If I flip a coin it is a 50:50 chance of Heads.
If I flip 10 coins, the chances all will be Heads is tiny.
But EACH coin still has a 50:50.
People struggle with one particular part.
If you flip one coin 10 times, and the first 9 flips all come up heads, then they think that 10th flip has a much higher chance of being tails, to balance the odds, right?
But the coin doesn’t know, and so it still has a 50:50.
The change in odds is in the conglomeration, the collective of all the flips, not in ANY individual flip.
In SEO, this means Google may look at a single signal that has an 80% chance to be spam, and decide it is too low threshold to penalize.
But 10 different signals…
Well, that’s when you get to odds with a probability around the 99.something percentage, and can absolutely decide to downrank it below stuff that doesn’t have that high probability of being spam.
Where this failure to understand shows up the most often in SEO is when people say “Google can’t KNOW they are paid links” or “Google can’t be SURE it is AI written” etc.
They don’t have to know.
They just need a bunch of various probability numbers to do math with.
Google don’t have to KNOW a thing is spam, or AI.
They can easily just include stuff they feel very sure ISN’T.
Or better, just downvalue anything iffy, so even if wrong, everything like it gets downgraded equally and no harm done.
Downvalue only shows when others weren’t downed.
What can really nudge your noodle is when a factor might be based on SERP interaction data (e.g. Bounce Back To SERP), and a site that is just as bad, but never ranked to clock up the negative signals before, will TEMPORARILY take its place.
Until the clicks start.
Google might put a positive spin on sites that didn’t rank before suddenly getting a trial in the SERP.
They might call it ‘surfacing hidden gems’ or something.
What matters isn’t what takes the place of spam immediately – what matters is what sticks there.
But hey, you can churn and burn sites that only last a few weeks, some will say.
Sure.
But the sites that appear briefly, then slide away to the same penalty, are training the ML as they go.
Same stuff won’t work long.