CASE_STUDY_01 · Digital wallet
Making AI visibility predictable and repeatable.
0% →0%
AI search visibility share · ChatGPT
Folio Wallet
Average rank 2.4
Total mentions 20
Top 10 competitors map · X average rank · Y total mentions
Client
Folio Wallet
Service provider
Turbine
Timeframe
10 weeks of optimization
+ 3 weeks of stabilization
Primary model
ChatGPT
Key metric
AI search visibility share
The client
Folio Wallet is a London-based digital wallet app for storing personal documents such as IDs, passports, payment cards, and medical records.
At the start of the project, Folio already had:
What they didn’t have was measurable visibility inside large language models, specifically, being mentioned or cited when users asked ChatGPT relevant questions in their category.
The timeline
AI search visibility share · Folio
0%
PHASE_01
Folio not mentioned
PHASE_01
We ran initial tests across a wide range of prompts. It confirmed two things:
Result: 0% visibility. Folio did not appear in model outputs.
PHASE_02
We started with two fundamentals:
We deliberately looked for prompts at the intersection of:
That intersection became the core prompt set we tracked going forward. We describe the general methodology for building prompt sets here.
We also chose ChatGPT as the primary model to measure. While AI visibility work can create spillover effects across multiple models, strategies differ by model. Given ChatGPT’s current query volume, we treated it as the reference point.
Result: the initial setup and technical fixes resulted in the first 18% of visibility share for the chosen set of topics and prompts.
PHASE_03
This was the main execution phase. We focused on three core thematic clusters:
What we did:
Results:
Week 1
20%
Week 2
37%
Week 3
50%
Following month
43%
PHASE_04
For three weeks, we made no changes at all.
We simply observed whether visibility would hold.
Result: it did. The gains were not short-term fluctuations.
Why this worked
AI visibility comes from systematic work on the topics and prompts where a product wants to appear. The payoff is more users: traffic from chatbots and customers who found the product through model recommendations.
Most tools just count mentions, and most agencies rely on experience. LLMs see the world through vectors, so we use maths to estimate in advance which texts will surface for a prompt.
Models fluctuate, but with regular tracking a move from 0 to the 30–50% range is clearly measurable.
What we did in practice
For every topic (and for every prompt within that topic) we followed the same procedure. Below is a simplified example based on a single prompt.
PromptEncrypted vault for personal documents
STEP_01Initial assessment
Visible brands sat in one semantic cluster (blue), Folio in another (white). In this state, Folio had a low chance of being shown.
Semantic clusters · current content
Two clusters · brand is not visible
STEP_02Content creation
LLM response · annotated entities
If you’re looking for free apps to securely store and manage your documents and cards, here are some options to consider:
DocStorer allows you to create a protected digital archive of personal files, including documents, notes, cards, and receipts. The appoffers PIN security to protect your information. (docstorer.com)
Folio Wallet enables you to scan and securely store various cards and documents. It offers features like PIN and biometric authentication, ensuring your sensitive information remains protected. (folio.id)
STEP_03Evaluation before publishing
We estimate the chances before publishing. The brighter the cell, the closer a text is to those already surfaced (first five rows).
Folio's current page peaked at 0.84, not enough to appear. The new version reached 0.90, on par with visible brands.
Semantic similarity matrix
| VGSSOFTWARE | KEEPERSECURITY | NORDLOCKER | PROTON | DROPBOX | FOLIO | FOLIO_NEW | |
|---|---|---|---|---|---|---|---|
| VGSSOFTWARE | 1.00 | 0.91 | 0.88 | 0.89 | 0.86 | 0.83 | 0.88 |
| KEEPERSECURITY | 0.91 | 1.00 | 0.88 | 0.89 | 0.86 | 0.82 | 0.90 |
| NORDLOCKER | 0.88 | 0.88 | 1.00 | 0.87 | 0.85 | 0.81 | 0.85 |
| PROTON | 0.89 | 0.89 | 0.87 | 1.00 | 0.85 | 0.79 | 0.85 |
| DROPBOX | 0.86 | 0.86 | 0.85 | 0.85 | 1.00 | 0.78 | 0.86 |
| FOLIO | 0.83 | 0.82 | 0.81 | 0.79 | 0.78 | 1.00 | 0.84 |
| FOLIO_NEW | 0.88 | 0.90 | 0.85 | 0.85 | 0.86 | 0.84 | 1.00 |
STEP_04Evaluation after publishing
After publishing, the model cited the new content and Folio started appearing consistently. Clustering confirmed it: Folio moved into the same cluster as the visible brands (pink).
Semantic clusters · after publishing
One cluster · high chances to get visible
Overall results
RESULT_01 · Leading position
Folio moved from low visibility to a leading position in its category. A clear narrative was established and reinforced.
Top 10 competitors map · Before
Top 10 competitors map · After
RESULT_02 · Primary citation source
The Folio website became a primary citation source when models talk about Folio.
Source connections · brand → cited source
RESULT_03 · New territory held
Strategically, Folio
All of this while competing with products like Google Wallet and Apple Wallet, as well as long-established players in the travel segment.
Knowledge base10 oct 2026
Why measuring vectors and semantics beats simple mention tracking
Knowledge base10 oct 2026
Turn SEO keywords into prompts and build your AI visibility tracking set
Research10 oct 2026
An empirical study comparing OpenAI, Gemini, and SBERT embeddings for GEO
Thanks — you’re in. Check your inbox soon.
