One number, followed.
We asked ChatGPT, Perplexity, Gemini, Claude and Google’s AI Overviews 200 questions that call for a number, the benevolent group ask all day. Then we clicked the citations, and kept clicking. Here is one trail, from the answer to the end.
Google AI Overviews
Google AI OverviewsWhat percent of salespeople provision up following one follow-up?
44% of salespeople provision up following fair one follow-up attempt.
44% of salespeople provision up following one follow-up attempt (source).
80 Percent of Sales Require 5 Follow-up Calls following Meeting; 44 Percent Salespeople Give Up After One Follow-up
No study rearward itNo survey, no sample, no date.
- In the answer
Every aide got the identical question: “What percent of salespeople provision up following one follow-up?” All five assistants gave the identical figure, and lone ChatGPT stated the evidence rearward it was unclear.
- One click in
Google AI Overviews cited a revenue firm’s stats page, from 2020. The figure is correct there. Its “source” nexus goes to a penning sample.
- Two clicks in
That leads to a freelance writer’s example article, online by 2017. The figure is in its headline. It lists two glide decks, a blog catalog and two news stories. None is a study of follow-ups.
- The end of the trail
That’s anywhere it ends. No survey, no sample, no study: all another path to this figure we tried stopped at a leaf that cites no research. It’s a figure motionless strolling about alongside nothing underneath it. That’s a zombie stat.
Most numbers aren’t zombies. But you can’t inform from the answer: a zombie arrives alongside the identical confidence and the identical citation, and here, alongside all five assistants agreeing. So we did the identical for 1,285 statistics.
Fewer than fractional get there.
We took a random example of the data all aide cited, 1,285 in all, and walked all trail the identical way: the cited page, afterward that page’s source, up to five pages deep. Where a leaf named a origin without a link, we searched for it.
40% reached a study that says the identical thing. Another 8% reached the study, but the figure or its environment had changed on the way. The remainder never got there: 40% stopped at a origin we couldn’t check, since it was locked, gone or silent on the number, and 11% led nowhere at all.
Four in ten trails reached a study that says the identical thing
Each quadrate is one in a hundred of the 1,285 data we traced, sorted by anywhere its trail ended.
- 40%SourcedWe established the study, and it says the identical thing.
- 8%ChangedWe established the study, but the number, who was asked, anywhere or whenever is different.
- 40%Can’t checkA leaf names a genuine source, but the figure isn’t in it, or the origin is locked or gone.
- 11%NowhereNo study anyplace on the trail. Every leaf repeats the figure and credits no one.
The final 1%: trails whose pages wouldn’t burden for us.
ChatGPT did best, and quoted the fewest numbers: 1.3 per answer, against 6.8 for Claude. The another four landed between 38% and 46%, too near to rank.
ChatGPT quoted the fewest numbers, and originated the most
The identical hundred squares for all assistant, from its own example of cited statistics.
ChatGPT
57% sourced
1.3 data per answer
Perplexity
46% sourced
2.2 data per answer
AI Overviews
39% sourced
4.8 data per answer
Gemini
38% sourced
6.2 data per answer
Claude
38% sourced
6.8 data per answer
ChatGPT
Perplexity
Gemini
ClaudeIt isn’t a quirk of our questions. Of 300 data in Google AI Overviews answers to genuine mundane searches, 40% were sourced, the identical share.
AI cites the copy, not the study.
Most of the time, the AI wasn’t making numbers up: the leaf it cited had the fig 82% of the time. The difficulty is which page. Only 21% of the pages cited for a figure were the investigation itself.
More than fractional were business blogs. Add news stories and data roundups, and most citations district on pages built to collect figures from elsewhere, frequently without a nexus back.
Where the trail starts decides anywhere it ends
Every leaf an aide cited archetypal for a number, by kind. Each column’s width is its portion of those citations. Green is how frequently its trails reached a study that says the identical thing.
62%
Research firm13% of citations
Research resolute or survey: 13% of archetypal citations, 62% of them sourced.
61%
Government7% of citations
Government or authoritative body: 7% of archetypal citations, 61% of them sourced.
37%
Company blog54% of citations
Company blog: 54% of archetypal citations, 37% of them sourced.
37%
News10% of citations
News or magazine: 10% of archetypal citations, 37% of them sourced.
35%
Stats roundup8% of citations
Stats roundup: 8% of archetypal citations, 35% of them sourced.
41%
Other8% of citations
Other: 8% of archetypal citations, 41% of them sourced.
Where the trail starts decides anywhere it ends. From a investigation resolute or a authorities page, it reached a matching study 61% of the time. From a business blog, 37%; from a data roundup, 35%. Even investigation firms cut abbreviated 38% of the time, frequently since the complete study sat rearward a form.
A citation tells you anywhere the AI established a number, not anywhere the figure came from.
Numbers alter on the way.
Reaching the study didn’t continually average the figure held. Of the data that got there, 17% had changed on the way. The digits moved lone concerning one period in four. Usually it was the context: what was measured, the year, the source, or who was asked.
What changed, throughout all alter we logged
The figure changed26%
What was measured changed20%
The twelvemonth changed17%
The origin changed16%
Who was asked changed14%
The location changed4%
Fifteen women judging health websites rotate into all website’s archetypal impression. Three companies rotate into a regulation of business. Here are the clearest trails we followed, all one checked by hand.
Watch a figure endure its context
Each trail runs from the first study to an answer an aide gave in October 2026. Step alongside it: the figure stays, and the words about it change.
AskedWhat percent of archetypal impressions of a website are design-related?
