I don’t know yet — and I’m not going to pretend I do. Both twin sites are live, with six mirrored articles published. Search Console is verified on both properties, indexing is underway, and measurement (GA4 + GTM) is running. Baseline comes next. The one-line result will sit here when I’ve actually earned it.
- Keyword & twin sites locked (private — not named here)
- Content live — 6 mirrored articles, post-name URLs
- GSC verified · indexing requested · GA4/GTM live
- Baseline period ← you are here
- Coin flip → engagement treatment vs control
- Results
The idea
You’ve heard the claim a thousand times: Google rewards pages people stick around on — more clicks from search, more time on the page, deeper scrolling — and that engagement somehow feeds back into rankings. It’s repeated everywhere. Google has never confirmed it.
Most of what gets called “proof” is correlation. Pages that already rank well also tend to look engaged. That doesn’t tell you which way the arrow points.
So I built twins. Site A and Site B are nearly identical on purpose. After a quiet baseline, only one site gets real engagement from search. A coin flip decides which. Then I watch Search Console and see what actually moves.
Hypothesis
On two near-identical new sites (Site A and Site B), deliberately raising on-page engagement — scroll depth, clicks, and time on page — on the treatment site alone will improve that site’s Google rankings and Search Console clicks and impressions relative to the control over the same period.
Null: Engagement differences will not produce a meaningful ranking or Search Console advantage once content, hosting, and technical SEO are held constant.
I’ll judge after a shared baseline and a coin flip that assigns which twin gets the engagement treatment. Search Console is the scoreboard; GA4 only checks that the treatment actually stuck.
I’m keeping the exact sites and keyword private until this ends. Outside engagement would skew the test, and I like my experiments uncontaminated.
How I’m setting this up
If you’re not an SEO, this is the full path — every real decision from idea to “live and waiting,” with what I did and why. Fun, digestible, no secret black box. Domains and the exact keyword stay unnamed on purpose (see the end).
1. I picked the topic on purpose
What I did: Both twins are Nature/science sites — curious, factual, not a hustle niche.
Why: Low commercial heat. I’m testing a ranking theory, not building an affiliate funnel. A money-page niche would pile on other ranking factors (offers, ads, link bait) and make engagement harder to isolate.
2. Domains with equal brand weight (names stay private)
What I did: I chose two domains with the same kind of words on both sides — equal brand weight in the name. Neither twin got a juicier exact-match domain than the other.
Why: Domain wording can be a free SEO edge. If one name were obviously “better” for the keyword, any ranking gap later could be the domain, not engagement. Equal weight keeps that off the board. (The actual strings stay off this page — see step 12.)
3. Same stack — but a split footprint
What I did: Same host plan, CDN, theme, and caching approach on both twins. Under the hood they’re still separate: different IPs, different locations, and different nameserver paths — one behind Cloudflare DNS, one on registrar DNS.
Why: For this experiment I want hosting quality held roughly constant so speed isn’t the hidden variable. But I also don’t want Google (or a DNS/CDN quirk) treating them like one blob. Separate IPs, locations, and DNS paths keep the twins clearly two sites. Fast-vs-slow hosting is a later experiment (Experiment 002) — not this one.
4. Same CMS and builder
What I did: WordPress + Oxygen on both Site A and Site B.
Why: Template and tech parity. If one twin ran a different builder or theme stack, a ranking gap could be “better templates,” not engagement.
5. Twin sites, six articles, three mirrored topics
What I did: Two near-identical sites. Same three questions on each — six articles total. Wording differs per site (not copy-paste clones). One topic on each side is a deeper science piece on how the mechanism works — not a kids explainer.
Why: Twins so the only intentional difference later is engagement. Six pages / three topics so each side behaves like a small real site, not a lonely one-pager. Mirrored topics keep the comparison fair. Unique copy avoids “duplicate content twins.” Same intent depth on the science piece so one side isn’t accidentally “thinner.”
6. Same structure — no cross-links
What I did: Parallel layouts. Sidebars only link to other articles on the same site. No links from Site A ↔ Site B.
Why: Cross-links would leak traffic and authority between twins. Then you couldn’t tell whether rankings moved because of engagement or because one site kept feeding the other.
7. Separate measurement
What I did: Each twin gets its own GA4 / GTM and its own Search Console property.
Why: Two clean scoreboards. Analytics checks whether the engagement treatment stuck. Search Console is the ranking/clicks/impressions scoreboard. One mixed bucket would hide which twin did what.
8. Published together — clean URLs, findable by Google
What I did: All six articles went live as one set. Pretty (post-name) URLs. Search Console verified on both. Sitemaps in play; indexing requested.
Why: Indexing and age parity. A staggered launch would give one twin a head start that looks like a treatment effect. Clean URLs and sitemaps help Google discover both sides the same way before anyone gets treated.
9. Quiet baseline — then the coin flip
What I did: Both twins sit live with the same setup for a quiet “before” period. Only after that do I randomly assign which twin gets the engagement treatment and which stays control. Treatment is never baked into the article copy.
Why: Baseline is the before photo. Without it I can’t honestly say what changed. The coin flip after baseline stops me from picking a favorite early or writing one site to “win.”
10. When treatment starts, engagement has to be real
What I did: The plan (after baseline) is real search-driven engagement on one twin — not bots, VPN farms, or click farms.
Why: Fake engagement would “prove” nothing useful. If the theory is about how Google reacts to real people from search, the treatment has to look like that.
11. What’s not in this experiment
What I left out on purpose: Changing UX or page design on only one twin. Unequal hosting speed (that’s Experiment 002). Baking “engagement tricks” into the articles themselves.
Why: Those would be different tests. Experiment 001 is about engagement after a fair baseline — not a redesign bake-off or a speed bake-off.
12. Kept private for now
What I did: Real domains and the exact target keyword stay off this public page until the experiment ends.
Why: If readers (or bots) pile onto one twin on purpose, that outside engagement contaminates the test. Privacy here isn’t secrecy for fun — it’s so Site A and Site B stay comparable.
What I’m measuring
Two layers — did the treatment stick, and did rankings move?
Engagement
GA4 / GTM — behavior on each twin
- Time on page / engagement time
- Scroll depth
- Pageviews and pages per session (including second pageviews)
- Engaged sessions
Search outcomes
Search Console — what showed up in results
- Impressions
- Clicks
- SERP click-through rate (CTR)
- Average position
I’ll report what changed after the coin flip — not vibes, not screenshots of a lucky week. If engagement moves rankings, the numbers should show it. If they don’t, that’s an answer too.
What happens next
When baseline settles, I flip the coin, run the engagement treatment on one twin, and find out whether engagement actually moves rankings — or whether a lot of marketing Twitter owes me an apology.
No spoilers. Just the result, when it’s real.
Get the result by email
When I know whether engagement actually moves rankings, you’ll hear it first.
One message when Experiment 001 finishes. Not a newsletter saga.
Does a faster server rank better?
Fast host vs slow host with identical content — server speed against rankings.