Track performance and adjust

Read a programmatic site by group of pages. A single city page getting four clicks tells you nothing, while the same number across 400 pages of one template tells you whether the template works. So the useful setup is one table that joins each URL to its template, its rollout wave and its data, and a short list of metrics that each trigger a decision.

I’ll stay with the same example: a coworking directory with 2,000 city pages, released in waves as described in the chapter on production and indexing.

Follow the funnel from indexed page to business result

Each metric below exists because a bad number points at a specific fix.

Stage	Metric	Source	A weak number points to
Indexing	Share of submitted pages indexed, per wave	Search Console, Page indexing	Template quality, discovery
Visibility	Impressions, queries per page	Search Console	Wrong targeting, too little authority
Ranking	Average position by query group	Search Console	Content depth, internal links
Attractiveness	CTR at comparable positions	Search Console	Title and meta templates
Engagement	Engaged sessions, average engagement time	GA4	Search intent, UX
Business	Key events, value per visitor, retention	GA4, CRM, product data	Offer, calls to action, lead quality

A few of these need careful reading. Average position is weighted by impressions and blends every query a page appears for, so compare it within a group of similar queries. GA4 has replaced the old time on page with engagement time, and its bounce rate is now simply the opposite of engagement rate. What it calls conversions are now key events.

Clicks and impressions from Google’s AI features are counted inside Search Console’s web totals. So a falling CTR on stable positions can mean an AI answer now satisfies the query, with nothing wrong on your page. So I read CTR by query type. My working assumption is that factual queries are the most exposed, and that queries needing a list, a comparison or a price from your data keep a better reason to click.

The bottom of the table is the part people skip, and it is the only part your business feels. The results I care about are business numbers: 107 signups in a typical week on Derrick, and a steady 35 to 50 leads per month on Deux.io. To measure that, pass the landing page through to your CRM or product database at signup. Otherwise retention and lifetime value stay invisible at page level.

Know what each tool answers

Tool	Question it answers
Google Search Console	Is the page indexed, for which queries does it appear, and where
Google Analytics 4	What visitors do once they arrive
Looker Studio	What the whole picture looks like, shared with the team
Screaming Frog	What is broken on the site: 404s, orphans, tags, internal links
Ahrefs, Semrush	How backlinks and competitors’ rankings are moving
Hotjar, Microsoft Clarity	Where people get stuck on a template

None of this is real time, and for a programmatic site that doesn’t matter. Search Console data arrives with a delay, and search results move slowly enough that a weekly read is plenty.

The Search Console limit that does matter at this scale is the 1,000 rows in the interface. Use the API, or better, the bulk data export to BigQuery, which is built for sites with many pages.

Referrals from AI assistants deserve their own channel in GA4. A custom channel group using this condition on session source does the job, and it will undercount, because part of that traffic arrives with no referrer at all:

^(chatgpt\.com|perplexity\.ai|copilot\.microsoft\.com|gemini\.google\.com|claude\.ai)$

Join search data to your own data

Export your page table to BigQuery with the full URL, then query it against the Search Console export. Replace project with your own project ID, and adjust dataset names to your setup.

SELECT
  p.template,
  p.wave,
  SUM(g.impressions) AS impressions,
  SUM(g.clicks) AS clicks,
  SAFE_DIVIDE(SUM(g.clicks), SUM(g.impressions)) AS ctr,
  SAFE_DIVIDE(SUM(g.sum_position), SUM(g.impressions)) + 1 AS avg_position
FROM `project.searchconsole.searchdata_url_impression` AS g
JOIN `project.pseo.page` AS p
  ON g.url = p.url
WHERE g.data_date >= DATE_SUB(CURRENT_DATE(), INTERVAL 28 DAY)
  AND g.search_type = 'WEB'
GROUP BY p.template, p.wave
ORDER BY p.template, p.wave;

Add any field from your database to that grouping and you start answering questions no SEO tool can. Do cities with ten or more listings earn a better CTR than cities with three? Do pages with a price summary rank higher? If they do, you know what to invest in, and you have evidence for raising your publication threshold. Plug the result into Looker Studio and you have your dashboard.

Turn symptoms into changes

What you see	Likely cause	What to try
Good position, low CTR	Generic title template	Bring real data into the title: count, price, freshness
Stuck on page two	Not enough depth or internal authority	Add data blocks, add links from stronger pages
A wave stays unindexed	Pages too similar or too thin	Enrich them, or raise the threshold and prune
Impressions on queries you never targeted	An unmet need	Add a field, a section or a new template
Clicks, no engagement	The page doesn’t match the intent	Check the template in Clarity, rework the top of the page
Engagement, no conversions	Weak or misplaced call to action	Test the offer on that template

Fast-growing queries and long-tail variations surface in the same export. They are your cheapest new pages, because the template already exists.

For the second row, add links from pages that have authority to pages close to breaking through, using the rules from the chapter on internal linking. Regroup pages into coherent clusters when the query data shows that people think about them together.

New formats are worth testing too: comparisons, interactive tools, FAQ blocks. Add FAQs because they answer real questions and because AI answers can reuse them. The FAQ rich result is restricted to authoritative government and health sites, so it won’t appear for a directory.

Updates should follow real changes in the data. Moving a date on a page that hasn’t changed isn’t freshness, and it damages the trust Google gives your lastmod values.

And if a template keeps producing pages that earn neither clicks nor engagement after several rounds of work, remove those pages. They are the profile the scaled content abuse policy describes, and they drag down your read of everything else.

Test on a wave before changing everything

Editing a template changes thousands of pages at once, which makes it impossible to know what caused what. Apply the change to one wave or half of a template, keep the rest as your control group, record the date, and compare both groups over the same 28-day windows so that seasonality affects them equally. Then roll out what worked.

Keep the rhythm simple:

• every week, scan for anomalies in indexing, clicks and server errors
• every month, review results by template and wave and pick one change to test
• every quarter, prune weak pages, review competitors and refresh your data

Most of this can run without you once it is set up. The chapter on automating SEO audits covers how to catch regressions across thousands of pages.

Your next step: connect the Search Console bulk export, load your page table next to it, and run the query above. Keep the first dashboard to one chart of clicks per wave, then choose the one template change you will test this month.