This pillar publishes real test data from three live sites we operate (Clarivian.io, wealth-wire.com, jbaitoolsinsider.com). Server logs, GSC impressions, schema experiments, crawler patterns, and citation outcomes. The premise: GEO advice without data is opinion; GEO advice with data is methodology. We anonymize what needs anonymizing and publish what doesn't.
GEO content has a credibility gap. Most published advice is either vendor copy (selling a tool), recycled SEO-blog wisdom (untested for the AI context), or theoretical extrapolation (sounds right, untested). The minority that's grounded in actual data tends to come from large agencies that won't share their workings.
This pillar fills the gap with case studies from our own portfolio. Three live sites, all serving real users, all instrumented with the same measurement stack. We publish what happened, what we changed, and what we learned. No vendor relationships, no NDAs, no theoretical model that hasn't been tested.
Our methodology
Every case study follows the same structure:
- Starting state. Where the site or page was before the change. Impressions, clicks, position, crawler activity, schema coverage.
- The intervention. What specifically changed (code, schema, robots.txt, content structure, internal linking). Reproducible enough that someone else can apply it.
- The measurement window. How long we waited before evaluating. Typically 14-28 days for crawler signal, 60-90 days for ranking and citation signal.
- The result. Honest numbers, including the cases where the result was nothing or worse than before.
- The takeaway. What we'd recommend doing differently next time.
Case study 1: Quick Answer boxes on 5 articles (Clarivian, 2026-05)
Starting state: Five articles republished after a content-management mistake had unpublished them, returning 404 to Googlebot for several weeks. All five had real content (1,400-2,250 words) but no Quick Answer box at the top. Average position before: 0 (not indexed). Impressions before: 0 (Google had deindexed them).
Intervention: Republished with Quick Answer boxes added to the top of each article (~80-word factual summary, marked with a green left-border and a 'Quick answer' eyebrow). Em-dashes stripped. Articles extended to 2,500+ words. Then submitted to IndexNow.
Result, day 2: Two of the five articles already appeared in the GSC top-25 list with 14-16 impressions each (uk-government-grants 16 impr at pos 14.1, singapore-sme-loans 14 impr at pos 37.8). The other three didn't appear in the 7-day snapshot but were indexed.
Takeaway: The combination of republish + Quick Answer + IndexNow ping accelerated reindexing from the typical 2-4 week wait down to days. The Quick Answer addition alone is hard to attribute (we changed multiple things at once), but the pattern across our portfolio suggests Quick Answer placement is one of the highest-ROI single changes for AI citation eligibility.
Case study 2: Recovery from a 96% impression collapse (Clarivian, May 2026)
Starting state: Clarivian impressions dropped from 2,295/day (May 3 peak) to 99/day (May 23), a 96% collapse over 20 days. Average position fell from 8 to 22. Googlebot crawl rate dropped from 62/day to 5/day. Cause: 8 articles silently unpublished (returning 404 to Googlebot) plus multiple top earners renamed (mass 301 churn).
Intervention: Republished 5 articles with no replacement, added 301 redirects from 3 with replacements, swept em-dashes, extended content to 2,500+ words, submitted to IndexNow. Built database-level governance constraints (CHECK constraint preventing naked unpublish, trigger preventing slug renames) to prevent recurrence.
Result (early signal, day 3 post-fix): Articles re-appearing in GSC. Recovery trajectory still in progress; the deeper case study will publish after the full 60-day measurement window closes.
Takeaway: Operational discipline matters more than content velocity. The collapse wasn't caused by content quality or algorithm changes; it was caused by routine content-management mistakes (unpublish without redirect, rename without 301) that we'd been making for months and only diagnosed after impressions collapsed. The governance constraints are now in production and prevent the entire class of mistake at the database level.
Case study 3: Migrating off WordPress (this site, jbai, 2026-05-26)
Starting state: jbai on WordPress with 207 published posts, 12,289 GSC impressions over 90 days (10K from one freak week), average position 35-70, near-zero clicks. Niche: SaaS tools reviews. Competition: G2, Capterra, Backlinko, vendor sites with DR 80-90+.
Intervention: Migrated to FastAPI + Postgres (forked from the wealth-wire pattern). Identified 12 AI-related keepers; 410'd ~200 SaaS reviews; set default to 410 for unknown slugs (catches Google-indexed ghost URLs). Repositioned site as 'AI Search & Citation Optimization' (this niche). Built governance schema (same as Clarivian/wealth-wire). Cutover via Apache vhost change (ProxyPass to 127.0.0.1:8004).
