Measurement frameworks for small-business websites

A measurement framework for a small-business website is the structured set of instruments, signal definitions, and reporting conventions used to decide whether the site is doing its job. Where the companion entry Decision-linked metrics versus vanity metrics addresses which metrics deserve a place on a dashboard, this entry addresses how those metrics are produced — the tools that collect them, the layers that organise them, the leading-to-lagging ladder that orders them in time, and the editorial discipline required to make data legible to a non-technical operator.

The frameworks documented here are drawn from primary Google documentation, independent research (Pew Research Center, Ahrefs large-N studies, peer-reviewed work), and a tightly-flagged band of vendor sources whose incentives are noted in line. They are calibrated for the small-business case: a site with one operator, modest traffic, and an owner who needs to know whether to keep investing.

The two-layer model: search visibility vs business outcomes

The foundational distinction in website measurement is that search visibility and business outcomes are two different questions, answered by two different tools, and must not be merged into a single chart.

• Layer 1 — Search visibility. Answered by Google Search Console (GSC). The Performance report exposes exactly four metrics: Impressions (how often a link to the site appeared), Clicks (how often someone clicked through), CTR (clicks divided by impressions), and average Position (an impression-weighted average of the topmost result's position). The question this layer answers is Is Google surfacing this site, and where?
• Layer 2 — Business outcomes. Answered by web analytics — Google Analytics 4 (GA4), Matomo, Plausible, Umami, or equivalent. The question this layer answers is Is the site producing business? This is where engagement, conversions ("key events" in GA4 vocabulary), and revenue live.

Source: Google Search Console Help; GA4 official Help. Confidence: Verified. Caveat: Blurring these two layers is the single most common analytics error in small-business reporting.

The reason the layers must stay strictly separate is mechanical, not stylistic. Google has stated repeatedly that it does not use Google Analytics data — including bounce rate — in its ranking algorithm. John Mueller, Webmaster Central, June 2022: "I think there's a bit of misconception here that we're looking at things like the analytics bounce rate when it comes to ranking websites, and that's definitely not the case."

Source: John Mueller (Google), Webmaster Central, June 2022. Confidence: Verified (direct first-party quote).

GSC describes Google's behaviour toward the site; GA4 describes visitors' behaviour on the site. A trend chart that mixes "impressions" and "engagement rate" produces incoherent reads in both directions: it implies that GA4 metrics are a search signal (they are not) and that GSC metrics are a business outcome (they are not). The operational rule, applied to every client dashboard, is that visibility and business outcomes get separate panels, separate trend lines, and separate narratives.

The instrument stack

A small-business measurement framework is built from a small number of well-understood tools, each pointed at a different question.

Google Search Console (visibility, free, first-party)

GSC is the canonical instrument for Layer 1. It reports impressions, clicks, CTR, and average position, plus the Pages / Indexing report (which URLs Google has discovered and stored) and URL Inspection (a live test that probes the index for a single URL). The Performance report is the daily working surface; the Pages report is the periodic health check.

GSC has known limitations that any practitioner must hold in mind — they are catalogued in the GSC instrument realities section below.

Web analytics (GA4, Matomo, Plausible, Umami)

Layer 2 is provided by a web-analytics tool. GA4 is the dominant free option and produces engagement-rate, engaged-sessions, key-event, and revenue metrics. Matomo and Plausible are privacy-respecting alternatives; Umami is a lightweight self-hosted option. The choice between them is primarily a privacy / data-sovereignty decision — the substantive measurement questions are answered similarly across all four.

Server logs (ground truth on crawl behaviour)

Server logs record every HTTP request the origin received, including Googlebot fetches. When GSC reporting lags or disagrees with reality, raw server logs are the ground truth on whether Google actually crawled a URL and when. Server logs are also the only instrument that captures bot traffic Google does not report.

URL Inspection live test (real-time index probe)

The URL Inspection live test in GSC bypasses the reporting cache and queries Google's current index for a single URL. It is the third leg of the cross-check stool when the Performance and Pages reports are stale.

