Methodology
QueryRev AI methodology How every number is calculated, labelled and verified.
QueryRev AI labels every revenue figure as one of four kinds. Known revenue is recorded in your store or CRM. Influenced revenue is revenue where organic search took part. Modeled revenue is an estimate, shown as a range. Incremental revenue is the lift above a modeled baseline after a change. A figure of one kind is never shown as another.
Every worked figure on this page is sample data, and is labelled so.
Updated
Value layers
Choose a kind. Its definition, where the figure comes from and the arithmetic follow.
Known Revenue
Known revenue is revenue recorded in your own store or CRM for journeys that began with organic search, under the attribution rule you selected. It is counted, not estimated.
Influenced Revenue
Influenced revenue is revenue from journeys in which organic search took part but was not necessarily the last touch. It is measured, and it is never shown as known revenue or added to it.
Modeled Revenue
Modeled revenue is an estimate of value that has not been observed, such as the upside of an opportunity, calculated from measured inputs. It is always shown as a range, never as a single point.
Incremental Revenue
Incremental revenue is the lift above a modeled baseline after a change: what happened, minus what the page would have earned anyway. Because the baseline is a range, the lift is a range too.
Revenue as Google Analytics 4 attributes it has a label of its own, GA4-reported. It is never shown as known or as influenced revenue.
In AI search, a visit from an AI answer that analytics can observe is observed value. Visibility in an answer that produced no click is modeled, and is shown as modeled.
Revenue at risk, revenue opportunity, expected value and protected revenue are views built on these kinds, and every figure in them carries the mark of its kind. The screens that show them are described under Revenue Intelligence.
Labels
Every number in QueryRev AI says how it was obtained. Three words cover it, and they are not interchangeable.
Measured
1,310 impressions· Search Console · final dataA measured number is a count read from a source as that source reports it: clicks from Search Console, sessions from Google Analytics 4, orders from your store, status codes from the crawl. It carries its source, its dates and how it was obtained.
Estimated
40 leads × $300 = $12,000· Expected value, before revenue closesAn estimated number is a measured count multiplied by a value that is stated beside it, such as the expected value of a lead. The count happened; the value per unit is a stated, probabilistic input for revenue that has not closed yet, so the result is an expectation, not revenue.
Modeled
$5.8k–8.1k a month· Opportunity · a rangeA modeled number is a projection of something that has not been observed: the upside of an opportunity, the baseline a change is compared with, visibility in an AI answer that produced no click. It is always a range, with the strength of its evidence beside it.
When the source itself is partial
A measured number is only as complete as its source. When Google reports that a day was sampled, thresholded or folded into an “(other)” row, or when a day reaches the row ceiling, the number carries that flag instead of looking complete. Clicks that Search Console counts for a page but assigns to no query are shown as “queries hidden by Google”, not spread over the queries that are known.
Search Console days
Search Console data normally appears two to three days after the day it describes, and Google can still revise it. QueryRev AI stores final data only, and goes back for revisions.
Final data only
A day is stored once Google reports it as final. Fresh data that is not final yet is not shown as if it were complete.
The dates are Google’s
Search Console days are in Pacific Time, as Google defines them, and are stored without conversion. Google Analytics 4 days are in the property’s own time zone.
Recent days are read again
Every day the last three stored days are read again, with any of the last ten days whose site total no longer matches Google’s. Once a week the same comparison runs over 90 days.
Three levels of detail
Site totals, page totals and page-by-query rows are stored separately, because Google leaves rare queries out of the detailed rows. Without the totals, sums would never match Search Console.
Google Analytics 4 settles later
The last four complete days are read again each day, because Google Analytics 4 may change data for up to 48 hours.
History is kept
The first connection loads the 16 months Google offers, newest first. After that every day is added, so the history grows past what Google itself keeps.
Managed URLs
A managed URL is the canonical, working HTML page of a site you connected that is indexable or had search traffic in the last 90 days. Plans are measured in managed URLs.
A URL is managed when all four are true
It is the canonical version. Not a redirect, not canonicalised to another URL, and not a tracking-parameter variant.
It works. Its last successful fetch answered 200 with HTML, and it is not a soft 404.
You have not excluded it. It matches none of your exclusion patterns.
It is indexable, or it had search traffic. No noindex and no robots.txt block; or at least one Search Console impression or one organic session in the last 90 days.
