For growth teams
More customers from search. Proved, not claimed.
Edition House picks the page with the most to gain, ships one scoped change, and measures it against a baseline frozen before it went out. You get a graded readout — including the months where the change did nothing.
- Frozen baselines
- Search Console readout
- You approve every change
The problem
Organic is the only channel you can't run a clean experiment on.
Paid has CAC. Lifecycle has holdouts. Onboarding has a control group. Organic has a line that went up, twelve things that shipped that quarter, and no way to say which one did it. The channel that compounds hardest is the one you can defend least.
No baseline
By the time anyone looks, the "before" is gone. Nothing was frozen, so there is nothing honest to compare against.
No attribution
Twelve posts, one algorithm update and a seasonal swing all landed in the same quarter. Which of them moved the number?
No honest negative
Every tool in this category reports a win. None of them tell you a change did nothing — so nobody ever stops doing it.
The readout
Five verdicts. Two of them are bad news.
A result is graded on the strength of the evidence, never on the direction you hoped for. An untrustworthy reading outranks everything else, because a confident wrong answer is worse than no answer.
| Verdict | What it means | What you do |
|---|---|---|
| Proven | The move clears the confidence interval on the real sample size. | Keep it — or roll it back, if the proven direction is down. |
| Directional | Enough data to see a move, not enough to call it. | Wait for more data, or repeat the change on a similar page. |
| Inconclusive | Enough data to have detected a move. There was not one. | Stop. This change did nothing, and that is a finding. |
| Underpowered | Too few impressions to detect a change of the size you care about. | No verdict is the honest answer. We will not invent one. |
| Confounded | Something else moved at the same time. The reading is not trustworthy. | Discard the reading, not the change. |
Direction and grade are separate things. A confirmed regression is a proven result pointing down — and it can generate a rollback draft, reviewed exactly like any other change.
How the change is chosen
A position-6 page and a position-50 page are not the same job.
Live rank sets the strategy before a word is written. Each band gets its own editing budget, its own structural permissions, and its own definition of success.
Positions 4–10
Breakthrough
Already on page one. Close the gap to the top three with the smallest set of high-leverage edits.
Up to 3 sections rewritten · no new sections · nothing removed
Positions 11–20
Optimisation
Earn a page-one position: fix the sections that under-serve the searcher, fill genuine gaps, cut dead weight.
Up to 5 sections rewritten · new sections allowed · removals allowed
Positions 21–40
Early traction
A substantial quality lift — while protecting every query the page already earns.
Wider rewrite budget · gaps closed · dead weight cut
Positions 41+
Diagnostic
A diagnostic rebuild within the existing architecture — and if the topic itself is wrong, it says so instead of polishing the wrong page.
Largest budget · the diagnosis may conclude "wrong page"
The most useful thing this product says is "you are improving the wrong page". No tool that bills per article will ever tell you that.
Every tier shares one safety model: code-owned structure, hard and soft preservation, and deterministic assembly. The engine cannot quietly delete the paragraph that earns your traffic.
Methodology
The questions you are going to ask.
You run experiments for a living. Interrogate it.
- What is the baseline, and when is it frozen?
- It is captured before the change ships and it is immutable afterwards. A denominator that keeps moving is not a baseline, it is a story.
- How do you separate my change from an algorithm update or seasonality?
- Movement that coincides with a sitewide or category-wide shift is graded confounded, and confounded overrides every other grade. We would rather hand you no verdict than a flattering one.
- What if the page was already trending upward before the change?
- That is exactly what the confounded grade exists to catch. A pre-existing trend is not your change, and reporting it as one would be the easiest lie in this category.
- How much traffic does a page need before you will call anything?
- Enough impressions to detect a move of the size worth caring about. Below that bar the grade is underpowered, and you get told that instead of a number with false precision.
- How long until a readout?
- Typically four to twelve weeks, set by the page's impression volume. Anything faster is not measurement, it is noise. If your team needs a weekly experiment cadence, organic will frustrate you and we would rather say so now.
- Can I run this on pages my team is already editing?
- Not usefully. Concurrent edits confound the reading, and you will get a confounded grade rather than an answer. Pick pages you can leave alone for the measurement window.
- What happens when a change loses?
- A proven negative can generate a rollback draft, reviewed like any other change. Losses roll back; that is the point of running it as an experiment rather than a publishing schedule.
- Where does the finished change actually land?
- As a WordPress draft, as a GitHub pull request, or held in Edition House for review. You approve every one. More destinations as we can verify them properly — we will not claim a delivery path we cannot prove landed.
Fit
This works when there is something to measure.
A good fit
- Search Console history with real impressions
- Roughly 30+ pages, so choosing one is a real decision
- A WordPress or Git-backed site
- Someone who can approve a content change
- A content programme already running that nobody can grade
Not yet
- Shopify or a hosted-only CMS — no delivery path we can verify today
- A new site with no search history — nothing to measure against
- Teams who want content published without a human review step
- Anyone who needs a result inside a month
If that is you today, it may not be in six months. The constraint is the setup, not the company.
The offer
We do not have a case study yet.
If we had a graded result to show you, it would be at the top of this page instead of down here. We do not, so here is the honest version of the deal.
- 10 design partners, free for 3 months.
- You get the work: the page selection, the change, the delivery and the readout.
- We get to publish what happened — including the changes that did nothing.
- Nothing is anonymised without your say-so, and you can stop at any point.
You sell experiments internally. You already know most of them fail. That is the number we are trying to earn the right to show you.