How do you measure SEO?

“Did SEO improve?” is too broad until the goal is defined. A publisher may value qualified readers and subscriptions. A local clinic may value suitable appointment enquiries. A store may value revenue, repeat purchase and profit. Measurement should connect search visibility to the result the page exists to support.

Write the measurement question first: Did non-brand discovery for repair services reach more suitable searchers? Did those visitors view service information and request help? This prevents a dashboard from turning every rising number into success.

The Performance report can show clicks, impressions, average click-through rate and average position for eligible Google Search data. It can be filtered by query, page, country, device, date and search appearance, subject to privacy and reporting limits.

If 1,000 counted impressions create 60 clicks, CTR is 6%. A lower CTR is not automatically bad. The page may be appearing for broader early-stage searches, a result feature may answer part of the query, or the average position may have changed.

Compare like with like: brand versus non-brand, mobile versus desktop, country, page type and query intent. A product page and a dictionary definition should not be expected to have the same click behavior.

Search Console averages the topmost position of the site for counted impressions. Results vary by query, location, device and result layout. An average of 7.4 does not mean every person saw the page at exactly position seven.

Use position as diagnostic context with impressions, clicks and query relevance. Do not report one manually observed search as a guaranteed market-wide ranking.

A website analytics tool such as Google Analytics can record sessions, page views, events and configured key events after visitors reach the site, depending on implementation, consent, blockers and identity rules. It does not reproduce Search Console’s query dataset.

Search Console clicks and analytics sessions will not match exactly. They use different definitions, time processing, consent behavior and counting methods. The difference is expected; investigate large unexplained gaps without forcing equality.

A conversion may be a completed purchase, qualified enquiry, booking, trial registration or another action that matters. A button click alone may be an intermediate event, not proof that a lead was delivered or useful.

Test forms and purchase tracking end to end. Prevent duplicate events, exclude internal tests where appropriate and avoid sending personal information into analytics. CRM or order systems can help evaluate lead quality and revenue after the website event.

A customer may discover a guide in search, return through social media, click an email and later contact the company directly. Attribution models assign credit using available recorded touchpoints. They do not reveal perfect causation or every offline influence.

Report assumptions honestly. Search assisted the journey according to the available data; it may not deserve all credit. Use controlled experiments where possible, longer-term trends and qualitative evidence to improve confidence.

Google states that traffic from AI Overviews and AI Mode is included in Search Console’s overall Web search reporting. Search Console does not currently provide a separate universal “AI citation ranking” report. Avoid third-party dashboards that present estimated AI visibility as exact fact.

Track pages, conversions, referral data where available, brand demand and observed citations carefully. Record the method and date because AI experiences change quickly.

• Check technical/indexing issues that affect important pages.
• Separate brand and non-brand search patterns where practical.
• Review relevant queries and pages, not only totals.
• Compare periods with seasonality and site changes in mind.
• Check whether important organic visits create useful actions.
• Verify tracking before explaining a sudden change.
• Record changes made and what evidence would support their effect.