Hypothesis-Driven SEO
Generate hypotheses by analyzing sites (and asking their SEO managers), then test to validate.
- Difficulty
- Moderate
- Time to result
- ~months to results
- Steps
- 4
- Confidence
- 90%
SEO advice is contradictory and constantly changing, so treat every claim, from Google, other SEOs, or your own hunch, as a hypothesis to be validated, not a truth to adopt. The highest-value activity is analyzing your own and competitors' site data to generate qualified hypotheses; testing merely validates them. A powerful shortcut is contacting the SEO manager at a consistently-winning site and asking what they've already tested.
Origin
Ethan Smith's operating philosophy, including his international SEO strategy built by studying TripAdvisor and interviewing its SEO lead Luke Levesque.
Core principles
- 01Data analysis of your and competitors' sites generates hypotheses; testing only validates them
- 02Value comes from analysis and observation, not brainstorming or blind testing
- 03Don't test before you've launched enough to have something to measure
- 04Watch leading indicators of ranking progress before scaling
How to run it
- 1
Analyze sites to form hypotheses
Pull competitors' top pages and keywords via ahrefs/semrush, look for patterns correlating with success (e.g. longer word count), and study your own best/worst pages.
Pro tip Interview the SEO manager at a repeatedly-winning site; they may already have tested what you're wondering about, saving you the experiment.
- 2
Launch before testing
If you're starting from zero, build and launch pages first; testing is a month-six-plus activity, not a starting move.
Watch out Testing with nothing launched wastes effort; there's no signal to read.
- 3
Watch leading indicators
Track whether new pages rank at least #20 for anything, then climb #15 to #12 to #8 to #5. Rising positions on a few pages predict you can scale the pattern to many.
Pro tip If 10 articles reach #5 for their keywords, that's evidence 1,000 more can too.
Watch out It's common to wrongly conclude failure; a first attempt that doesn't rank usually means the strategy was mis-implemented or authority is too low, not that SEO can't work.
- 4
Test to validate, then scale
Run a split or sequential test to confirm the hypothesis, then scale the validated pattern across the topic set.
Pro tip Bucketing is the main work in testing; a random sample of pages is a fine control, no special tool needed.
In the wild
Smith analyzed which companies repeatedly succeeded outside the US, found TripAdvisor consistently outperformed, formed hypotheses about why, then tracked down and interviewed TripAdvisor's SEO lead Luke Levesque who answered many of them directly.
→ He obtained validated answers without having to run the tests himself, because someone had already built and tested it.
Common mistakes
Concluding failure after one attempt
A first try that doesn't rank is usually a mis-implemented strategy or insufficient authority, not proof SEO won't work; abandoning it is a false negative.
Testing before launching
Running experiments from zero, with nothing built, produces no signal and burns time that should go into building pages.
Is it for you?
Best for
SEO practitioners drowning in contradictory advice who need a way to decide what to believe
Not ideal for
Teams that haven't yet launched enough pages to have data to analyze
From the transcript
“the test validates the hypothesis”
“the hypothesis comes from analyzing other people's sites and analyzing my own site”
“contact the person who runs SEO at that company and then set up a meeting”
“are we ranking at least number 20 for anything”
“it's common to have uh to incorrectly conclude failure”
From the episode
The ultimate guide to SEO
Ethan Smith (Graphite)