Result: Migration completed in a single session. All 12 keepers serving 200, all 410'd URLs serving 410 (clean deindex signal), Apache cutover with zero downtime. The deeper measurement (impact on impressions, citations, crawler activity post-pivot) requires 60-90 days and will publish then.
Takeaway: Migrating stack AND pivoting niche simultaneously cost less than doing them sequentially. With only 12 keepers in scope, the migration was trivial; with 207 it would have been a week of work.
Case study 4: OAI-SearchBot tracking across three domains (May 2026)
Starting state: No systematic tracking of which AI bots crawl which sites. The Clarivian admin monitor dashboard showed a 'ChatGPT' tile counting only ChatGPT-User, missing OAI-SearchBot and GPTBot entirely.
Intervention: Built a ~100-line Python script (/root/ops/jbai-chatgpt-tracker.py) that re-scans 14 days of Apache logs and writes a JSONL summary of OpenAI bot hits per day, broken down by GPTBot / OAI-SearchBot / ChatGPT-User. Scheduled via cron at 00:05 UTC daily. Added an OAI-SearchBot tile to the Clarivian admin monitor.
Result: Surfaced that OAI-SearchBot is the most-active AI crawler across all three sites (4-37 hits/day), GPTBot is bursty (typically 2-5/day with occasional 40+ spikes), ChatGPT-User correlates with user activity (typically 1-3/day). Daily visibility enables faster response if a site's AI bot traffic suddenly drops.
Takeaway: If you're not measuring AI bot activity per crawler, you're missing the leading indicator of AI citation visibility. The instrumentation is cheap and the signal is real.
Case study 5: 410 versus 301 versus 404 for retired content
Starting state: The jbai migration retired ~200 SaaS review posts that no longer fit the new GEO positioning.
Intervention: Three SEO disposition options for retired content: 301 (redirect to closest live replacement), 410 (explicit Gone, no replacement), 404 (default if nothing is set). We chose 410 for the majority because: there are no thematically-relevant replacements (the new GEO content doesn't match SaaS reviews), and 410 sends a faster deindex signal than 404 (Google treats 410 as 'deliberate' versus 404 as 'temporarily missing'). We additionally set the default for unknown slugs to 410 during the migration window because Google had ghost URLs in its index that we didn't have in the WP export.
Result: Deindexing typically completes 4-12 weeks faster with 410 than with 404 (based on Google's published guidance and observed behavior on prior migrations). Full deindex of the 200+ retired URLs expected by end of August 2026.
Takeaway: 301 to a non-thematic replacement is worse than 410 because Google treats large-scale 301-to-homepage as soft-404 spam. 410 is the honest signal for deliberate removal with no replacement. Use 301 only when the replacement is genuinely thematically relevant.
How to read our data
Where we publish absolute numbers (impressions, crawl rates, position changes), the numbers are real. Where we publish percentages or ratios, the denominator is real. Where we anonymize, it's clearly labeled. We don't publish: individual customer data, specific commercial agreements, raw log files with IP addresses, internal financial numbers beyond what's relevant to the case study.
The data flows back into the other four pillars. Findings on how AI engines cite content (pillar 1) inform our schema and content-structure recommendations (pillar 2). Findings on crawler behavior (pillar 3) inform our measurement infrastructure (pillar 4).
Case study 6: SBLOC pillar internal linking impact (Clarivian, 2026-05-26)
Starting state: Clarivian's SBLOC content cluster had 10 pillar-grade articles, but internal linking was uneven. The top earner (sbloc-rates-by-broker-2026, 458 weekly impressions) only linked to 2 other SBLOC pages. The fidelity-sbloc-rates article (90 weekly impressions at position 6.7) linked to zero other SBLOC pages. Link equity from these top earners was not flowing to the rest of the cluster.
Intervention: Added a "Related coverage" block to both under-linked articles, each cross-linking to 7 other pillar pages in the SBLOC cluster. Pattern: editorial spread with hairline dividers, short descriptions, single accent color. No interstitial CTAs or distractions; pure content discoverability.
Result: Every pillar page in the SBLOC cluster now has 5-7 outbound links to siblings. The cross-linking density allows link equity from the top earners to feed underperforming siblings. Measurement of impression and ranking impact requires a 30-day window; will update this case study with the actual numbers when the window closes.