Page-speed and Core Web Vitals tools

Page-speed measurement uses PageSpeed Insights, Chrome User Experience Report (CrUX), the Search Console Core Web Vitals report, and field-data tools that surface real-user LCP, INP, and CLS. The thresholds and instrument caveats are detailed in the Page-speed measurement section below.

The six-rung SEO ladder: leading to lagging signals

SEO measurement is best organised as a ladder of signals that move sequentially. Each rung is a precondition for the next, and each rung is measurable on a different cadence. Treating any single rung as the success metric in isolation produces a vanity read; treating the ladder as a whole produces a leading-to-lagging diagnosis.

Rung 1 — Crawled / Indexed

What it measures. Whether Google has discovered and stored the site's pages. Where measured. Search Console (Pages / Indexing report, URL Inspection). How early it moves. Earliest — days to a few weeks. High-authority sites index in 24-48 hours; new small sites typically 1-4 weeks. Trustworthiness. High as a go/no-go gate (no indexing means nothing else can happen), but it proves only eligibility, not success.

John Mueller: "Most high-quality content is typically indexed within about a week."

Source: Google Search Console documentation; John Mueller via Search Engine Journal. Confidence: Verified. Caveat: Indexing is not guaranteed. No indexing by week 4 indicates a crawlability problem, not patience.

Rung 2 — Impressions

What it measures. How often the site's URLs appear in search results. Where measured. GSC Performance report. How early it moves. Very early — days to weeks after indexing; rises first for long-tail and low-competition queries. Trustworthiness. Moderate — confirms Google is starting to surface the site, but inflated by non-target queries and (historically) by bot impressions. Growth in impressions is not growth in business.

Source: Google Search Console Help; Performance report documentation. Confidence: Verified (metric definition); Industry-consensus (interpretation). Caveat: Two recent reporting events muddy impression trend reads — the September 2025 &num=100 retirement and the May 2025-April 2026 impression logging bug, both detailed below.

Rung 3 — Query count growth

What it measures. The number of distinct queries generating impressions — a proxy for topical reach. Where measured. GSC Performance, query dimension. How early it moves. Early — weeks. Trustworthiness. Moderate — breadth shows topical reach expanding, but counts low-intent queries equally with commercial ones.

Source: Google Search Console Performance report documentation. Confidence: Industry-consensus.

Rung 4 — Average position for non-brand queries

What it measures. Where the site ranks specifically for discovery (non-brand) terms — isolated from brand demand. Where measured. GSC Performance, using the native branded-queries filter Google announced on November 20, 2025 and expanded to all eligible sites on March 11, 2026, or a regex brand exclusion. How early it moves. Weeks to months. Trustworthiness. Higher than raw position, because filtering out brand terms isolates SEO discovery from existing demand.

Per Search Engine Land, approximately 44% of Google searches are branded — meaning raw-position averages overstate discovery health for any brand with notable traffic.

Source: Google Search Console branded-queries filter announcement (Nov 20, 2025); Search Engine Land on the ~44% branded share. Confidence: Industry-consensus. Caveat: Position is an impression-weighted average across users, geographies, and devices — not a fixed rank.

Rung 5 — Clicks / CTR

What it measures. Actual visits delivered from search, and the rate at which impressions convert to clicks. Where measured. GSC Performance report. How early it moves. Follows position — months. Trustworthiness. Higher — clicks are real traffic; CTR diagnoses title and snippet quality.

Source: Google Search Console Help; Ahrefs research on AI Overviews and CTR. Confidence: Verified (metric); Industry-consensus (zero-click effect). Caveat: Zero-click SERPs and AI Overviews break the rank-to-click link. Ahrefs associates AI Overviews with a 58% lower average CTR for the top-ranking organic page. This is a vendor-flagged single-source figure; the directional finding is consistent with the broader 2025-2026 zero-click trend, but the magnitude is not independently corroborated.

