Analytics & Forecasting

The skill that separates senior SEO specialists from competent practitioners is not another tactic. It is translating search data into business decisions and credible predictions. This guide covers the measurement model, the 2025-2026 changes to the data itself, and forecasting without lying. How to turn all of this into a roadmap and a stakeholder report is covered in SEO Strategy & Stakeholder Reporting.

The KPI hierarchy#

Structure metrics in three tiers; report upward, diagnose downward:

The classic failure is reporting tier-3 metrics to tier-1 audiences. "Average position improved 0.4" means nothing to a CFO; "organic signups up 23% quarter on quarter, now 4× cheaper than paid" funds your roadmap.

Working with Search Console data honestly#

Search Console is the source of truth for the search side, with known sharp edges:

• Privacy filtering. Query-level data omits anonymized queries, so the totals exceed the sum of visible queries. The UI shows at most 1,000 rows, so use the API or bulk export to BigQuery for serious analysis.
• AI features are blended in. Since June 2025, AI Mode clicks and impressions count in the Web search type, and an AI Overview occupies one position shared by all its links. A page cited in an AI Overview can show position 1 with almost no clicks.
• Positions are averages across impressions. "Position 8.2" may be #1 for one query and #40 for another. Diagnose at the query level.
• Use the right time grain. Weekly and monthly views (December 2025) smooth out day-of-week noise for trend charts.

Branded vs non-branded#

Brand traffic follows brand strength, not your SEO work. Mixing them flatters or hides everything. Search Console now has a branded queries filter (Performance report: open the filter bar, choose Query, then Branded or Non-branded), available to all eligible sites since 2026-03-11. It is AI-assisted rather than regex, so it catches misspellings and brand products, and an Insights card shows the branded share of clicks. It is not available for sub-properties (such as https://example.com/blog/) or for sites with few impressions. Keep a regex filter as a fallback for smaller sites:

rankyl|rankil|rank yl|ranky l

Annotate every break#

An un-annotated traffic chart is a Rorschach test. Search Console supports custom annotations (since 2025-11-17): right-click the Performance chart, add up to 120 characters, visible to everyone on the property. Annotate:

• confirmed Google updates (from the Search Status Dashboard)
• migrations, redesigns and major releases
• reporting changes, above all the one below

Measuring AI visibility#

Three first-party sources now exist, and each covers a different slice:

The GA4 channel is not retroactive, so earlier AI visits stay in Referral. If you see AI referrers that are not on Google's list in your Referral report, keep a custom channel for them. Measuring AI Visibility covers prompt tracking and log analysis for the gaps.

Attribution reality#

SEO's contribution is systematically under-credited: organic often opens journeys that close via direct or paid brand clicks, and zero-click and AI surfaces deliver value without sessions at all. Defenses:

• Use data-driven attribution (GA4's default) rather than last click when arguing budget.
• Track branded search demand (the branded filter's clicks and impressions) as a first-class outcome of content and PR work.
• Count AI impressions and citations where clicks cannot happen.

Forecasting SEO impact#

Forecasts get budgets approved. The honest method is scenario-based and bottom-up:

For each target cluster (from the keyword map):
  volume        monthly searches (tool estimate, haircut by 30%)
  position now  current avg position (GSC), else 0
  position then scenario: conservative / expected / optimistic
  CTR(pos)      apply a CTR curve built from your own GSC data,
                split by SERPs with and without AI Overviews
  Δ traffic     = volume × [CTR(then) − CTR(now)]
  Δ value       = Δ traffic × conversion rate × value per conversion
 
Sum across clusters → ramp over 6–12 months (SEO compounds,
it doesn't step) → present all three scenarios with assumptions.

Rules that keep forecasts credible:

• Present ranges, never points. "Expected case: +8k to 12k sessions a month by Q3, assuming we ship 12 of the planned guides" is defensible; "+47.3% traffic" is astrology.
• State assumptions as the deliverable. The forecast is the model; the conversation you want is about the assumptions.
• Discount for AI answers. Click-through for top results has fallen where AI Overviews appear. Ahrefs measured 34.5% lower position-1 CTR in April 2025 and reported 58% lower in its February 2026 rerun. Studies differ in method, so treat these as directional and use your own data where possible.
• Never train a model across a data break. Start impression-based baselines after September 2025, or model clicks only.
• Calibrate quarterly: forecast vs actual, in public. Being seen correcting your own model is what makes the next forecast trusted.

The reporting cadence that works#

• Weekly (internal): anomaly watch: indexation, CWV, manual actions, big query movers. No meeting; an alert channel.
• Monthly (team): performance vs plan by section and cluster, branded vs non-branded, AI impressions and referrals, shipped work, next month's priorities.
• Quarterly (stakeholders): business-tier results, forecast calibration, roadmap and resourcing ask.

What to put in each, and what to leave out, is the subject of SEO Strategy & Stakeholder Reporting.

Put it into practice#

The foundation is the stable part: be crawlable, be fast, be structured, be genuinely the best answer, and be measurable. Everything new is a variation on those themes. To practise all of it on a real site, work through the Capstone: Audit Your Own Site, and start your prompt portfolio now: AI visibility baselines get more valuable the earlier they start.