The last rule is deliberate. A page that drew search traffic and then got a noindex or a robots.txt block by mistake is exactly the page that has to be flagged, so it stays in the count.
Before any of this, every URL is normalised by one rule: known tracking parameters such as utm_source and gclid are removed from the query string, and so is any query parameter whose value contains “@” or is 32 or more hexadecimal characters. The path is not inspected. The http and https versions of a page, and its www and bare-host versions, are different URLs; they are joined by the redirects and canonicals the site declares, never by assumption.
The count is taken once a day for each site, at 00:00 UTC, and a workspace’s figure is the sum of its sites. Whether a plan is measured against the peak or the average of a billing period is not decided yet; it will be stated on the pricing page.
Verification
Published is not an outcome. A change is checked in four layers, and its effect is the difference between what happened and a modeled baseline: what the page would have earned anyway.
Technical
The change is live, valid and accessible on the page.
Search
Search has responded: the page was recrawled and indexed, and ranking and click-through moved. For AI search, the answers are checked again as well.
Business
Traffic, conversions, leads, orders and revenue are measured over the verification window.
Outcome
The result is classified: successful, neutral, monitoring, failed or rollback candidate.
The baseline is modeled, so it is a range, and so is the lift. Neither is shown as a single certain figure. An agent or a provider reporting that a task is complete means only that execution finished; it says nothing about the outcome.
Stronger causal methods are used where the data and the sample size allow them, and where they do not, the result says so rather than claim a cause.
Confidence
A language model can say it is 95% sure and be right far less often. Until a model’s stated confidence has been calibrated against real outcomes, QueryRev AI does not show it as confidence and does not let it drive automation.
What stands in its place is evidence strength, computed from data: the measurements behind the decision, the results of the validators, and the risk class of the action. A policy threshold reads that, not the model’s own figure.
The model’s stated confidence is still recorded with every decision, with the provider and the model version, so that it can be calibrated later, one decision type at a time, against verified outcomes and human review. What can be computed from data is computed, not asked of a model.
The evaluation harness that does this is part of QueryRev Intelligence, which is built on frontier AI models from a major US provider, behind our own Decision Gateway.
Limits
The rules a number has to pass before it is shown, written as refusals.
A modeled figure as a known one
Known, influenced, modeled and incremental revenue are never added into one total or shown under one label.
A point where there is a range
A modeled value is shown as a range. The midpoint alone is never presented as the answer.
Published as done
A change that was applied is not an outcome. It counts when it has been verified.
A model’s confidence as ours
Stated confidence is not product confidence until it is calibrated against outcomes.
GA4’s attribution as known revenue
Revenue as Google Analytics 4 reports it keeps its own label, GA4-reported.
A cause the data cannot support
Where a lift cannot be separated from the baseline, the result is neutral or still monitoring, not a win.
Methodology FAQ
Short answers first. Ask anything else at business@queryrev.com.
What is the difference between known and modeled revenue?
Known revenue is recorded in your own store or CRM for journeys that began with organic search. Modeled revenue is an estimate of value that has not been observed, such as the upside of an opportunity, and is always shown as a range. QueryRev AI never shows a modeled figure as a known one.
What are the four kinds of revenue in QueryRev AI?
Known, influenced, modeled and incremental. Known is recorded revenue from organic journeys. Influenced is revenue where organic search took part without being the last touch. Modeled is an estimated range. Incremental is the lift above a modeled baseline after a change.
What do measured, estimated and modeled mean?
Measured is a count read from a source as that source reports it. Estimated is a measured count multiplied by a stated value, such as leads times the expected value of a lead: an expectation, not closed revenue. Modeled is a projection of something that has not been observed, always shown as a range.
How does QueryRev AI count a managed URL?
A managed URL is the canonical version of a page on a connected site that answered 200 with HTML, is not a soft 404, is not excluded by you, and is indexable or had search traffic in the last 90 days. The count is taken once a day for each site.
How does QueryRev AI verify that a change worked?
In four layers: the change is live, search responded, the business result moved, and the outcome is classified. The effect is the difference between what happened and a modeled baseline, so it is reported as a range.
Why does QueryRev AI not show an AI model’s confidence?
Because a model’s stated confidence is not reliable until it has been calibrated against real outcomes. Until then QueryRev AI shows evidence strength computed from data, and records the model’s figure only so it can be calibrated later.
Keep exploring