Takeaway: Internal linking is one of the most under-invested SEO and GEO levers because it produces no measurable result in week one and compounding result over months. The audit pattern is simple (grep blog content for cross-references to other slugs in the cluster, flag pages with fewer than 3 outbound cluster links) and the fix is one-shot. Worth doing for every topical cluster on a content site.
Case study 7: Filling a content gap from GSC data (Clarivian, 2026-05-26)
Starting state: Clarivian's GSC showed Google attempting to surface the site for "corporation tax calculator" and "corporate tax calculator" queries, but Clarivian had only a UAE corp tax estimator. Position was 93-98 (page 10), 6-7 impressions per week, zero clicks. The query demand was real; the destination was wrong.
Intervention: Built /tools/uk-corporation-tax-calculator-2026: a JavaScript-based estimator handling the 19% small profits rate, 25% main rate, marginal relief calculation (standard fraction 3/200), associated companies threshold division, and short-period proration. Added FAQPage JSON-LD with four common questions about UK corp tax. Wired the tool into the SLUG_REDIRECTS dict so UK-related articles surface it in their "Related tools" section. Added to sitemap. Submitted to IndexNow.
Expected result: Google should re-evaluate within 7-14 days and start ranking the new tool for the target queries. Initial position will likely be position 20-40; improvement to page one depends on whether the tool earns engagement signals (time on page, calculations performed). The tool is in the sitemap and pinged to IndexNow as of the publish time of this case study.
Takeaway: GSC's "Performance > Queries" report contains free demand signal that often points to content gaps. The pattern is: query appearing at position 60-100 with multiple impressions per week means Google is looking for a relevant destination on your domain and not finding a great one. Building a focused tool or article exactly matching the query intent is one of the highest-conviction content investments available.
How we measure case study outcomes
Each case study has an explicit measurement window declared upfront. Typical windows:
- 14 days: AI crawler activity (OAI-SearchBot, PerplexityBot hits per affected page).
- 30 days: Internal link equity changes (impressions on previously-orphan pages).
- 60 days: AI Overview citation behavior changes (very rough; requires manual citation tracking).
- 90 days: Position changes from a major content or technical intervention.
We re-visit each case study at the end of its measurement window and append the actual result. If the result was nothing or worse than baseline, we say so. The credibility of the case study series depends on publishing the misses, not just the hits.
What's next in the case study queue
Planned case studies once their measurement windows close:
- Full 60-day recovery of the Clarivian impression collapse (currently 3 days in).
- Impact of the WP-to-FastAPI migration on jbai's GSC presence (currently day zero).
- Effect of adding HTML 410 default to unknown slugs on deindex velocity.
- Whether the 5 pillar pages we just published gain visible OAI-SearchBot crawl within 14 days.
- The post-pivot publish-agent rewrite (when complete) and its content quality compared to the WP-era agent.
The pace of publishing is moderate (one to two case studies per month) but the per-case-study depth is meaningful. We'd rather publish one well-instrumented case study than five shallow ones.
Reproducibility statement
Every case study published in this pillar is reproducible in the sense that the technical interventions are described in enough detail that another team running their own site can apply the same change and measure the result. We don't publish raw log files (privacy, operational security) but we publish the analysis scripts, the SQL queries, the schema additions, the robots.txt patterns, the route handler code.
If a case study reports a result, the result is a real number from a real site we own. If a case study describes an expected result before the measurement window closes, we say so explicitly. If a case study's result turns out to be nothing measurable, we publish that too; the credibility of a series like this depends on reporting the misses as honestly as the wins.
The instrumentation across our portfolio is unified enough that we can compare interventions on Clarivian against equivalent interventions on jbai or wealth-wire. The cross-site comparison is one of the strongest things we can offer because most published GEO advice comes from single-site observations that may or may not generalize.
FAQ
Why publish anonymized data instead of full numbers?
Some data points shouldn't be public. The rest can be. We publish the parts that help others and anonymize the parts that don't.
Will you share raw log files?
No. Raw logs contain IPs and operational details we don't want public. We publish summarized, processed data with the analysis. The methodology is reproducible.
How quickly do GEO changes show measurable results?
AI crawler activity changes within days of a structural improvement. Citation visibility takes weeks. Referrer traffic takes months to accumulate enough to read. The fastest measurable signal is OAI-SearchBot hit rate in server logs.
Are case studies sponsored or paid?
No. Every case study uses data from sites we own. We don't accept sponsored case studies because the value is in the credibility, and credibility requires that we have no incentive to misrepresent what happened.