Measuring Whether Link Building Worked

Agencies report links delivered, which is an activity count, and clients want revenue, which link building alone cannot be attributed to. Both sides then argue using numbers that answer different questions and the decision gets made on confidence rather than evidence. The only honest test is whether the pages that received links moved relative to comparable pages that did not, and that comparison has to be designed before the campaign starts, because it cannot be reconstructed afterwards.

The skill file

---
name: link-acquisition-measurement
description: Use when a link building programme is up for renewal or cancellation and nobody can say what it actually produced.
---

# Measuring Whether Link Building Worked

Agencies report links delivered, which is an activity count, and clients want revenue, which link building alone cannot be attributed to. Both sides then argue using numbers that answer different questions and the decision gets made on confidence rather than evidence. The only honest test is whether the pages that received links moved relative to comparable pages that did not, and that comparison has to be designed before the campaign starts, because it cannot be reconstructed afterwards.

## What you need first

- the delivered link list with date, target URL and acquisition method for each
- a control set of comparable pages agreed at the start that will receive no new links
- ranking and organic click history per target URL from before the campaign began
- the change log for the same period so content and technical work can be separated out

## Method

1. Fix the control set before the campaign runs: pages of similar type, age and starting position that deliberately get no new links. Without it, every movement in the period is claimed by whoever presents first.
2. Recrawl every delivered link and count only those that are live, followed, indexed and on a page with traffic of its own. Links delivered and links surviving are routinely different numbers by a wide margin.
3. Measure at target URL level, not site level. Link programmes point at a handful of pages, so a site-wide traffic chart is dominated by everything else that happened that quarter.
4. Allow the lag your own site has previously shown between links landing and rankings moving, taken from earlier batches. Reviewing at 30 days reliably cancels programmes that were on track.
5. Compare targets against the control over identical dates, so an algorithm update or a seasonal swing lands on both sets and cancels out instead of being read as effect.
6. Subtract anything the content or technical teams shipped to the same URLs in the window. A page rewritten and linked in the same month proves nothing about either piece of work.
7. Report a cost per surviving editorial link alongside the target versus control movement and the confidence you genuinely have, then make one recommendation: renew, change method, or stop.

## What this produces

A one page campaign read with surviving link count, cost per surviving link, target versus control movement over the lag-adjusted window, and a stated recommendation.

## Where this goes wrong

- reporting links delivered as the outcome, when the number that matters is links still live and followed six months later
- running with no control set, which leaves every claim unfalsifiable in both directions
- judging the programme inside the lag window your own site has already demonstrated
- comparing target pages against a previous period rather than a control, so an update inside the window is read as campaign effect

---

From the QuQi skill library - https://www.quqi.io/skills/link-acquisition-measurement

What you need first

• the delivered link list with date, target URL and acquisition method for each
• a control set of comparable pages agreed at the start that will receive no new links
• ranking and organic click history per target URL from before the campaign began
• the change log for the same period so content and technical work can be separated out

Method

1. 01 Fix the control set before the campaign runs: pages of similar type, age and starting position that deliberately get no new links. Without it, every movement in the period is claimed by whoever presents first.
2. 02 Recrawl every delivered link and count only those that are live, followed, indexed and on a page with traffic of its own. Links delivered and links surviving are routinely different numbers by a wide margin.
3. 03 Measure at target URL level, not site level. Link programmes point at a handful of pages, so a site-wide traffic chart is dominated by everything else that happened that quarter.
4. 04 Allow the lag your own site has previously shown between links landing and rankings moving, taken from earlier batches. Reviewing at 30 days reliably cancels programmes that were on track.
5. 05 Compare targets against the control over identical dates, so an algorithm update or a seasonal swing lands on both sets and cancels out instead of being read as effect.
6. 06 Subtract anything the content or technical teams shipped to the same URLs in the window. A page rewritten and linked in the same month proves nothing about either piece of work.
7. 07 Report a cost per surviving editorial link alongside the target versus control movement and the confidence you genuinely have, then make one recommendation: renew, change method, or stop.

What this produces

A one page campaign read with surviving link count, cost per surviving link, target versus control movement over the lag-adjusted window, and a stated recommendation.

Where this goes wrong

• reporting links delivered as the outcome, when the number that matters is links still live and followed six months later
• running with no control set, which leaves every claim unfalsifiable in both directions
• judging the programme inside the lag window your own site has already demonstrated
• comparing target pages against a previous period rather than a control, so an update inside the window is read as campaign effect

Use this skill in your own AI

The download is a plain markdown file with the name and trigger in its frontmatter. Where an assistant supports skills it can load itself, that frontmatter is what it reads to decide this one applies.

Questions about this skill

More in Links & authority