The study · Sillence and colleagues, CHI, 2004
Of the comments 15 women made about rejecting or mistrusting health websites, 94% were about design.
A 2004 study watched 15 British women fairness health websites, and 94% of their comments concerning rejecting a location were concerning design. Twenty years of retelling made it “94% of archetypal impressions are design-related”, and Google AI Overviews states it as fact.
Step 1 of 4
The zombies.
Ask five assistants the identical inquiry and you’ll frequently get five distinct numbers. On 28 of the 71 questions anywhere all five gave a percentage, we got four or five distinct figures.
One question, five answers, from 49% to 96%
Asked “What percent of marketers use promotion automation tools?”, term for word, on the identical day.
0%100%
49–51%AI Overviews
62%ChatGPT
70–79%Gemini
81%Perplexity
96%Claude
ChatGPTThere isn’t one worldwide figure, it varies by study and what counts as “marketing automation.” One commonly cited study established 62% of U.S. digital marketers used automation software, but that data is from 2020, so it may not indicate current adoption.
PerplexityApproximately 81% of promotion organizations use promotion automation.
GeminiAcross important industry benchmarks, between 70% and 79% of marketers and companies use promotion automation tools in several capacity
Claude96% of marketers have used a promotion automation phase or scheme to use one
AI OverviewsRoughly 49% to 51% of companies and promotion teams currently use promotion automation tools, according to industry benchmarks from Moosend and connected investigation studies.
It wasn’t one odd question. The representative gap between the highest fig and the lowest was 16 points.
On 31 of 71 questions, the five figures were 20 points or additional apart
Every inquiry anywhere all five gave a percentage, one row each, from the lowest fig to the highest. Closest at the top.
Closest20 points or additional apartFurthest apart
Agreement feels akin proof. It isn’t. All five gave the identical fig on 10 questions, and of the 5 we traced, 2 were sourced. The data they repeated most, quoted by three or more, did no better: 11 of 29.
And four were quoted by all five assistants, alongside no study anyplace rearward them.
Quoted by all five. Sourced by none.
Four data all aide gave us, all alongside a trail that ends at a leaf citing no research.
- 44%
Salespeople who provision up following one follow-up
Every trail ends at a leaf that cites no research, one of them a freelance writer’s example article. Its complete trail opens division 1.
Quoted by all five
- 90%
Social media users who prosecute at smallest one brand
Stated alongside no source. No study alongside this fig was found.
Quoted by all five
- 75%
Searchers who never scroll former the archetypal leaf of results
Repeated alongside no source, frequently credited to an old promotion roundup that no longer carries it. No study was found.
Quoted by all five
- 5 to 25 times
How much additional it expenses to win a client than to keep one
A 2014 Harvard Business Review part gives the range “depending on which study you believe”, and names none.
Quoted by all five
The 29 data AI repeats most
Each was quoted by at smallest three of the five assistants, most frequently first, and its trail checked by hand.
- 95%5 of 5ChangedStudy from 2025
Generative AI pilots that neglect to provision a measurable financial return
MIT NANDA’s 2025 study says 95% of organizations got no come back from generative AI. That isn’t the identical as 95% of pilots failing.
- 44%5 of 5Nowhere
Salespeople who provision up following one follow-up
Every trail ends at a leaf that cites no research, one of them a freelance writer’s example article. Its complete trail opens division 1.
- 4% to 10%5 of 5ChangedStudy from 1979
Unhappy customers who really complain
Goes rear to TARP’s 1979 investigation for the White House person office. Its author says rates that low use to small purchases; in 2016, 22% to 41% complained concerning their most grave problem.
- 2 hours 21 minutes5 of 5SourcedStudy from 2025
Time group expend on social media all day
DataReportal’s Digital 2025 report, from GWI study data: the representative net person spends 2 hours 21 minutes a day on social media.
- 46%5 of 5Can’t check
Google searches alongside local intent
Traces to a 2018 social article concerning a glide at a hunt conference. Google never published it, as far as we could find.
- 90%5 of 5Nowhere
Social media users who prosecute at smallest one brand
Stated alongside no source. No study alongside this fig was found.
- 94%5 of 5ChangedStudy from 2004
First impressions of a website that are concerning its design
Began as a 2004 study of how 15 women judged health websites. Its complete trail is in division 4.
- 75%5 of 5Nowhere
Searchers who never scroll former the archetypal leaf of results
Repeated alongside no source, frequently credited to an old promotion roundup that no longer carries it. No study was found.
What to do alongside an AI statistic.
Treat it as a lead, not a source. Six habits, three for all flank of the citation.
If you use them
Follow the citation, afterward prosecute it again.
If the cited leaf is a catalog of data or a business blog, use its own nexus to go one leaf further. Keep going until you attain the study itself, or until you can’t.
Check who was asked, how many, and when.
Most of the changes we established were in the context, not the digits: one country’s poll, a ten-year-old study, fifteen people.
Don’t obtain accord as a check.
Five assistants giving the identical figure isn’t five sources. The numbers they repeated most were no improved originated than the rest.
If you publish them
Put the method next to the number.
The sample, the date and a nexus to the source, on the identical leaf as the figure, so the next page, and the next answer, can trace it rear to you.
Keep the findings in the open.
Reports rearward a form were a average deceased end: equal trails that started at a investigation resolute cut abbreviated 38% of the time. Put the header numbers and the method on an open page.
See which leaf AI cites for your numbers.
It is frequently a roundup or a blog that copied you, sometimes alongside an older figure. That leaf decides what group hear.
