Real SEO A/B tests. Controls suggested for you.
An experiment without a control isn’t an experiment. It’s a memory of a deploy. SERPVoyager picks a matched control for you, so test and control move side by side. The deploy stops being a guess.
Title rewrite · comparison-style on /running-shoes/ PDPs
Avg Pos
−1.4
vs control: −0.1
Traffic
+18.6%
vs control: +1.2%
The control group is the experiment.
Almost every “SEO test” you’ve read is a before/after on one URL. The deploy lines up with a Google update on day three, traffic moves, nobody knows whether it was your change or the algorithm. The post still gets shared.
The fix is a control group that matches the test on the dimensions that decide ranking volatility: search volume, page template, category neighborhood, and starting position. When Google sneezes mid-test, the control catches the same cold. The treatment’s outperformance is what you’re left to explain.
The hard part has always been picking that control by hand. SERPVoyager scores every other keyword in your project (vol match + position match + page-type bonus + category bonus) and presents the top candidates. You confirm with a click.
- best running shoes for marathons1.921,180 vol · Guide · running-shoes · #12
- asics gel-kayano 31 review1.811,420 vol · Product · running-shoes · #10
- saucony endorphin pro review1.74980 vol · Product · running-shoes · #13
- 4best stability running shoes1.621,310 vol · Guide · running-shoes · #9
- 5hoka clifton 10 review1.481,090 vol · Product · running-shoes · #14
The auto-suggested control panel. The top three are auto-picked; the rest stay in the list for one-click swaps if you want a different mix.
Seven change types, one workflow.
The change-type field exists so reports filter cleanly: your title-tag tests, your schema tests, your page-speed tests — each lined up next to its own history.
Title tag
Comparison-style vs benefit-style. Length tweaks. Brand prefix in/out.
Meta description
Coverage rate, CTR delta, whether Google still rewrites the description after your edit.
Content
On-page rewrites, FAQ blocks, comparison tables, intro length, schema-rich answers.
Internal linking
Hub-page adjustments, sidebar links, footer changes, anchor-text variants.
Schema
Product, Article, FAQ, HowTo markup. Eligibility for rich results and AIO citations.
Page speed
LCP, INP, CLS tied to ranking deltas. Useful for finding the page-type-wide payoffs.
Other
Anything you can deploy and roll back. Custom change type with a free-text description.
Keyword set or target page
Run a test on a curated keyword set, or on a target page (and let the keyword set be inferred from the URLs you choose). Same workflow either way.
Duration: 7 d min · 28 d typical
Four neighbors that make experiments trustworthy.
Experiments inherit the data model, the daily cadence, and the click-weighted scoring used everywhere else in the product. The result is one number that reconciles with the rest of your dashboard.
Page-type matching depends on the URL model.
“Same page type” is meaningless on a flat keyword list. The control suggester only works because every keyword carries its page type, primary category, and URL.
Daily rankings, daily verdicts.
Test and control track the same daily crawl that powers rankings and SoV. Day-by-day chart, not weekly buckets that hide the inflection point.
Traffic impact uses the published CTR curve.
The “+18.6% traffic” number on the result table is volume × ΔCTR using the same shared curve as SoV and Movers. The three reports reconcile.
Wins surface in Movers automatically.
A winning experiment is usually visible as a cluster of same-page-type movers a week before the result table calls it. Both views show the same underlying lift.
The four questions we get every time a team plans an experiment.
How long should an experiment run?
Minimum 7 days (a week of SERPs absorbs most weekly cyclicality), typical 28 days, longer for low-volume keyword sets. Anything shorter than a week and normal day-to-day SERP noise eats your signal. The wizard enforces the 7-day floor; the default is 28.
What about confounders — a Google update mid-test?
That’s exactly what the control group is for. When Google rolls out an update, both groups absorb it equally; the treatment’s residual outperformance is what you’re measuring. Without a control, you’re reading raw noise and calling it a win.
Can I pick the control myself?
Yes. The auto-suggester returns the top candidates ranked by score, but every candidate is one click away from being in or out. You can also ignore the suggester entirely and hand-pick the control set — the rest of the workflow (variants, baseline, daily readout, completion outcome) is the same either way.
How is the outcome decided — automatically or by me?
By you. When the experiment ends you mark it Positive, Neutral, or Negative with a free-text notes field. The dashboard surfaces “needs review” when an active experiment runs past its end date so nothing gets forgotten, but we never auto-conclude. Outcome lives next to the chart so future you can read why past you decided.
Test your next SEO change. With a real control group.
Plug in your keywords today. Pick a scope and the change you’re shipping; SERPVoyager scores every other keyword on volume, page type, category, and position to suggest a matched control. Watch test and control move side by side.
14-day free trial. No credit card.