Build, own and operate.
A Munich dance school's website: built in January 2025, measured for 16 months, rebuilt on what the numbers said, and relaunched on 23 September 2026 with a scheduling plugin I wrote.
- Built
- One system for eight course types, a redirect map from 16 months of Search Console data, a timetable rendered from records instead of an image.
- Measured
- Organic sessions +169% year over year on the 2025 build. No pre-launch baseline. Four weeks after the relaunch: Google AI-feature impressions about 3x. Readout due 4 Nov and 16 Dec.
- Still open
- Hand-built teacher profiles, no English version; the mobile timetable shipped as 1.11.21. At the end, with what I would do.
Headline result

- Site built
- Tracking begins
- Operating
- Rebuilt on staging
- Live
- Readouts
Chapters
Measured on the 2025 build. The relaunch went live on 23 September 2026.
No 2024 baseline exists: kayjays.de was the first site I built and tracking was set up as part of building it. The figures above are a like-for-like comparison, not a rebuild-uplift claim.
The home page also cites, from the same Search Console property, non-brand, 4 May - 31 Jul 2026 against the same weeks of 2025: clicks 59 → 146, average position 30.7 → 17.8, CTR 0.5% → 2.3%.
On the 2025 build, before the relaunch
“Since the new website went live, my sign-ups have at least tripled. Meili finds the right solution for every challenge - and keeps going until everything is really right.”
Jessy, owner, KayJaysTranslated from GermanRelaunch readout · pending
Baseline captured 21.09.2026 · readouts 4 Nov and 16 Dec 2026
| Metric | Before | After |
|---|---|---|
| GSC clicks, 30 days before / after | 265 | Pending |
| Pages indexed | 27 | Pending |
| Organic engagement rate | 70.6% | Pending |
| Rich results vs schema baseline | Pending | Pending |
The relaunch went live on 23 September 2026. Three weeks later:
Recommended first by ChatGPT
Jessy expressed that she wanted to attract more adult students to her hip-hop classes. The relaunch gave AI assistants something to read: one source of truth for every class time and price, and a course page that answers the questions a beginner asks. The method worked. In October 2026, a logged-out ChatGPT session asked for adult hip-hop classes in Munich returned KayJays as its number-one recommendation, in German and in English, with a “best for beginners” label. Both answers draw the class time, the trial price and the address from kayjays.de.
ChatGPT, logged out, October 2026. Left: German prompt. Right: English prompt.
Early signal, four weeks after the relaunch. The formal readout above follows on 4 November.
AI answers show the new site three times as often.
Search Console's generative AI features report counts how often kayjays.de appears in Google's AI Overviews and AI Mode. In the three weeks before the switch it averaged about 11 impressions a day; in the four weeks after, about 34, and about 42 in the last two of them. From August to mid-October, AI-feature impressions grew about 4.5 times while all Google impressions grew about 2.4 times, so AI visibility rose roughly twice as fast as search visibility overall. The new course pages are already among the cited pages. About 28% of AI impressions still land on old URLs that now redirect: Google has not finished swapping them for the new ones.

Four weeks in · early signal
Before and after the switch on 23.09.2026 · formal readout 4 Nov 2026
| Metric | Before | After |
|---|---|---|
| Google AI-feature impressions per day | ~11 (1-22 Sep) | ~34 (23 Sep - 19 Oct) |
| All Google impressions per day | ~150 (Aug) | ~350 (1-14 Oct) |
| Search Console clicks per day, October | ~4 (Oct 2025) | ~15 (1-14 Oct 2026) |
| Average position, October | 13.3 (Oct 2025) | 8.8 (1-14 Oct 2026) |
| Search Console clicks, September (switch on the 23rd) | 131 (2025) | 352 (2026) |
| Trial-class bookings per day | ~1 (May - Jul, old form) | ~4 (since launch, plugin) |
Limits: four weeks is short, and the autumn course season started around 13 September, before the switch, so part of every rise here is season. The trial-booking comparison uses a different form before and after and has no autumn 2025 baseline, because form entries before 15 April 2026 were deleted. The AI report counts impressions only, not clicks. These figures are not the readout; the 4 November and 16 December readouts in the table above still stand.
Eight course types, one system.
Intent, audience, schedule, location and teacher are separate elements, so a course page is assembled, not written. That is what made the 2026 relaunch a re-composition rather than a rewrite.
Average position by term · German terms rank, national heads don't
Left of the dashed line is page one (position 10 or better).
- shuffle
- 6.65
- First page
- kpop
- 5.47
- First page
- breakdance kinder
- 8.39
- First page
- dancehall
- 7.08
- First page
- bauchtanz
- 16.82
- Opportunity
- kindertanz
- 20.71
- Opportunity
- kinderballett
- 29.06
- Opportunity

