Methodology

Precision isn't a feature. It's the foundation.

How we measure, and what we refuse to guess

Search Console is the most authoritative record of your search performance, but it is not a clean spreadsheet. It samples, it lags, it hides rare queries. This page explains exactly how SearchVault handles each of those realities so the numbers you act on mean what they appear to mean.

First-party only

Every number originates in Google Search Console, Google’s own record of your impressions, clicks and positions. Nothing is scraped from a results page and nothing is estimated from a competitor index.

Matched days, not matched labels

Comparisons align the number of days on both sides, so a 28-day February is never set against a 31-day January and a partial month is never compared with a complete one.

Totals from Google’s own aggregates

Property totals come from the aggregate row Google reports, not from adding up the visible query rows, so a long tail that the interface truncates cannot shrink your headline numbers.

No imputation

A day Google reported nothing for stays empty. We never zero-fill, interpolate or back-estimate a missing day to make a chart look continuous.

The data source

What Google does to the data before you see it

Four behaviours are built into Search Console itself. None of them is a flaw in your account; they are how the product works. A tool that ignores them will hand you confident numbers that are quietly wrong.

Sampling and row limits

For large properties Search Console can report sampled or filtered data, and its interface caps how many rows it will display or export in one view. A single export is therefore a partial view of a big site: the head terms are there, the long tail is truncated. SearchVault stores what Google reports and never pretends the visible rows are the whole picture. Where a figure is derived from a partial row set, we say so instead of presenting it as a property total.

Processing delay

Search Console data is not final the moment a day ends. Google keeps refining the last few days, so the most recent two to three days can still move. That is why a drop in the last 48 hours is not treated as a trend: SearchVault anchors its windows on the newest day that actually has data, and the header shows when the property last synced.

Anonymization and thresholding

To protect searcher privacy Google omits some queries entirely, and very low-volume queries can be withheld. Those impressions still count toward your totals but never appear as a named keyword. The result is a small, structural gap between the sum of named queries and the property total, which is exactly why we report totals from Google’s aggregate rather than from the keyword list.

A 16-month window

Google deletes Search Console history after roughly 16 months. Any comparison older than that is impossible inside Search Console itself. SearchVault keeps collecting and storing your data permanently, so year-over-year comparisons stay available after Google has forgotten them.

Comparisons

Matched days versus full periods

Most “your traffic dropped 40%” alarms are calendar artefacts. Comparing a partial period with a complete one, or a 28-day month with a 31-day month, produces a large percentage change that has nothing to do with performance. We align the two sides by day count and show the exact ranges used.

Month to date vs previous monthCompares 12 days with a complete 31-day month and reports a collapse.Compares the same 12 days with the first 12 days of the previous month, and states the exact date ranges used.
February vs JanuaryA 28-day month against a 31-day month, so clicks look down even when daily performance is flat.Either matched by day count or normalised per day, with the alignment mode named on the report.
A week with a missing dayThe missing day is treated as zero, which fakes a one-day crash in the trend line.The day stays blank, the comparison window is anchored on real data, and the report notes the gap.

On the Reports page you can switch the alignment explicitly: matched days, clipped to the same calendar window, or full period. The report always names the mode it used, so a percentage change can never be read out of context.

Totals

Why we keep sentinel rows

A Search Console export is a list of rows: one per query, one per page, one per country. Add those rows up and you get less than the total Google reports, because the export is truncated and because anonymized queries have no row at all. If a tool sums the visible rows, your headline clicks are simply too low.

The naive approach

Sum every query row in the export. The result silently excludes truncated rows and anonymized queries, so clicks and impressions come out understated, and the shortfall grows with the size of the site.

The SearchVault approach

We store Google’s own aggregate row alongside the detail rows and use it for every property total, KPI and trend. Detail rows are used only for breakdowns, and they are filtered out of keyword and page lists so an aggregate can never appear as a “keyword”.

Analytics

The same discipline, applied to your traffic data

Search Console tells you how you ranked. Analytics tells you what happened after the click: sessions, engagement and conversions. The analytics layer has its own ways of quietly distorting a number, so SearchVault reads it with the same rule it uses for Search Console: totals come from the platform’s verified aggregate, never from summing a breakdown.

