Skip to content
Get Ranked On TopIllustrative case studySignal plan

A fictional business. The figures are a modelled scenario, not a client result.

Florhaven Glasshouse Trust · Botanical garden & plant database · Ghent

How a Victorian glasshouse got its 12,000-species database back into search

+130% visits from search, 91% of species pages indexed

16 months · Europe · Signal plan · Canonicalisation · Crawlable pagination · Core Web Vitals · Log-file analysis

3,100 → 10,90010,900 of 12,000 species pages indexed
Fig. 1 · The Palm House, its twelve broken panes reglazedbefore: 3,100

The challenge

Twelve thousand species under glass, and the glass had clouded over.

A legacy CMS served Florhaven’s plant database through duplicate URLs, broken pagination and slow pages. Most of its 12,000 species pages were never indexed.

Florhaven’s plant database is the garden’s second collection: 12,000 species, each with a herbarium scan, a hardiness note and its bed in the glasshouse. Gardeners, students and curious visitors search for those names every day. But the CMS behind it had been patched for fifteen years, and every patch gave the same plant another address. Google found 71,400 URLs, spent its visits on the copies and gave up on the rest.

-33%change in visits from search over the two years before the work
One species, six addresses

Victoria amazonica

  • /plants/view.php?id=4471
  • /plants/view.php?id=4471&sort=az
  • /plants/view.php?id=4471&view=grid
  • /plants/view.php;jsessionid=8F2A0C…?id=4471
  • /Plants/View.php?ID=4471
  • /plants/euryale-amazonica

Canonical URL/species/victoria-amazonica/

Every copy split the signals for the same page. Now each one redirects in a single hop to /species/victoria-amazonica/.

Index coverage before the work · one cell = 100 species pages
  • Indexed · 3,100
  • Not indexed · 8,900

Diagnosis

What the survey of the ironwork found.

Duplicate URLs

71,400

URL variants for 12,000 species: sort and view parameters, session IDs, upper-case paths and old synonym names that each had a page of their own. Not one canonical tag on the site.

Finding 1 of 4

Pagination

380

Listing pages Googlebot could not reach: the A–Z index loaded more names only through JavaScript, and family listings pointed the canonical of page 2 onwards at page 1.

Finding 2 of 4

Page speed

5.8 s

Mobile LCP at p75 on species pages. Herbarium scans went out as 4–9 MB TIFFs, every view queried the database uncached, and an embedded garden map blocked rendering.

Finding 3 of 4

Crawl budget

64%

Share of Googlebot requests spent on parameter and session URLs in twelve weeks of server logs. Another 2,100 species had no internal link at all.

Finding 4 of 4

Googlebot requests, before · log-file sample
  • Parameter & session URLs
  • Pagination loops
  • Species pages
  • Other
Googlebot requests, month 16 · log-file sample
  • Parameter & session URLs
  • Pagination loops
  • Species pages
  • Other

The strategy

Four repairs, one cast-iron frame.

Built on our Signal plan: the technical engagement for sites where crawl, rendering and architecture decide the outcome.

  1. Canonicalisation

    One URL per accepted name: /species/<genus-species>/. All 59,000 variants mapped and 301-redirected in one hop, parameters stripped from every internal link, a self-referencing canonical on every page. Synonyms redirect to the accepted name and are listed on its page.

  2. Crawlable pagination

    Real links behind the A–Z and family listings, a canonical on each page that points at itself, a 404 beyond the last page, and genus hubs that link every species. No species is more than three clicks from the home page.

  3. Core Web Vitals

    Herbarium scans turned into responsive AVIF and WebP, species pages served from an edge cache, the garden map deferred until someone opens it. Field data watched for each part of the database.

  4. Log-file analysis

    Server logs ingested daily from the first week: which URLs Googlebot asked for, what it got back and how long it waited. Each repair was confirmed in the logs before the next one went live.

Replace the panes

Twelve broken panes, twelve faults. Reglaze them and watch the index fill.

Each cracked pane is one fault from the audit, marked with its kind. Click a pane on the facade, or use the work order below, to replace it; press again to break it and compare.

Species pages indexed
3,100of 12,000
Organic visits / month
17,0000%
Mobile LCP (p75)
5.8 sgood ≤ 2.5 s

Choose a broken pane, on the facade or in the work order, to read the fault behind it and how it was repaired.

Work order · twelve panes

Duplicate URLs

0 of 4

Pagination

0 of 4

Slow pages

0 of 4

Month by month

16 months, pane by pane.

  1. Months 1–2

    Survey

    Full crawl of 71,400 URLs, twelve weeks of server logs, Search Console coverage and a field-data baseline.

  2. Months 3–4

    Canonical model

    One URL per accepted name, a redirect map for 59,000 variants, synonyms merged together with the curators.

  3. Months 5–7

    Pagination

    Crawlable A–Z and family listings, genus hubs and one XML sitemap per plant family.

  4. Months 8–10

    Speed

    An image pipeline for the herbarium scans, the edge cache, the deferred garden map, backups moved off the live server.

  5. Months 11–16

    Reglazed

    Monitoring in the logs and Search Console; new species pages published straight into the clean structure.

Results

A clear roof, and the light gets in.

Species pages indexed

3,100 → 10,900

3,100 → 10,900 of 12,000 species pages

Organic visits / month

+130%

17,000 → 39,100 visits / month

Duplicate URLs retired

59,000

71,400 URL variants → 12,400 canonical URLs

Mobile LCP (p75)

5.8 s → 1.9 s

75th percentile on mobile; “good” is 2.5 s or less

Modelled scenario for a fictional business — not a client result. Rankings and traffic are never guaranteed; see our Terms.

Species pages indexed, by month
04,0008,00012,000M1M4M7M10M13M1610,900
Organic visits per month
020,00040,000M1M4M7M10M13M1639,10017,000

Core Web Vitals, field data

LCP (p75, mobile)5.8 s → 1.9 s
TTFB (p75)2.4 s → 0.3 s
CLS (p75)0.31 → 0.03

The vertical rule marks the “good” threshold.

Positions for searches the garden cares about

QueryBeforeMonth 16
victoria amazonica bloei#34#3
plantentuin gent#9#2
serre victorienne gand#22#4
titan arum flowering belgium#41#5
trachycarpus fortunei winterhard#57#8
nepenthes rajah care–#11

Queries as people typed them, in Dutch, French and English. A dash means not in the top 100.

Get a plan like this

Is a large database hiding from search behind duplicate URLs, broken pagination or slow pages? Start with the Signal plan, or tell us what your site holds.

← All case studies