Rung 6 — On-site engagement

What it measures. Whether arriving visitors do something meaningful — scroll, navigate, dwell, micro-convert. Where measured. GA4 (engagement rate, engaged sessions, key-event rate). How early it moves. Immediate once clicks arrive. Trustworthiness. Mixed — useful as a diagnostic of traffic-content fit, weak as a success proxy because GA4's engagement definition is arbitrary (a 10-second default threshold).

Source: GA4 official Help. Confidence: Single-source / Directional for any causal claim from engagement to conversions. Caveat: The causal link from on-site engagement metrics to conversions and revenue is unproven in the general case. Engagement is diagnostic, never the primary KPI.

The implicit Rung 7 — conversions and revenue — sits in GA4 plus CRM and is the only rung that maps directly to business value. See Decision-linked metrics versus vanity metrics for the conversion / vanity distinction.

GSC instrument realities

Search Console is the most important free instrument in the stack and the most frequently misread. Five mechanical facts shape every legitimate use of GSC data.

Average Position is an impression-weighted average, not a fixed rank

GSC's average Position is an impression-weighted average of the topmost result's position across users, geographies, devices, and time. Google's documentation is explicit that position "is an average," that data is attributed to the canonical URL, and that newest data may be preliminary.

A reported "position 4.2" is the average of every impression event in the window. It does not mean the site appears at position 4 or 5 for any specific query at any specific time. Single-keyword rank tracking is misleading for the same reason: position fluctuates by query, geography, device, personalisation, and time of day, and a daily check captures one slice of a noisy distribution.

Source: Google Search Console Help. Confidence: Verified (first-party documentation).

The native brand filter (November 2025 / March 2026)

Google announced a native branded / non-branded query filter for Search Console Performance reports on November 20, 2025, and expanded it to all eligible sites on March 11, 2026. Filtering for non-brand queries — the discovery layer — became a one-click native operation rather than a regex exercise.

Source: Google Search Console product announcement, November 20, 2025; March 11, 2026 expansion. Confidence: Verified. Caveat: Per Search Engine Land, approximately 44% of Google searches are branded — meaning raw position averages overstate discovery health for any brand with notable traffic.

The &num=100 retirement (September 2025)

In September 2025, Google retired the &num=100 URL parameter that had allowed rank-trackers and crawlers to request 100 results per SERP page. This removed bot-inflated deep-rank impressions for positions 11-100 that no real user saw.

Search Engine Land (Danny Goodwin, September 18, 2025), citing LOCOMOTIVE's Tyler Gargula analysis of 319 properties: 87.7% of sites lost impressions in Google Search Console and unique ranking terms (query count) fell for 77.6% of sites.

Source: Search Engine Land (Danny Goodwin, Sept 18, 2025); LOCOMOTIVE / Tyler Gargula 319-property analysis. Confidence: Verified (event); Single-source on the headline percentages. Caveat: The 87.7% / 77.6% figures rest on a single 319-site study. The directional finding (industry-wide drop from a reporting change) is consistent across observer reports; the precise percentages are not an established benchmark. Year-over-year impression comparisons spanning September 2025 must be annotated.

This was a reporting cleanup, not a traffic loss — but it makes any naïve impression-trend comparison across the September 2025 boundary structurally misleading.

The impression logging bug (May 2025 – April 2026)

Google confirmed (and resolved) a logging error: "A logging error prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026. This issue has been resolved... Only impressions and related metrics — CTR and average position — were affected; clicks were not affected."

Source: Google Search Console official announcement. Confidence: Verified (first-party Google confirmation). Caveat: Impression, CTR, and position trends spanning May 2025 – April 2026 must be read with this annotation. Clicks remain the most trustworthy GSC metric across the discontinuity.

The practical implication is that clicks are the only GSC metric that can be trusted unannotated across the bug window. When reviewing a 2025 baseline, normalising against the bug window is a precondition of legitimate trend analysis.

