Chapter 7: Analytics and Tooling

Weekly and monthly review SOP, timed

1. Top line, five minutes. Non-branded organic clicks, AI referral traffic, organic conversions. Compare week over week and year over year — the year-on-year comparison is what removes seasonality from your first impression.
2. Locate the swing, fifteen minutes. In Search Console, compare the last 28 days against the previous 28, sort by click difference ascending, and take the ten pages and twenty queries that lost the most. Do the same for gains; losing sight of why something went up is how wins get destroyed by accident.
3. Diagnose the shape of the change. The pattern tells you where to look:
   • Impressions and position both fell → algorithm update, an accidental noindex, broken internal links, or a competitor genuinely outranking you now.
   • Position and impressions held, clicks and CTR collapsed → check the SERP in a private window. New AI answers, more ads, or a competitor taking the rich result will push you below the fold without touching your rank.
   • Everything flat except a specific day or two → seasonality or a local holiday; confirm with Trends before you open a ticket.
   • Search Console fine but analytics sessions or conversions at zero → the tracking broke. Check the tag, the consent banner and the checkout.
4. Write the action, not the screenshot. Every chart in the report carries two lines: what it means, and which pages change next week because of it. A report without a decision attached is a newsletter.

Interview questions and answer frameworks

Q1: Your boss asks why organic traffic dropped 30% last week. You have one hour. What do you do?

Framework: Five steps, and lead with the ones that rule things out fastest. First, is the data real? Compare Search Console clicks against analytics sessions. If clicks are flat and sessions fell, the tracking broke — a tag lost in a deploy, a consent banner blocking scripts. Compare year over year to rule out a seasonal dip. Second, segment before theorising. Split the drop by directory, country, device and brand versus non-brand. A single template or a single market failing is a different problem from a site-wide decline. Third, check for self-inflicted damage. Manual actions, security issues, a spike in 5xx, and — most common of all — ask what shipped last week: a robots change, a canonical change, a rendering change. Fourth, check the outside. Algorithm update trackers, then look at the SERP for the queries that lost the most: new AI answer, new ads, a competitor rewrite. Fifth, report in three parts: what happened, how much of the business it touches, and what is already being fixed with an expected timeline. Never answer with a chart and no conclusion.

Q2: Why do Search Console clicks and analytics organic sessions never match?

Framework: They measure different things at different points. Search Console counts the click on the result page; analytics counts a pageview that fired a script. Clicks that bounce before the page loads, redirects, and blocked or unconsented scripts are counted in one and not the other. Deduplication differs too — several clicks from one person within the session window become one session. And privacy tooling blocks analytics far more often than it blocks the search engine’s own logging. The practical answer: expect clicks to exceed sessions by a noticeable margin, treat the gap as a health indicator rather than an error, and investigate when the gap suddenly changes — a widening gap is often the first sign that a script stopped loading.

Q3: How do you measure GEO, where the click may never happen?

Framework: Accept that the unit of measurement changes from sessions to mentions. Track four things: share of answer — for a fixed panel of prompts in your category, how often you are cited, measured on a schedule with the same prompts; citation quality — whether you are the primary source or a footnote; sentiment and accuracy — whether the model describes you correctly, because being cited wrongly is worse than being absent; downstream — branded search volume and direct traffic, which is where answer-engine visibility shows up when it works. Then be honest about limits: prompt panels drift, models change between sessions, and none of this attributes cleanly to revenue. Report it as a trend with a stated method, not as a precise number.

Q4: What is the most common mistake you see in SEO reporting?

Framework: Reporting activity instead of outcome, and outcome without a control. “We published twelve posts and fixed 240 issues” is activity. “Non-branded clicks on the commercial cluster rose while the untouched control cluster stayed flat” is evidence. The second version requires deciding in advance what you expect to move and what you will compare against, which is why most teams skip it and then argue about whether their work caused the chart to go up. Say that you build the comparison group before the work starts, not after the results arrive.