Sampling

When a query covers more events than a processing threshold, Analytics returns a sampled estimate instead of an exhaustive count. SearchVault reads the sampled flag on every response, flags days that were sampled in Data Health, and never presents an estimate as if it were exact.

Thresholding

When identity signals such as demographics are enabled, Analytics withholds rows that could identify an individual. Those sessions still count toward the headline total but vanish from the breakdown, which is exactly why the headline has to be read from the aggregate rather than rebuilt from the rows.

Cardinality and the (other) row

When a dimension has too many distinct values, Analytics folds the overflow into a single (other) bucket, so individual landing pages or sources drop out of the detail while still counting in the total. We keep the detail for breakdowns and the aggregate for totals, so the (other) bucket can never shrink a KPI.

The naive approach

Build the dashboard from the breakdown export and call the sum of the rows a total. Sampling, thresholding and cardinality limits all pull that sum below the real figure, and the shortfall grows with the size of the property.

The SearchVault approach

We store a verified anchor row, the platform’s own reported totals for sessions, users, engaged sessions and key events, next to the detail. Every KPI reads the anchor, breakdowns are used only for breakdowns, and a day’s integrity is checked by comparing the anchor against the detail it should roughly reconcile with.

Pairing the two first-party records lets a query be followed to a ranking, to a session, to a conversion, with no scraped estimate anywhere in the chain.

Limits

What we do not claim

  • •Position is an average. Search Console position is weighted by impressions across devices, locations and personalization. It is not a single point-in-time rank for one searcher.
  • •There is no competitor data. First-party data describes your site only. Competitor and local grid visibility need a different source, and we do not pretend otherwise.
  • •Recent days are provisional. Anything inside Google’s processing window can still be revised. We label it rather than hide it.
  • •Estimates are labelled as estimates. Opportunity sizing such as striking-distance upside is a projection from your own CTR curve, and it is always presented as an estimate, never as a measured result.
Site Health

Link Intelligence, Index Health and Anomaly Detection

Beyond query-level analytics, SearchVault pulls structural data from three additional Search Console APIs and cross-references it with your query performance.

Link Intelligence (GSC Links API)

We pull external and internal link data directly from Google's Links report. Important caveat: Google samples backlink data heavily and updates it infrequently (often weeks behind), so link counts are directional, not exhaustive. We surface them alongside your traffic data so you can correlate link changes with ranking shifts, but we label every link metric as "Google-reported" rather than claiming completeness.

Index Health (URL Inspection API)

The URL Inspection API returns Google's actual indexing verdict for each URL: indexed, crawled but not indexed, excluded by robots, redirect, and more. Google enforces a rate limit of 2,000 inspections per property per day, so SearchVault prioritizes URLs by traffic volume and inspection age. High-traffic pages are checked first; low-traffic pages cycle through over multiple days. Each verdict is timestamped so you know how fresh it is.

Anomaly Detection (z-score baseline)

Anomalies are detected by comparing each day's clicks, impressions, position and CTR against the property's own 90-day rolling baseline. We compute the z-score (how many standard deviations from the mean) rather than using a fixed threshold, because a 20% traffic swing that is normal for a volatile site would be alarming for a stable one. Severity tiers: |z| >= 3 is critical, >= 2 is a warning, >= 1.5 is informational. All thresholds use your own history, not an industry benchmark.

Composite Site Health score

The overall Site Health score is a weighted composite: Index Coverage 35%, Links 25%, Crawlability 20%, Sitemap Health 20%. Each sub-score is computed from the API data above, and the weights reflect how directly each factor impacts discoverability. The score is a diagnostic tool, not a ranking factor, and we state that clearly in the UI.

Smart Reports and the White-Label Report Builder both draw from this same pipeline, so every insight in a client report uses the same methodology described above.

Ready for 100% data parity?

Every figure in SearchVault traces back to your own Search Console record, with the method stated. No estimates, no truncated tables.