LEARN

Measurement Discipline: Four Rules

Four upright partitions of differing thickness, each with an opening at a different height, separate streams meeting at one fixed base, representing data sources that each declare their own basis before meeting one acceptance standard.
Measuring discipline involves four rules: fixed denominator, not adding different sources, separate reporting by platform/language, and marking estimates as such

IN ONE SENTENCE

Four measurement rules: fix the denominator, never sum different sources, report per platform and language, and label estimates as estimates.

Data in this field easily looks like it is moving while saying nothing. Four rules prevent most failures: fix the denominator, never sum different sources, report per platform and language, and label estimates as estimates.

OUR POSITION

Credibility of a metric matters more than how good it looks. A report that can explain where a change came from is useful even when the numbers are unflattering; a flattering report whose definitions do not reconcile dies the first time someone asks.

01

The four rules

Fix the denominator: question set, platform mix and sampling runs settled before the first run. Changing them mid-flight means it is no longer the same measurement — a bigger problem than sampling noise.

Never sum different sources: first-party analytics, server logs, webmaster tools and web analytics measure different things. Explain a gap rather than closing it.

Report per platform and per language: blending hides a decline on one side. A rising blended mention rate may be one platform moving.

Label estimates: where sources are not displayed, citation share can only be estimated, and referral-less traffic can only be estimated. Labelled, they remain useful; unlabelled, they mislead.

02

The two most common failures

Comparing cumulative with per-period figures. Stretch the window and almost every brand approaches 100%, destroying the signal. Report daily-average or monthly and label the column.

Treating third-party estimates as an acceptance baseline. Third-party traffic and difficulty are estimates — fine for order-of-magnitude and competitive comparison, not for acceptance.

03

Why position bands are the thing to watch

Yield concentrates in the top ten: roughly 13 monthly clicks per keyword at 1–3, 3.4 at 4–10, 0.1 at 11–20, near zero beyond 21.

So review movement between bands rather than decimals in an average. An average position improving from 24 to 19 makes no business difference.

Data behind this page

1.08%

AI referrals as a share of total site traffic

Source：Conductor, across 13,770 domains，2025

13.1 / 3.4 / 0.1 / ≈0

Monthly clicks per keyword at positions 1–3 / 4–10 / 11–20 / 21+

Source：Our keyword-level analysis of 4,074 non-branded keywords for a leading site in this category，2026-06

Sources

  1. [1]Google users are less likely to click on links when an AI summary appears．Pew Research Center．2025-07-22
  2. [2]Optimizing your website for generative AI features on Google Search．Google Search Central．2026-05-15

Updated 2026-08-10