The detail: growth, architecture, local intent and language
Growth. Organic sessions rose 169% year over year on a like-for-like window, and organic moved from 47.6% to 56.6% of all sessions. Engagement on organic traffic held at 70.5%. Search Console recorded 2,833 clicks over the same period. Growth came from the course architecture and from posts that ranked on style and audience terms, a split that turned out to matter.
Architecture. Eight course types were modelled as one system rather than eight page templates. Intent, audience, schedule, location and teacher are distinct elements, so a course page can be assembled from them instead of written from scratch.
Local intent and language. German is the language of the demand. Style names stay in their original form where searchers use them, and audience and location terms are German throughout; the naming follows the query, not the brand's preference.
The traffic sat on posts, not the architecture.
The site's largest organic assets were a K-Pop post and a Shuffle post, not the course pages the system was built around. So the relaunch merged posts into course pages, put audiences under style hubs, targeted the terms that convert for one studio, and opened every course FAQ with the price.
The K-Pop course itself came out of the data: keyword research showed local K-Pop demand the studio wasn't serving. The owner had been weighing the course already; the numbers made the case, she launched it, and the K-Pop post became the site's second-biggest page.
The call. Style × audience × Munich long tails, the local pack as the channel, a small national "learn this style" layer at zero difficulty. Not the biggest numbers: volume is not demand the studio can meet.
Found in the baseline. Two thirds of homepage clicks were landing on http://www, because the old server sent http traffic to the homepage and dropped the path. One host rule fixes it; measured again at the first readout.
The detail: the strategy, the refused terms, the four decisions, the redirect map
The strategy the data produced. Win on style × audience × Munich long tails, where nobody has built a real page and where the studio actually has the class to match the query. Treat the local pack as the primary channel, not the blue links: it fires on almost every commercial query in this niche. Add a small national "learn this style" layer where difficulty is zero and volume is five to twenty times the local terms, for topical authority and AI-answer visibility the local pages cannot earn on their own.
Not targeted, on purpose. "tanzschule münchen" (900 searches a month) and "ballett münchen" (2,000) are both traps, for different reasons: the first is a local-pack query a studio in Berg am Laim cannot win against the city centre; the second is owned by ballet schools and would bring searchers the studio's two children's classes cannot serve. Volume is not demand the studio can meet.
The single highest-leverage move in the plan was not content. It was the Google Business Profile, because the local pack sits above the organic results on nearly every query that converts.
The four decisions. Merge posts into course pages, so the ranking asset and the conversion page are the same URL. Style × audience as hub and children, so each style hub carries its audience pages beneath it rather than competing with them. Conversion terms over national heads: kindergeburtstag münchen and the local pack, in place of national style heads that produce out-of-market clicks. Price question first: all 15 course pages open their FAQ with the price, because that is the question being asked.
Migration discipline. The redirect map was built from the Search Console Pages export covering 16 months and 166 URLs, not from the sitemap or a crawl, both of which were incomplete. 26 path redirects plus one host rule for www and http went live; 9 of 26 were sampled on launch day, all single-hop 301. About 120 URLs left behind by an old events plugin were deliberately not redirected: they had 15 clicks in 16 months and no page on the new site matches them, so they return 404 instead of pointing somewhere misleading.
Largest organic assets, 16 months to 19.09.2026. K-Pop post 380 clicks (position 7.96), Shuffle course page 353, Shuffle post 119.
The timetable was one image. Now it is 27 records.
Times, teachers and prices used to be typed on pages and drawn in a PNG with no alt text, and they drifted apart. Now they are entered once in the plugin and rendered everywhere: schedule, teacher grid, course panel, structured data. The studio owner moves a class, cancels a date or changes a price in one place, and every page follows. The plugin is written so a small business can keep its own site correct without a developer.
One record, four surfaces
- Schedule blockPer course, teacher and location
- Teacher gridAnd course-teacher cards
- Course panelDynamic prices and sign-ups
- JSON-LDCourse schema from the same records as the display
Why it mattered
A schema plugin was hand-carrying 13 Course and 7 Person entries on one page, one for a teacher who had left, and suppressing the SEO plugin's own graph. Removed before launch.
Tested on live
Sign-up and cancellation with both confirmation mails, logged out, within 20 minutes of the switch.
The detail: the stale-fact story and the rule
A fact written twice goes stale once. Class times, teachers and ages were hand-written in prose on teacher and course pages and also rendered by the schedule. One Tuesday class moved to 18:30 while the page still said 17:50. A fact sweep across 50 pages also found invented facts on pages written before fact-gathering, including phone hours nobody had confirmed.
The rule adopted. Schedule, course and price information lives only in the plugin backend. No class time, course teacher or price is typed by hand on a course or schedule page. Prices render via shortcode from one price record. Teacher profile pages are the one place this does not hold yet; generating them from the records is next.
One config line decided which site served.
The hosting could not repoint the domain, so the new database sat beside the old one and one line chose between them. Rehearsed both ways on a copy first. Rollback kept two weeks, never used. On the way I found the old server config sent every plain-http request to the homepage, dropping the path; the new one keeps it and redirects in one hop.
What operate means.
- 52 → 0 pre-publish check boxes across 45 pages, before launch. Open questions collected in one document for the owner rather than sent ad hoc.
- 45 pages swept before launch: no missing description, no over-long title, no duplicate.
- 3 legal gaps found in the audit: no Impressum on the live site, cookie tooling installed but never configured, privacy text citing only the BDSG.
- Retention runs itself: drop-in sign-ups deleted 30 days after the class, form entries after six months, newsletter records on unsubscribe. Photo consent moved to a separate opt-in.
Still open, and what I would do.
- Mobile timetable: shipped. At launch it still side-scrolled on a phone; below 1120px the grid now switches to day tabs (plugin 1.11.21, built 24 September, live on kayjays.de by 28 September).
- Teacher profiles are still hand-built pages, the one place the one-source rule does not hold yet. Generating them from the records is next.
- No English version. Not until the language split in Search Console shows demand for one; a second language doubles every future edit.
- Course-category strength and homepage impressions were addressed in the relaunch; the readouts say whether it worked.
- Pre-launch numbers from the old builder site are unrecoverable. Stated, not worked around.
Sources and limits.
Figures come from GA4 and Google Search Console for kayjays.de, on like-for-like windows, each stated beside the figure it supports. The relaunch baseline was captured two days before the switch; readouts follow on 4 November and 16 December 2026, and anything not yet measured is marked Pending. A second analytics plugin that could also load the GA4 tag was removed at launch, so sessions before and after that date are compared only once a single tag is confirmed. Key-event figures are not quoted; the definition changed mid-period.