Reporting glitches: Oct-Dec 2025 as a named instance

Confirmed GSC reporting glitches in the October-December 2025 window: the Performance report lagged 50-70+ hours (against a normal 2-6 hours), and the Page Indexing report froze for nearly a month, with data stuck around November 21 and the freeze not fixed until December 18, 2025.

Source: Google's public acknowledgments + community tracking, Oct-Dec 2025. Confidence: Verified (Google-acknowledged glitches). Caveat: This is the most recent named instance; assume similar episodes will recur.

Google confirmed these were reporting-only glitches that did not affect actual crawling, indexing, or ranking. But any site change made on the stale dashboards was a change made on bad information.

The cross-check rule

The composite implication of the four points above is operational: when GSC reports are visibly lagging or stuck, cross-check before acting. The three-instrument cross-check is server logs (were Googlebot fetches actually happening?), web analytics (was Google referral traffic landing?), and the URL Inspection live test (does Google currently see the URL as indexed — the live test bypasses the reporting cache). No drastic site change is made on stale reporting data.

Source: Synthesis of GSC documentation + October-December 2025 glitch window. Confidence: Verified.

If GSC and server logs disagree, server logs are usually closer to ground truth on the crawl side; URL Inspection live tests are closer to ground truth on the index side.

Success definitions: matching the framework to the site's situation

The same dashboard cannot evaluate both a brand-new domain and a replatformed established site. The two situations have inverted success definitions and demand different reporting postures. See also SEO J-curve and new-site ramp for the broader ramp dynamics.

New domain — growth from zero

For a brand-new domain, success is the chain starts moving from zero. The primary question is whether indexing → impressions → non-brand clicks → conversions are beginning to climb. The judgement is by direction and leading indicators, not absolute volume.

New domains face a widely-observed (though Google-unconfirmed as a deliberate hold) "sandbox" or trust-evaluation delay, commonly 3-6 months and longer in competitive niches. Rand Fishkin reported Moz experienced approximately 9 months of sandbox-like effects despite 12,000+ backlinks — note this is a single high-profile anecdote, not a base rate.

Source: Compass artifact research document; Rand Fishkin (Moz) anecdote. Confidence: Industry-consensus. Caveat: The Fishkin / Moz 9-month figure is a single anecdote and should not set expectations for a typical small-business launch. The honest framing is a trust-accrual window of "a couple of months to half a year or more," not a precise timeline.

Replatform — regression not detected

For a replatform or redesign replacing an existing site, success is regression not detected. The primary question is whether rankings, traffic, and conversions were preserved through the cutover, and whether cannibalisation or loss was detected fast.

The dominant risk is self-inflicted: missing 301 redirects, lost metadata or structured data, performance regression, accidental noindex directives, or robots.txt blocks. Botched migrations "lose 30-50% of their organic traffic almost overnight" (per numentechnology.co.uk — vendor incentive flag: migration-agency source). Clean migrations show only minor temporary volatility, typically stabilising within 14-30 days.

Source: Compass artifact research document; agency reporting (vendor incentive flagged). Confidence: Industry-consensus. Caveat: Applying the growth playbook to a migration hides traffic loss until revenue drops. The framing is the inverse of new-domain measurement: success is the absence of regression, not the presence of growth.

The two definitions impose different dashboards. A new-domain dashboard leads with directional leading indicators (impressions trending, query count expanding, first non-brand clicks appearing). A replatform dashboard leads with comparative deltas against the pre-cutover baseline (clicks within ±5%, top-50 ranking URLs preserved, no orphaned pages, no noindex regressions).

Realistic progress shape over twelve months

The measurement framework only earns its keep if the dashboard reads against a realistic time profile. The shape below is illustrative, not a benchmark or promise — and survivorship bias applies to every published progress curve.

Month 1 — Foundation, not results

Indexing completes; first impressions appear, mostly for long-tail and brand terms. Little to no meaningful clicks should be expected. The honest signal of progress here is technical: pages indexed, impressions beginning, no crawl errors.

