Measurement
Part of How to decide what technical SEO metrics to measure and report
Before you pick technical SEO attribution methods, settle the rules
Compare technical SEO attribution methods for UK teams: last click, modelled and experiment-led, with the evidence each needs and a worked pound example.
What to take away
- Attribution is a claim about cause, and most teams pick a method before agreeing what a fix may claim, then argue about the numbers.
- Last click suits single-page fixes such as redirects. It overstates the work when other channels moved in the same period.
- Modelled and experiment-led methods need a control group, a baseline and a stated confidence level.
- Record the rule, the data source and the review date on the same page as the claim.
Why the method choice goes wrong
What the claim must specify first
A redirect fix, a schema change and a speed improvement each need different evidence. Teams that start from a tool dashboard end up defending a number they cannot explain.
Three attribution methods compared
Last click
- What it credits
- Final interaction
- Evidence needed
- Click data
- Best for
- Single-page fixes
- Causal strength
- Weak
- Cost to run
- Cheap
Modelled
- What it credits
- Spread by rule/model
- Evidence needed
- High volume, written model
- Best for
- Stable high-volume claims
- Causal strength
- Medium
- Cost to run
- Moderate
Experiment-led
- What it credits
- Changed vs unchanged group
- Evidence needed
- Holdout group
- Best for
- Large or contested claims
- Causal strength
- Strongest
- Cost to run
- Most expensive
For example: "Moving 240 legacy URLS to 301 redirects recovered £2,400 a month in organic revenue." That sets the metric, the window and the baseline.
Three methods and what each one needs
Last click credits the final interaction. It is cheap and familiar. It assumes GA4 conversion events, or another click-level source, plus conversion tracking on the final action.
Modelled attribution spreads credit using a rule or a statistical model. It needs enough conversion volume to stay stable, plus a written model description a reviewer can challenge. It assumes GA4 modelled conversions or a documented marketing mix model.
Experiment-led attribution compares a changed group with an unchanged one. It assumes a holdout of comparable URLS, server logs and Search Console data.
Google's guidance on redirect types and their search impact is useful here, because it sets out what a redirect change does and does not affect.
How to compare the methods on evidence
Match the metric to the fix
Performance work should be reported against Core Web Vitals thresholds and metric definitions, not a blended traffic figure. If the fix targets loading speed, the claim should name the metric that moved.
Our measurement and reporting guide sets out the baseline and review cadence each method assumes, which is worth reading before you commit to a model.
For template-wide changes, use a holdout. For single-URL changes, last click is usually enough. Where a fix touches consent, tracking or AI-assisted tooling, check the ICO's guidance on artificial intelligence before relying on inferred data.
The table maps the fix type to the method, the evidence you need and the confidence you can claim.
Method
- Single URL
- Last click
- Template
- Modelled or holdout
- Consent or tracking
- Experiment-led
Evidence needed
- Single URL
- GA4 conversion event, redirect date, no overlapping campaign
- Template
- GA4 model description, baseline period, control pages, server logs
- Consent or tracking
- Lawful basis, consent records, holdout, Search Console checks
Confidence level
- Single URL
- Medium
- Template
- Medium to high
- Consent or tracking
- High if the control holds
A worked example with labelled figures
An English retailer redirects 240 legacy URLS. Organic revenue for the affected section was £18,000 a month before the change and £20,400 after, a raw difference of £2,400 a month.
Last click attributes all £2,400 to the redirect work. A modelled view, after removing a seasonal uplift of £600 a month, attributes £1,800. An experiment-led view using a holdout of 60 comparable URLS attributes £1,400.
The £2,400, £1,800 and £1,400 figures are illustrative, not UK benchmarks. The method, not the traffic, decides the number you publish, so report the range and the method rather than one headline figure.
Our benchmark research on English sites sets out its own sample and method, so check it before you cite a UK figure.
What to write down before you report
The attribution record
Keep one page per claim. Record the fix, the URLS affected, the metric and the periods compared, then the method, the data source, the reviewer and the re-check date.
Choosing between methods
Use last click for narrow, single-page fixes. Use modelled attribution when conversion volume is high enough and the model is documented. Use an experiment when the claim is large, public or contested.
Never mix methods inside one comparison table without labelling each row.
Common questions
Which attribution method is most accurate for technical SEO?
No method is accurate in absolute terms. Experiment-led attribution gives the strongest causal evidence, but it needs a control group and enough traffic.
Can I use last click for redirect fixes?
Yes, for a small set of URLS where the redirect is the only change and no campaign ran in the same period. Document the baseline and the review date.
How long should I wait before reporting a result?
Long enough to cover one full reporting cycle after the change, and for the affected URLS to be recrawled.
Do AI tools change what I can attribute?
They can change how data is collected and inferred, so the compliance position matters as much as the model. Check the lawful basis for any inferred profiling before using it in a published claim.
