is no longer a source to offer, and nothing about the sentence tells a reader that.

This is the part worth internalising. A dead link is a visible failure. A number that has shed its link is invisible, and it reads as more confident than the version that had a citation, because uncited assertions look like common knowledge.

For completeness, the origin was not much better when it existed. According to the audit trail, the page was a bylined opinion column whose entire methodology statement was that the figures came from "different studies carried out at different times, in different places, by different market research companies". No study named, no sample, no date. That description is relayed rather than read here, for the obvious reason.

The companion figure, the cascade of how many reps quit after one, two or three attempts, is attributed to a "National Sales Executive Association", an organisation that nobody has been able to show ever existed. A trail leads to a 1957 trade volume, and it has not been reached at source, so it stays at medium confidence rather than being written up as an origin.

Numbers move, and the people who publish them move them quietly

The other failure mode is not a dead source. It is a live source that has changed its answer.

Salesforce's State of Sales report is the usual citation for how little of a rep's week is spent selling. The figure in wide circulation is 28 to 30 percent. The seventh edition of the report, downloaded on 14 August 2026, contains a chart that reads "60% Not selling, 40% Selling".

Ten points, from the same publisher, in the same report series. Anyone whose slide says thirty percent and credits Salesforce is now being contradicted by Salesforce.

Both editions are worth reading with their methodology attached, which is stated plainly in the report and almost never quoted alongside the number. The seventh edition surveyed 4,050 sales professionals across 22 countries, in an anonymous survey conducted August through September 2025, and states: "All respondents are third-party panelists." That is a self-reported estimate of how people spend their week, from a paid panel. It is a legitimate thing to publish and a fragile thing to build an argument on, and a ten-point move between editions is exactly what fragility looks like from the outside.

Correlation with a causal verb on the front

The most current example is happening right now, in the numbers about AI Overviews and clicks.

The figure everyone quotes comes from Ahrefs, published 4 February 2026, drawn from 300,000 keywords comparing December 2023 with December 2025. The finding, in the study's own words: "the presence of an AI Overview now correlates with a 58% lower average clickthrough rate for the top-ranking page."

Fetch the page and search it. It contains two occurrences of "correlates" and zero occurrences of any word beginning "causa". So the sentence people quote is carefully hedged. The title of the same page is "Update: AI Overviews Reduce Clicks by 58%", and the closing line says "AI Overviews reduce the organic click-through rate for position one content by 58%". The method is a before-and-after comparison against a forecast counterfactual, over two years in which a great many things about search changed. Ahrefs sells SEO software.

There is now a causal number, and it is smaller. A randomized field experiment assigned 1,065 US desktop Chrome users to see or not see AI Overviews and observed 68,089 searches. Where an overview appeared, removing it raised outbound organic clicks and cut zero-click searches: a 39.8 percent reduction in outbound organic clicks attributable to the overview, a 34.5 percent rise in zero-click searches, and no measurable movement in sponsored clicks.

Two honest notes about that paragraph. First, the two figures are not measuring the same thing, so 39.8 does not simply replace 58; one is a two-year observational drift, the other is a within-study contrast. Second, the working paper sits on SSRN and SSRN returns HTTP 403 to this machine, verified twice. The numbers above are relayed from press coverage of the paper rather than read in it. That is a weaker link than the rest of this article and it is marked rather than smoothed over.

The caveat the author wrote and nobody carries

Sometimes the source is fine and the problem is that everyone stopped reading two sentences early.

The 95:5 rule, that only five percent of business buyers are in market at any time, comes from John Dawes at the Ehrenberg-Bass Institute. His own page shows the derivation: corporations change service providers "around once every five years on average", therefore twenty percent are in market across a year, therefore roughly five percent in a quarter. It is arithmetic from an assumed replacement cycle, not a measurement of buyer intent, and the page says so in the next paragraph:

The 95% figure is not meant to be a precise rule. We're using it as a heuristic to get the idea across.

The underlying idea is almost certainly right. Most of a market is not buying today. The failure is in the repetition: a heuristic gets quoted as a measurement, then applied to categories whose replacement cycles look nothing like five years, then used to justify a budget.

What to do with a number before you use it

Three questions, in order, and a number that fails any of them does not go in the deck.

Can you reach the source? Not the roundup that cited it. The document. If the trail terminates in another roundup, you do not have a source, you have a chain.

Does the source contain the number? This is where more of them die than anyone expects. A figure attributed to a named report is frequently absent from that report.

Does the source say what the number means? Sample, date, method, and the qualifier the author attached. "Correlates with" is not "causes". A panel's self-estimate is not a measurement. An author's own heuristic label is part of the finding.

There is some evidence this is becoming normal rather than eccentric. Statistic roundups, which were the transmission vector for every myth above, have started publishing what they refused to carry. One of them states: "We chased each figure to its origin rather than quoting another roundup. Where a widely-shared number couldn't be traced to a real study, we dropped it."

And the same rule applied here

The internal document that supplied most of the trails above corrected itself in its own text. Its first draft stated that the EU AI Act's high-risk obligations took effect on 2 August 2026. They did not; they were deferred by a regulation that came into force the previous week, and what applied on that date was the transparency article. The author wrote the correction into the document rather than quietly fixing the sentence, which is the only reason it is possible to say so here.

This company has no statistics of its own to offer. There are no customers to count and no results to report, which is a constraint rather than a virtue. It does mean there is nothing to be gained here by borrowing somebody else's number, and one obvious thing to lose. A page whose whole argument is that claims should be checkable cannot afford to be carrying a figure that dies on a DNS lookup.