Source: Compass artifact research document. Confidence: Industry-consensus. Caveat: Illustrative shape with wide variance. No indexing by week 4 indicates a technical crawlability problem.

Month 3 — First leading signals

Rising impressions, growing query count, first rankings for low-competition long-tail terms, early trickle of non-brand clicks. Google's Maile Ohye (former Developer Programs Tech Lead), in the "How to Hire an SEO" video: "In most cases, the SEO will need four months to a year to help your business first implement improvements and then see potential benefit."

Source: Maile Ohye (Google), "How to Hire an SEO" video; compass artifact research document. Confidence: Industry-consensus, citing Google. Caveat: Mid-funnel rungs moving at month 3 is the realistic shape for a foundation-and-content-sound site. Conversions (Rung 7) are not expected yet.

Month 6 — Possible inflection for low-competition terms

If the foundation and content are sound, this is where clicks and first conversions often begin to compound — but competitive terms typically remain out of reach. The industry framing of "3-6 months for early results" is consistent here.

Source: Compass artifact research document; industry consensus framing. Confidence: Industry-consensus. Caveat: "Possible inflection" is the right framing — not a guaranteed month-6 milestone. Many sites do not inflect until month 9+ or never.

Month 12 — Compounding, if it is working

Multiple pages ranking and supporting each other; non-brand clicks and conversions trending up. But the base rate is low: Ahrefs' May 2025 study (Patrick Stox) found only 1.74% of pages rank top-10 within a year, down from 5.7% in 2017.

Source: Ahrefs' May 2025 study (Patrick Stox); compass artifact research document. Confidence: Verified (the percentage figures); Industry-consensus (the "compounding" framing). Caveat: There is no single, reliable inflection point — where one appears, it is usually between months 3 and 6 for low-competition terms and much later or never for competitive ones.

AI Overviews and the zero-click reality

The 2025-2026 measurement landscape is reshaped by AI Overviews and AI Mode citations. Two consequences matter for the framework: prevalence must be cited as a range with an anchor, and the rank-to-click chain is materially weaker than the six-rung ladder implies.

Prevalence: anchor on Pew, cite vendors as range

AI Overview prevalence has no single authoritative number. Independent research (Pew, March 2025) and vendor trackers (16-48%, 2025-2026) diverge by definition and sample.

Pew Research Center (Athena Chapekis), analysing 900 US adults across 68,879 Google searches: "Some 18% of all the Google searches in our study generated an AI summary as part of the search results." The companion figure: 58% of US adults conducted at least one Google search in March 2025 that produced an AI-generated summary.

Source: https://pewresearch.org/short-reads/2025/07/22/ — Jul 22, 2025. Confidence: Verified (independent research org; no commercial incentive).

The vendor range across the same period:

• Semrush (10M+ keywords): 6.49% (Jan 2025) → peak ~24.61% (Jul 2025) → ~15.69% (Nov 2025).
• Conductor (21.9M queries): 25.11%, up from 13.14% in March 2025.
• BrightEdge (tracked queries): ~30% (early 2025) → 48% by Feb-Mar 2026 — a 58% year-over-year rise.

Source: Semrush blog 2025/26; Conductor.com 2026; BrightEdge "AI Overviews at the One-Year Mark." Confidence: Single-source per vendor (SEO platforms with incentive to show large numbers). Caveat: "Tracked queries" is a vendor-selected sample; vendor figures are not directly comparable to Pew's real-user search log. The direction (rising) is consistent across every source; the level requires an anchor + range treatment.

The reconciliation of the gap is largely definitional — different keyword samples, "tracked queries" versus all queries, inclusion of niche and long-tail terms. The reconciliation is itself the finding worth citing.

The discipline: cite Pew's independent 18% as the level anchor and the vendor trackers as a 16-48% range. Never cite a single vendor figure as "the" level.

