SEO Talks

How do you tell a real ranking signal from a correlation everyone repeats?

You tell them apart by asking for a plausible causal mechanism and real evidence, not by counting how many people repeat the claim. A real ranking signal has a reason Google would actually use it, a mechanism that makes sense for ranking quality results, and ideally Google’s own confirmation or controlled evidence that isolates the effect. A repeated correlation is just an observed association in the data, a pattern someone noticed between a feature and rankings, with no demonstrated cause behind it. The discriminator is causal-mechanism-and-evidence versus correlation-with-a-confound, and the test is “what is the mechanism and is it confirmed,” not “does everyone say it.”

The classic giveaway of a correlation masquerading as a signal is the confound. Studies find that higher-ranking pages tend to have shorter URLs, or more words, or some other measurable trait, and the trait gets repeated as a ranking factor. But the trait usually travels with something else that is the real cause. Shorter URLs sit on better-structured, more authoritative sites; longer content correlates with thoroughness that satisfies intent. The measured feature rides along with the actual driver, so optimizing the feature itself does nothing. An association without an isolated cause is a confound, not a signal.

This is why widely-repeated correlations should not be treated as signals just because they are everywhere. Repetition is evidence of popularity, not of causation, and SEO advice propagates claims faster than it tests them. A claim can be stated confidently by many sources and still be a confounded correlation that nobody actually verified. The honest move is to demand the mechanism and the evidence before you act on a supposed factor, because acting on a confound spends effort on a lever that is not connected to anything. Consensus is not confirmation.

To apply the test, take any claimed signal and ask two questions. Is there a plausible mechanism Google would actually use to reward this? And is there confirmation or controlled evidence, or only an observed association that could be explained by a confound? Treat the claims that pass as real signals worth acting on, and the ones that fail as correlations to ignore no matter how often you hear them. Test for mechanism and evidence, not consensus.

Related posts:

  1. Why do shorter URLs tend to rank better, correlation or cause? It is correlation, not cause. Shorter URLs tend to accompany higher-ranking pages because they travel with cleaner site structure, better click-through, and easier sharing and linking, and those are the......
  2. How do you tell a seasonal dip from a real ranking problem? To tell a seasonal dip from a real ranking problem, compare year over year rather than just month over month, and check whether your positions held while traffic fell or......
  3. Is the “EEAT score” a real ranking factor or a framework? There is no "EEAT score," so the answer is framework, not factor. Google's ranking system does not hold a number called EEAT, does not compute one, and does not feed......
  4. Should anchor text repeat the exact keyword, or does that trigger a penalty signal? Repeating an exact keyword in internal anchor text is far lower risk than the backlink version of the same worry, and most of the penalty fear is imported from the......
  5. Discovered vs Crawled “not indexed”, which is the worse signal and why? "Discovered, currently not indexed" is usually the worse of the two signals, because it means Google has not even judged the page worth crawling yet. The two labels sit at......
  6. What does Google actually read as a “quality” signal it can measure? Google cannot measure "quality" directly, so it approximates it through measurable proxies. There is no quality meter that reads a page and returns a verdict; instead the system infers quality......
  7. What’s the smallest change that actually adds an experience signal? The smallest change that actually adds an experience signal is one concrete piece of genuine first-hand evidence: a specific result you got, a tested example, an original photo of the......
  8. What does a spike in “crawled not indexed” pages actually signal? A spike in "crawled, currently not indexed" pages usually signals a quality or duplication problem at scale, not a technical glitch. The status means Google crawled those pages and then......
  9. Is a sudden drop more likely an update or a technical break, how to tell? Tell them apart with one test: check the drop's timing, its scope, and your own change log together, because the two causes leave opposite fingerprints on all three. A technical......
  10. How do you tell if crawl budget is your problem or just an excuse? Read your crawl stats and index coverage data, and if your important pages are being crawled regularly and are indexed, then crawl budget is not your problem, full stop. A......
  11. How do you tell which update hit you when several rolled out? Tell which update hit you by matching the exact dates of your traffic movement to documented update timelines, then checking which content pattern was affected. Date alignment narrows the field:......
  12. How do you tell transactional from informational intent on the same keyword? Search the keyword yourself and read the result mix Google returns, because the SERP is where Google has already published its verdict on intent. The same words can carry either......
  13. How do you tell if Google is crawling a page but choosing not to index? Check Search Console: run the page through URL Inspection and read the index coverage report, because the exact status tells you whether Google crawled the page or never got that......
  14. How do you tell if a domain is under a trust deficit vs a content problem? Check the scope of the problem: a trust deficit shows up as broad, site-wide suppression, while a content problem is page or topic-specific. That single test, site-wide versus isolated, separates......
  15. How do you tell if the SERP’s intent shifted, not your page? Compare the current SERP's content types to what they used to be and to what your page is, and the comparison tells you which problem you have. The discriminator is......