Myths and unverified statistics

There's a particular kind of statistic that runs this industry. Precise, memorable, attributed to nobody, and repeated until everybody treats it as common knowledge.

Try tracing one. Pick any confident percentage about click-through rates, zero-click searches or AI referral behavior and follow the citations backward.

The usual path: a blog post citing an agency roundup citing a webinar slide citing an analysis nobody published. Often it ends in a circle.

The four questions

Who measured it? A platform publishing about its own system is strong. An independent study with a published method is good. A vendor measuring the thing their product improves is weak.

On what? A sample from one industry doesn't generalize to yours. "We analyzed a million keywords" tells you nothing until you know which million.

When? These numbers decay fast. A figure measured before a major interface change describes a world that doesn't exist anymore.

What were they selling? Not disqualifying. It just tells you which direction the error runs.

Why this keeps happening

There's advice going around that every section of a page should contain a data point, because figures get quoted. It isn't unreasonable advice. It also creates direct pressure to put a statistic in whether or not you have one.

That pressure is how made-up numbers get into honest work. Nobody sits down and decides to invent a statistic. Somebody needs a figure for a section, half-remembers one, and types it.

We fixed that structurally rather than by willpower. A central ledger holds every figure with its source, publisher and the date we looked. A number in the text that isn't in the ledger fails the build. Every section is welcome to carry a figure. An unsourced one can't ship. A page with no sourced figure carries none, and that's fine.

Numbers we won't repeat

Precise zero-click percentages. Quoted everywhere, method opaque, and definitionally unstable — what counts as a "search" and a "click" varies between every source.

"AI traffic converts N times better." Asserted constantly. Measuring it would need attribution that our own measurement section explains isn't available.

Click-through rate by position curves. They exist, they vary enormously by query type and interface, and they get quoted as though they're universal.

That Glassdoor sentiment is becoming an E-E-A-T factor. Turns up in practitioner material with nothing behind it. We grade it speculative and label it a guess.

Any specific formatting rule for getting cited. Word counts, sentence lengths, optimal list sizes. No platform publishes a formatting spec. Where we give formatting advice, it's because the rule follows from how extraction works — not because somebody spotted a correlation.

What the ledger does hold

Core Web Vitals thresholds. Sitemap and robots limits. The robots cache duration. The effective date of the FTC's reviews rule. The number of crawlers OpenAI documents.

Each one records the figure, the source URL, the publisher, and the date we retrieved it.

That's a short list for a site this size. The shortness is the point.