Spoke
User engagement
These are the most over-interpreted numbers in search marketing. Worth watching, and they don't mean what your dashboard implies they mean.
That engagement metrics feed ranking has never been documented and has been disputed by Google representatives for over a decade. The diagnostic uses here are our judgment.
Graded speculative, because the central claim people make about engagement has never been established, and the metrics themselves are weaker than they look.
Does time on page or bounce rate affect rankings?
Nothing documents engagement metrics as ranking inputs, and Google representatives have disputed the claim repeatedly for over a decade. Treat any confident assertion otherwise as unsupported.
They're still useful for something else: as diagnostics on your own site, read in context, to find pages where something is obviously broken. Worth doing whether or not any engine ever looks at them.
Why the numbers are ambiguous
Somebody lands, reads for twenty seconds, gets the answer, leaves.
Depending on your dashboard that's a bounce, a low time on page, and a failure. It's also a complete success — they came with a question and left with an answer.
There's no way to tell that apart from somebody who arrived, found nothing, and gave up. Both produce identical numbers. Any reading that treats one as obvious is adding an assumption the data doesn't contain.
Measurement artifacts
Time on page is usually computed from the gap between pageviews, so the last page of a session often records nothing. The page somebody read longest can register as zero.
Bounce rate counts single-page sessions, which mashes together satisfied-and-done with arrived-and-fled.
Click-through rate from search varies enormously by position, query type, device and what else is on the results page. Comparing yours to a published curve compares you to somebody else's query mix.
None of that makes the numbers useless. It makes them context-dependent, which is not how they get reported.
What they're actually good for
Comparisons inside your own site, over time, where the confounding factors stay roughly constant.
A page far below its neighbors. Worth a look — not because the number is bad, but because it's odd, and odd points at something.
A sudden change after a redesign, a template change or a content update. This is the most reliable use of engagement data and the least glamorous.
Scroll depth on long pages tells you where people stop, which tells you where the page stopped earning attention.
Used that way they're a smoke alarm. Used as a scorecard they're a target to be gamed.
The gaming problem
Make bounce rate the objective and there are well-known ways to improve it that make the site worse. Split the content across pages to force clicks. Auto-play something. Put an interstitial in the way of leaving.
Every one of those improves the number and degrades the experience the number exists to approximate.
That's the general case with proxy measures, and it's why the measurement section argues for outcomes — qualified leads, revenue per visit — over behavioral proxies.
What would change this page
A platform documenting engagement signals as ranking inputs, with specifics.
The claim has circulated for well over a decade without that. At some point the absence is itself evidence.
More in E-E-A-T & YMYL
What E-E-A-T is
It comes from the instructions Google gives the people it pays to grade search results. Knowing where it comes from explains both why it matters and why most advice about raising it is confused.
The Experience signal
The first E is the newest and the most neglected. It's also the only one you can't buy, study for, or fake — which is exactly why it's worth building content around.
YMYL
Topics where bad information costs somebody their money, health, housing or legal position get held to a higher bar. If you publish in one, the review process isn't overhead. It's the product.
Reputation & reviews
Reviews are among the strongest trust signals available and among the easiest to acquire illegally. The FTC rule on consumer reviews is in force, and several widespread marketing practices fall inside it.
Human vs AI content
The most argued-about question in search right now, split into the part Google has put in writing, the part people assert with no evidence, and the part nobody outside the AI companies can answer.
AI content workflow
How to use AI for the parts it's good at, keep it away from the parts it's bad at, and put a human check in the one place that matters.
Author bios
A byline says somebody stands behind this. That only means something if they're real, the credential fits the subject, and the same identity shows up everywhere else.
Backlinks
A link is somebody deciding your page was worth sending a reader to. Everything useful about links follows from that, and so does everything wrong with trying to manufacture them.
Mentions & citations
Getting named with no link attached is now a common kind of visibility. What it proves is clear. What any AI engine does with it is not, and this page keeps the two apart.