Card no. 01 · Content Engine
Tumblegrain — Plant-based meal-kit subscription · Melbourne
How a Melbourne meal-kit startup built a 400-recipe content moat
15 months · Melbourne, Australia · Hospitality & Food · modelled scenario
1The challenge
A brilliant recipe box, nailed shut.
Sign-ups depended entirely on paid social, and acquisition costs kept rising. The recipe archive was behind a login, so there was no organic footprint.
Tumblegrain had written 418 tested plant-based recipes for its weekly kits, the exact thing people in Melbourne search for every night at 5 pm. But only paying members could open them. Each new subscriber was bought one ad click at a time, and every rise in auction prices went straight onto the cost of a trial.
- sign-ups from paid social
- 100%
- sign-ups from paid social
- indexable pages
- 11
- indexable pages
- recipes Google could see
- 0
- recipes Google could see
2Diagnosis
The kitchen inspection.
Six findings from the crawl, the query map and a look at who AI answers were citing instead. Each one became a card in the plan.
- TechnicalCritical
418 recipe URLs behind a login
Every recipe 302-redirected to /login. Googlebot could reach 11 marketing pages and nothing a hungry person searches for.
- SchemaCritical
No Recipe structured data
Ingredients lived inside a JPEG of the printed card. No cook time, yield, nutrition or rating, so no eligibility for recipe carousels.
- ArchitectureHigh
Organised by delivery week, not by hunger
Recipes were filed as “Week 37 box”. Nobody searches that; they search “high protein vegan dinner” or “air fryer tofu”.
- AI OverviewsHigh
Nothing quotable
Method text was one long paragraph. AI answers for plant-based queries cited big publishers whose steps were numbered and sat near the top.
- ContentMedium
38 stale launch-era posts
Thin listicles with 2022 prices and discontinued kits, three of them competing for the same “vegan meal prep” intent.
- PerformanceMedium
LCP 5.1 s on mobile
2.4 MB hero photos and a client-rendered recipe app injecting ingredient lists after hydration.
3The strategy
Open the box. File every card where hunger looks for it.
Built on our Content Engine plan: recipes became public, members kept the planner, swaps and delivery. Four methods did the rest.
- 1
Topic clusters
2,300 plant-based queries grouped into six alphabetised clusters (A, B, H, L, S, T). Each cluster has one pillar guide and a drawer of recipes, and every recipe links up to its pillar and across to two siblings.
- 2
Recipe schema
Server-rendered Recipe JSON-LD on every card: ingredients, numbered steps, total time, yield, nutrition and aggregate rating from verified kit buyers. Validated weekly, so a template change can't silently break eligibility.
- 3
AI Overviews
Each recipe opens with a 40-word answer block (time, protein per serve, the one trick) and numbered steps. Pillars carry comparison tables AI answers can lift and cite.
- 4
Content refresh
38 launch-era posts were merged, redirected or rewritten into pillars. Seasonal recipes get refreshed every quarter with Victorian produce timing, before demand peaks.
4The index box
Six tabs. 418 cards. Watch them rise.
Each tab is a topic cluster; shown are its four hero recipes. Drag through the 15 months and the cards climb out of the box as their Google position improves.
A — Air-fryer plant-based · 58 recipes in the cluster · 4 of 4 hero cards on page one
Cards rise as they climb Google: buried = not ranking, halfway = top 20, fully out = top 3. Stained cards are top performers. Press a card to flip it.
5Method
15 months, one step at a time.
- 01Month 1 · Jul · 9,000 sessionsKitchen inspection: crawl, gated-URL inventory, 2,300-query map
- 02Month 2 · Aug · 10,200 sessionsUngate the archive: recipes public, planner & swaps stay members-only
- 03Month 3 · Sep · 12,000 sessionsServer-rendered recipe template with Recipe schema
- 04Month 4 · Oct · 15,000 sessionsClusters A & B live: air-fryer and batch-cooking pillars
- 05Month 5 · Nov · 19,000 sessionsClusters H & S live; first carousel placements
- 06Month 6 · Dec · 25,000 sessionsAnswer blocks rolled out to 200 recipes for AI Overviews
- 07Month 7 · Jan · 32,000 sessionsContent refresh: 38 old posts merged or redirected
- 08Month 8 · Feb · 41,000 sessionsClusters L & T live; image pipeline cuts LCP to 1.9 s
- 09Month 9 · Mar · 52,000 sessionsRecipe → trial CTA tested on cards (‘cook this in a kit’)
- 10Month 10 · Apr · 63,000 sessionsAutumn seasonal refresh ahead of the laksa & soup peak
- 11Month 11 · May · 74,000 sessionsInternal-link pass: 3,100 contextual links between siblings
- 12Month 12 · Jun · 86,000 sessionsRatings from verified buyers feed aggregateRating
- 13Month 13 · Jul · 97,000 sessionsWinter refresh; second wave of answer blocks
- 14Month 14 · Aug · 108,000 sessions400th recipe published; schema validator in CI
- 15Month 15 · Sep · 118,000 sessionsSpring refresh; CAC review shifts budget from paid social
6Results
What came out of the box.
Modelled scenario for a fictional business — not a client result. Rankings and traffic are never guaranteed; see our Terms.
The ten biggest cards
| Query | Tab | Searches / mo | Month 1 | Month 15 | SERP feature |
|---|---|---|---|---|---|
| air fryer tofu | A | 14,800 | not indexed | #2 | Recipe carousel |
| salt and pepper tofu | T | 12,100 | not indexed | #2 | Recipe carousel |
| high protein vegan dinner | H | 9,900 | not indexed | #2 | AI Overview cite |
| tofu scramble recipe | T | 9,900 | not indexed | #5 | Recipe carousel |
| vegan laksa recipe | S | 8,100 | not indexed | #1 | Recipe carousel |
| vegan meal prep for the week | B | 6,600 | not indexed | #3 | AI Overview cite |
| vegan risotto recipe | S | 6,600 | not indexed | #4 | Recipe carousel |
| vegan lunchbox ideas | L | 5,400 | not indexed | #5 | AI Overview cite |
| how to press tofu | T | 5,400 | not indexed | #10 | People also ask |
| air fryer chickpeas | A | 4,400 | not indexed | #9 | Recipe carousel |
“high protein vegan dinner”, before and after
Month 1
Tumblegrain: not in the top 100. The recipe existed, behind a login.
Month 15
Two carousel slots plus the AI citation.
Get a plan like this
Have an archive of great content nobody can find? We’ll map it, open it up and file it where people search.