"90% of Startups Fail": How to Find the Original Source of a Claim

You write a sentence like "roughly 90% of startups fail within five years" because you've read it a hundred times. Then you go to cite it, and every page you open cites another page, which cites a blog post, which cites nothing. The claim is everywhere and the source is nowhere. This is one of the most common ways a draft stalls — and the fix is a specific research skill, not more Googling.

Why some claims are hard to trace

Widely repeated statistics tend to travel through a chain of secondary sources. A journalist paraphrases a report, a blog quotes the journalist, a textbook cites the blog, and the number drifts a little at each step. By the time it reaches your draft, the figure may have lost its sample, its year, its country, and sometimes its original meaning.

Three failure modes are worth recognising early:

• Orphaned statistics. The number is real but the original study is old, paywalled, or published as a government report that never got indexed well.
• Drifted numbers. The source said "about half of new businesses survive five years"; the internet rounded it into "90% fail."
• Zombie claims. No original study exists at all. The figure was an estimate someone gave in an interview and it calcified into a fact.

Knowing which one you're dealing with changes what you should write.

How to trace a claim back to its source

Start with the most credible page that repeats the claim — a review article, a government publication, a university page — and read its citation, not its text. Follow that citation to the next document and repeat. You're walking backwards through the chain until you hit something with primary data: a dataset, a survey, a cohort study.

A few things speed this up:

• Search the phrasing, not the topic. If a claim originates in a specific paper, the distinctive wording often survives in the abstract. Searching a near-quote in Google Scholar frequently lands you on the source faster than searching keywords.
• Use the number as a search term. Combining the statistic with a discipline word ("survival rate five years cohort") often surfaces the underlying study directly.
• Check who is cited repeatedly. If four secondary sources all name the same author or agency, that's your target, even if none of them link to it properly.
• Look for the dataset behind the number. Many economic and health statistics come from national statistical agencies rather than journal articles — the citable object is a report or data release.

What to do when the trail runs out

If you can't find a primary source, do not cite a source that merely repeats the claim as though it produced it. You have better options.

Cite the strongest source you can verify and adjust the sentence to match what that source actually supports. "Failure rates for new firms are high, with roughly half closing within five years (source)" is a defensible sentence. The unverifiable version is not.

If the specific number matters to your argument, replace it with a claim you can support and note the range of estimates. Reviewers respect a hedged accurate sentence far more than a confident unsourced one. And if you must reference a secondary source, use the "as cited in" convention honestly rather than implying you read the original.

Finding sources for a statement you've already written

Most of the time, the real task isn't detective work on one famous statistic — it's finding research papers supporting a statement you wrote from memory. Instead of guessing keywords, you can hand over the sentence itself and get back real, indexed literature that actually addresses it.

That's what Citely's Find Source does: paste the claim as you wrote it, and it searches academic literature for papers that support it, so you can see immediately whether the sentence stands up — or needs rewriting before it reaches your reviewer.