Behavioural consequence: clicks halved when AI summaries appear

Pew's behavioural data measures the click consequence directly. When an AI summary appears, users click a traditional link 8% of the time versus 15% without — and only 1% click a link inside the summary. Session-end behaviour follows the same pattern: 26% of users ended the search session after seeing an AI summary, versus 16% without.

Source: Pew Research Center, https://pewresearch.org — Jul 22, 2025. Confidence: Verified.

The framework consequence is that rank is no longer a sufficient success metric. The decoupling is multi-source and large: BrightEdge data shows ~17% overlap between AI Overview citations and the organic top-10; Ahrefs data shows page-1 overlap fell from 76% to 38% in 6-7 months; ~80% of cited URLs are not in Google's top 100 across non-Google AI surfaces; Moz data puts 88% of AI Mode citations outside the top 10.

The operational reframe: client reporting leads with three parallel metrics — rank for high-intent commercial queries, AI-citation rate on tracked queries, and branded / direct traffic trend. See AI Overview citation patterns (GEO/AEO) for the citation-pattern details and How Google crawls, discovers, and indexes pages for the crawl and index plumbing that precedes any visibility at all.

Page-speed measurement and Core Web Vitals

Page-speed measurement is its own sub-framework, anchored on Google's Core Web Vitals (CWV) thresholds and a small set of canonical case studies.

The thresholds (and a circulating rumour)

Google's official "Good" thresholds remain LCP ≤ 2.5s, INP ≤ 200ms, CLS ≤ 0.1 (developers.google.com, last updated December 10, 2025). A widely-repeated SEO-blog claim that March 2026 tightened LCP to 2.0s and "elevated" INP is unconfirmed by Google and contradicted by current documentation — the claim traces to a single Digital Applied post citing a Search Central blog entry that does not exist. Search Engine Roundtable (April 8, 2026) confirmed Google's only public communication on the matter was a Search Status Dashboard note.

Source: developers.google.com (Dec 10, 2025); Search Engine Roundtable (April 8, 2026). Confidence: Verified (Google documentation). Caveat: Treat the March 2026 tightening as unconfirmed rumour.

Field data realities

CWV is Chrome-only field data. Safari only added LCP and INP in Safari 26.2 (December 2025). For iPhone-heavy demographics, CrUX may not represent the audience accurately. Low-traffic small-business sites often show no URL-level CrUX data — Google falls back to origin-level — so the moat may be invisible at very small scale.

Per Web Almanac 2025: 48% of mobile origins and 56% of desktop origins pass all three CWV thresholds on July 2025 CrUX data. LCP is the bottleneck (62% of mobile origins good versus 77% INP and 81% CLS). The platform spread is wide: Duda 83.63%, Shopify 75.22%, Squarespace 67.66%, WordPress 43.44%. Elementor sub-segments perform worse (25-35%).

Source: Web Almanac 2025 (HTTP Archive). Confidence: Verified.

Per Dan Taylor (Search Engine Land, January 13, 2026, n=107,352): CWV is a gate, not a growth lever. Severe failure suppresses AI Overview citation; "Good" to "great" does not lift it. Correlations: LCP r = −0.12 to −0.18, CLS r = −0.05 to −0.09.

Business-impact evidence

The cleanest causal evidence linking page speed to revenue is Vodafone's 2021 A/B test: a 31% LCP improvement produced 8% more sales, 15% better lead-rate, and 11% better cart-rate. Most "performance affects revenue" claims are correlational; Vodafone is a controlled A/B test of a single LCP intervention.

Source: https://web.dev/case-studies/vodafone (2021, still cited in 2025+). Confidence: Verified.

The framework consequence is that page-speed measurement deserves its own panel — separate from GSC visibility and GA4 outcomes — with the CWV pass/fail gate, field-data LCP/INP/CLS distributions, and a documented baseline against which optimisation work is judged.

