Skip to content
Join the waitlist

Help Centre

Verification and Outcomes

A change is done when its effect is verified, not when it is published. When a decision is approved, QueryRev freezes a plan of what it will measure. Once the change is live, it compares the changed pages with what would have happened anyway, and the outcome says Successful, No Effect or Failed only after the final read.

By QueryRev AIPublished Updated3 min read

Key takeaways

  • The plan of what will be measured is frozen when the decision is approved.
  • An interim read is shown so a large loss can be caught early; it is never a result.
  • Before-and-after measurement adjusted for the season is low confidence and is never shown as Verified.
  • Every read is kept: a later read never rewrites an earlier one.

The plan, frozen at approval

When a decision is approved, its plan fixes what will be measured: the primary metric, such as clicks to the changed pages, the guardrails, the smallest effect that counts, the loss that would make it a rollback candidate, and the days of the first and final reads. Fixing it before the result is known means the method cannot be picked to suit the result.

The window starts the day after the change goes live. Depending on the kind of change, the final read comes 28 to 90 days later; a change aimed at AI answers is read over weekly checks instead of days.

How the effect is measured

Each outcome names its method, so you can see how strong the result can be.

The measurement methods an outcome can use.
  • Split test

    How it works
    The changed pages against a frozen group of comparable pages that were not changed.
    Can it be Verified?
    Yes
  • Counterfactual forecast

    How it works
    The page against a forecast made from unchanged comparable pages, the rest of the site or the page a year earlier.
    Can it be Verified?
    Yes
  • Before and after

    How it works
    The page before and after the change, adjusted for the season.
    Can it be Verified?
    No: low confidence
  • Checked live only

    How it works
    The change is too small for any method on this site; QueryRev checks it is live and correct.
    Can it be Verified?
    No: not measurable

A split test compares like with like at the same time, so a site-wide trend or a Google update moves both groups and cancels out. A method whose backtests have not passed shows at most Monitoring.

Interim and final reads

While a change is measured, its status reads Measuring, with the day of the window. An interim read is an estimate on the way: it is shown so a large loss can be caught early, and Successful, No Effect and Failed wait for the final read. If the data of a read is not complete yet, such as a day Search Console has not finished, the read waits for it.

What Else Happened lists the confounders found in the window, such as a Google update or another change on the same page, breaking ones first. Technical Checks confirm the change is live as approved and was recrawled by Google.

What each outcome means

The outcomes of a measured change.
  • Done · Successful

    What it means
    The whole range of the estimated effect is above zero.
  • Done · No effect

    What it means
    The whole range is within the smallest effect that counts, either way.
  • Done · Failed

    What it means
    The whole range is below zero.
  • Needs you · Rollback Candidate

    What it means
    The effect is below zero by the rollback threshold or more. A person decides whether to roll it back.
  • Needs you · Revenue fell

    What it means
    Search rose, but revenue on these pages fell. A person should look at why.
  • Closed · Inconclusive

    What it means
    The data could not tell, for the reason given, such as an interval too wide either side of zero.

When another change landed on the same page in the window, the result is joint: it belongs to both decisions, says so, and neither has an Incremental figure of its own. Every outcome keeps all its reads; a later read never rewrites an earlier one.

Search, AI search and business results

An outcome page shows the search result first, then the AI search result where the change aimed at AI answers, then the business result: the revenue of the page, on two lines, never folded into one. The difference against the modeled baseline is incremental revenue, and it is labelled Incremental, apart from Known revenue from your store and from Modeled figures.

Open an outcome from its decision or its change. What This Teaches counts how the verified changes of the same type on the site have turned out so far.

Cite this article

A suggested citation. Its link goes to this page, whose address does not change.

QueryRev AI. “Verification and Outcomes.” QueryRev AI, 7 Nov 2026. https://queryrev.com/docs/verification

Read next