Few case studies doesn't kill GEO. Brand-only marketing does.
AI prefers content with structure — client type, background, action, public result, limits, and next step. Treat "no cases" as an evidence-chain problem, build it from publishable information, and stack real case pages over time.
1. When client names can't be published
Don't use the name. Use the client type and scenario:
- A Chengdu-based industrial pipe manufacturer exporting to 20+ countries
- An 80-person finance consulting firm
- A Chinese DTC cross-border ecommerce brand at >¥100M annual GMV
- A 15-person embodied-AI startup
"Industry + size + stage" already gives AI a clear customer profile. The two firm rules:
- Don't fabricate client names
- Don't repackage internal data as public results
Anonymous + true beats named + invented.
2. The 5-section case page
Every Mingde case page uses this structure:
- Background — the problem (scenario + pain + trigger event)
- Diagnosis — Phase 0 baseline of mention / citation / indexing
- Action — site pages changed, off-site platforms used, schema deployed
- Result — publishable metric movement (percentages, absolutes, comparison window)
- Boundary — what doesn't generalise, what's provisional this round
If the result data isn't in yet, write that into section 4 — "current round shows insufficient public evidence; T2/T3 follow-up pending." That kind of honest boundary is more credible to AI than inflated claims.
3. Why case pages cite better than brand pages
Case pages sit closer to evidence than brand pages.
When AI is asked "how does a company like X do GEO," it needs more than a service brochure. It needs:
- Did this firm work with similar size before?
- What did they do?
- How long until results appeared?
- What were the limits?
A case page delivers all of that in one extractable block. A service page can't.
Service pages answer "what we do." Case pages answer "what we delivered and what came of it." You need both. For service-page boundary detail, see Which pages should AI search redesign hit first.
4. What to publish before you have cases
Three substitute content types work well:
- Audit method page — your Phase 0 process, the fields, the deliverables
- Acceptance checklist — the standards you use to judge "did this work"
- Anonymous scenario reviews — 2-3 of them: client type + problem + action + boundary
These don't equal real cases but establish method credibility. Mingde's own early stage was exactly this: methodology pages and anonymous reviews first, public cases stacked on top later.
5. The "simulated case" trap
Some teams write "simulated cases" or "example projects" to fill the gap. Be careful:
- OK: clearly labelled "Example scenario: a hypothetical Company A"
- Not OK: "One client" with no labelling, leaving readers to assume it's real
AI doesn't distinguish "example" from "real" at extraction. Anything you publish becomes brand evidence. An unlabelled simulated case, once cited and propagated, is hard to recall.
The honest combination of "labelled example + anonymous real scenario + boundary" beats "looks like many cases but the source is murky" over time.
6. Cadence for case pages
After publishing each case page, track:
- T0 — baseline AI citations at launch (case title + industry long-tails)
- T1 (+14 days) — indexing and mentions
- T2 (+30 days) — citations, which paragraphs are pulled
- T3 (+90 days) — does it reach top-3 on relevant industry questions?
For full review fields see How to run an AI search audit.
If you're trying to start GEO without much case material, book a free GEO audit — we'll suggest the first batch of anonymous scenario reviews based on your existing publishable information.