Trust signals as a measurement target

Trust-signal coverage is the often-overlooked third measurement axis. The macro condition is hostile: the Edelman 2025 Trust Barometer (33,000 respondents, 28 countries) found 7 in 10 respondents believe government officials, business leaders, and journalists deliberately mislead them; 6 in 10 hold "moderate to high" grievance against government and business.

Source: edelman.com/news-awards/2025-edelman-trust-barometer-reveals-high-level-grievance (January 19, 2025). Confidence: Verified.

Unsourced claims are read as adversarial by default in 2026. Every claim without a name, date, or source is processed by a reader who already assumes leaders mislead them. Sourcing is not extra credit; it is the floor.

Trust-signal interventions are measurable. The Trust Project / Reach Plc (UK) reported: "Across two surveys, Reach Plc (UK) found that trust in its flagship outlet, The Mirror, jumped eight percent after it added the Trust Indicators to its site."

Source: Trust Project — thetrustproject.org; SCU Markkula Center — scu.edu/ethics. Confidence: Verified.

The framework consequence is that trust-signal coverage is a measurable axis: visible sourcing, dated claims, author identity, and credentials are interventions that can be deployed, tracked, and evaluated. A trust-coverage rate — the percentage of public claims on a site that carry an explicit, datable source — is a defensible metric to add to a small-business dashboard.

A related demand-side anchor: Gartner's CSO Update (2019) found 77% of B2B buyers say their latest purchase was "very complex or difficult." The site that helps buyers reduce complexity — clear vertical entry points, decision aids, trust signals at the point of friction — earns the relationship. Trust signals are not decorative; they sit at the conversion path.

Source: Gartner — gartner.com/en/sales/insights/b2b-buying-journey. Confidence: Verified.

Operating rules

The framework distils to seven operating rules that govern day-to-day measurement work.

1. Keep GSC and GA4 strictly separate. Visibility and business outcomes are different questions answered by different tools. Never put impressions and engagement rate on the same trend chart. When a client asks "is the site working?" the answer comes in two halves: visibility (GSC) and business outcomes (GA4 plus CRM).

2. Read the six rungs as a ladder, not a leaderboard. Each rung is a precondition for the next. A site that ranks #2 with no clicks has a Rung-5 problem (snippet, intent match, or zero-click suppression), not a Rung-4 win.

3. Cross-check GSC during known lag windows. Server logs, analytics, and URL Inspection live tests are the three-instrument check. Do not make architectural decisions on stale reporting data. The Oct-Dec 2025 glitch window is the most recent named instance; assume similar episodes recur.

4. Match the success definition to the situation. New domains are judged on growth-from-zero direction. Replatforms are judged on regression-not-detected against a pre-cutover baseline. Applying the growth playbook to a migration hides traffic loss until revenue drops.

5. Decouple from Google rank as the sole success metric. Add AI citation rate and branded / direct traffic as parallel metrics. The 8% versus 15% click rate with and without an AI summary (Pew) is the empirical anchor for the reframe.

6. Cite Pew as the AI-prevalence anchor; cite vendor trackers as the range. The default citation pattern: "Pew (independent, March 2025): 18% of Google searches produced an AI summary; vendor trackers across 2025-2026 put the range at 16-48% depending on sample."

7. Treat engagement as diagnostic, not as a success KPI. GA4 engagement-rate is useful for diagnosing traffic-content fit. The causal link from on-site engagement to conversions and revenue is unproven in the general case.

Sister entries

• Decision-linked metrics versus vanity metrics — which metrics to choose and which to discard; this entry is the what to measure, the present entry is the how to measure.
• How Google crawls, discovers, and indexes pages — the crawl, render, and index plumbing that precedes Rung 1; required reading for any framework that begins with "is the site indexed?"
• SEO J-curve and new-site ramp — the longer-form ramp dynamics behind the realistic-progress-shape section; the trust-accrual window and the J-curve mechanics belong together.
• AI Overview citation patterns (GEO/AEO) — the citation-pattern detail behind the rank-decoupling rule; the AI Overviews section here is the measurement framing, that entry is the citation mechanics.