Task
I was checking a statistic in an AI-generated article.
The claim was: "According to a 2022 industry report, 68% of small businesses now use at least one AI tool."
It looked specific. It looked sourced. It looked verifiable. So I searched for it.
I found five pages that repeated the claim. Five different websites. Five different articles. All of them said the same thing. All of them cited the same report.
That's when I got suspicious.
If five pages all say the same thing, they might be independent confirmations. Or they might be repeating each other. I wanted to know which.
Here's what I found.
Tool and Date
Tool used to generate the draft: A general-purpose AI assistant (client's choice, paid tier)
Date of draft generation: Not provided. Article was recent.
Date of my audit: September 16, 2026
Tool I used to trace the source: Web search, Wayback Machine, direct site visits
Original Input
I don't have the original AI prompt. The client shared the article draft, not the prompt.
The claim in the article was:
"According to a 2022 industry report, 68% of small businesses now use at least one AI tool."
The article didn't include a link. It didn't name the report. It just said "a 2022 industry report."
Claim Under Review
The claim is: "68% of small businesses use at least one AI tool, per a 2022 industry report."
I wanted to know three things.
Which report did this come from?
Does the report actually exist?
Does the report actually say 68%?
Evidence
I searched for the exact phrase and the exact statistic. Here's what I found.
Page 1: A marketing blog. The page said: "According to a 2022 industry report, 68% of small businesses now use at least one AI tool." No link. No report name.
Page 2: A software review site. The page said: "Studies show that 68% of small businesses are now using AI." No link. No report name.
Page 3: A business news article. The page said: "A 2022 industry report found that 68% of small businesses now use AI tools." No link. No report name.
Page 4: A consultant's blog. The page said: "Recent research indicates that 68% of small businesses use at least one AI tool." No link. No report name.
Page 5: A vendor's landing page. The page said: "68% of small businesses are already using AI." No link. No report name.
Five pages. Same number. Same year. No source.
I checked the publication dates. Page 1 was published in May 2023. Page 2 in August 2023. Page 3 in November 2023. Page 4 in February 2024. Page 5 in June 2024.
Page 1 came first. The other four pages were published after. I checked the wording. Pages 2, 3, 4, and 5 all used slightly different wording, but the statistic was identical.
I checked Page 1 for a source. It didn't have one. I checked the author. The author was a freelancer. I checked the author's other work. No other references to the statistic.
I searched for the actual report. I searched for "2022 industry report 68% small business AI." I searched for "68% small business AI adoption 2022." I searched for "small business AI adoption report 2022." I checked Google Scholar, the U.S. Chamber of Commerce, the Small Business Administration, McKinsey, Deloitte, Gartner, Forrester.
I found no report that said 68%.
I found several reports that said something different. One said 37%. One said 43%. One said 51%. None said 68%.
I checked the Wayback Machine. I looked for the original Page 1 to see if it had ever cited a source. It hadn't.
Conclusion: The 68% statistic appears to have originated on a marketing blog in May 2023 with no source. Four other pages repeated it without verification. The statistic does not appear in any primary source I could find. It is unsupported.
Finding
Confirmed error — unsupported statistic.
The 68% figure appears to be a number that was invented or mistakenly transcribed by a marketing blog, then repeated by four other sites. It is not from a real 2022 industry report. It is not supported by any primary source I could find.
The AI that wrote the article didn't invent the number. It repeated a number that had already been repeated. The failure is upstream.
Risk
If the article had been published with this statistic, it would have continued the chain. Another reader might have cited the article. Another AI might have picked up the number. The fake statistic would keep spreading.
This is how misinformation compounds. Not through a single lie. Through repetition. Five pages repeating the same unsourced number look like five confirmations. But they're actually one unsourced claim, copied four times.
The risk isn't just one wrong number. It's the amplification of a wrong number.
Lesson
Here's what I learned from tracing this claim.
Repeated claims are not confirmed claims. If five pages say the same thing, that doesn't mean five sources agree. It might mean one source said it, and four other sources copied it.
Look for the earliest version. Find the oldest page that repeats the claim. Check whether it cites a source. If it doesn't, the claim may have originated there.
Look for the link. If a page cites a report but doesn't link to it, that's a red flag. A real source can be named and linked. An invented source can't.
Search for the actual report. If a claim references a specific report, search for that report directly. If you can't find it, the claim may be unsupported.
Check the number against other sources. If three credible reports say 37%, 43%, and 51%, and one blog says 68%, the blog is probably wrong.
Use the Wayback Machine. If a page has been updated, the Wayback Machine shows the original. Sometimes the source was added later. Sometimes it was removed.
Trace the claim, not the AI. The AI didn't invent the number. It repeated a number that was already circulating. The AI is downstream from the original failure. Trace the chain to its source.
What I'm Asking the Community
How do you distinguish a real primary source from a page that merely repeats an AI-generated claim?
Here's what I do now.
Find the oldest page that repeats the claim.
Check whether it cites a source.
If it does, follow the source.
If it doesn't, mark the claim as unsupported.
Search for the actual report by name.
Compare the number against credible alternatives.
This took me about 45 minutes for one claim. That's too slow for every claim. Is there a faster way?
I'm also curious about tools. Is there a way to find the earliest version of a claim automatically? Is there a way to check whether a statistic appears in a primary source? Is there a tool that traces repeated claims across the web?
And one more question: what should I do with the finding? Should I tell the client? Should I post the case here? Should I contact the original blog that started the claim? I'm not sure how far to take the correction.
If you have a faster method or a better framework, please share it. I'd rather build a habit now than after another unsupported statistic goes out the door.
Comments
No comments yet — be the first to share a thought.