Taiwan · 2023 to 2025 · MA interaction design research
GreenFilter 綠濾
Every purchase already moves money toward a kind of future. Green Filter explored how an AI companion could make that consequence visible at the moment a young adult chooses what to buy, then connect the same decision to saving, investing, health, climate, and company accountability.
Eligible completed responses used in the published results.
48+
universities represented
Twenty-one campuses were visited in person during recruitment.
32
moderated tests
Face-to-face prototype sessions across seven universities.
100+
self-tests
Anonymous prototype trials at more than twenty universities.
7
expert interviews
Sustainability, finance, policy, technology, and behavior perspectives.
The product problem
Caring is not the same as being able to act.
The research found a wide gap between environmental concern and purchase behavior. Students believed their spending affected the environment, yet the evidence needed to act was scattered across labels, company reports, news, product pages, and unfamiliar financial language. Price, convenience, health, trust, and product quality still had to be resolved in seconds.
The existing experience
A product page makes the transaction easy and the consequences invisible.
Commerce interfaces are excellent at showing price, images, reviews, color, payment plans, and urgency. They rarely explain where a product came from, who made it, what materials it contains, what risks or externalities sit behind it, or what kind of company receives the money.
The Green Filter proposition
Turn hidden product history into one usable decision layer.
Green Filter combines environmental, health, labor, product, company, and financial signals, translates them into plain language, and gives the user a practical next move: understand, compare, avoid, choose an alternative, or ask a deeper question.
Research system
The prototype came after the fieldwork.
Green Filter was developed as a mixed-method research program at National Cheng Kung University. The work moved from systems literature and expert perspectives, through a large behavioral survey, into prototypes tested inside real online shopping journeys.
01
Frame the system
Connect ecology to everyday financial behavior.
The research began with literature on planetary health, sustainable consumption, financial behavior, AI, and the Theory of Planned Behavior. Expert interviews helped identify where product data, incentives, regulation, and daily habits collide.
Literature review + 7 expert interviews
02
Map the people
Study attitudes before inventing the assistant.
A 63-question survey was distributed across Taiwan. It covered shopping, saving, investing, environmental concern, trust, AI, future plans, and feature preferences. The work moved beyond average answers by clustering 36 attitude fields into three behavioral personas.
Put the intervention inside a real shopping decision.
The Green Filter Chrome extension was tested on Momo product pages, where students were already looking at price, images, reviews, specifications, and discounts. A linked conversational prototype let them ask deeper questions about products and companies.
32 interviews + 100+ anonymous self-tests
The published research archive documents the process, methods, results, testing notes, and limitations in public. Image source: Green Filter research methodology.
01
Shop
Understand the product, company, materials, origin, and alternatives.
02
Save
See how repeated choices shape money, resources, risk, and personal impact.
03
Invest
Recognize spending as support for the companies and systems behind the product.
The product thesis
Shopping can become the first lesson in investing.
Green Filter calls this “Shopping-as-Investing.” The phrase reframes a purchase as more than consumption. Money becomes a vote of confidence in a company, a production system, a supply chain, and a future cost structure. The product begins with a familiar object, then widens the user’s view from price today to value and consequences over time.
Behavioral segmentation
One sustainability message would fail three different people.
K-means clustering across 36 attitude questions produced three broad personas. The clusters did not divide people into good and bad consumers. They exposed different trade-offs, motivations, and levels of perceived control that the interface must respect.
Values-led
n=278
Eco-Friendly
More willing to act or pay for environmental goals, but still needs credible evidence and usable alternatives.
Product implication: Offer depth, provenance, and the ability to compare impact without hiding practical product quality.
Undecided
n=356
Moderate
Balances competing concerns and may care about sustainability without making it the dominant purchase criterion.
Product implication: Make the better choice easy to understand, easy to compare, and compatible with price and convenience.
Cost-first
n=242
Frugal
Protects affordability and is cautious about paying a green premium, even when environmental concern exists.
Product implication: Lead with total value, durability, safety, savings, and lower-impact alternatives at a realistic price.
The personas were derived from response patterns rather than invented demographic profiles. Image source: Green Filter survey results.
What students asked for
Transparency becomes useful when it changes the next click.
The strongest preferences were concrete and product-level. Students wanted to avoid harmful products, understand origin and production, and receive feedback about their own behavior. Social feeds, generic news, and abstract carbon tracking ranked lower.
63%
Avoid the most polluting products
The strongest requested feature was a product-level filter that makes harmful options easier to identify and avoid.
41%
See where a product comes from
Origin mattered because it connects transport, local production, accountability, and perceived trust.
40%
Understand how it was produced
Students wanted production information translated into a simple signal, rather than another report to decode.
25%
Receive a monthly eco-score
A personal summary could turn isolated purchases into feedback about longer-term spending behavior.
One project, four surfaces
The research, recruitment, intervention, and conversation became one system.
Each surface answered a different question. The public archive made the evidence inspectable. Ziran supported recruitment and local communication. The extension reached the shopping moment. The AI prototype allowed deeper exploration after the first signal.
01
greenfilter.app
Public research archive
The complete thesis became a navigable public website rather than a PDF hidden in a university repository. Methods, charts, quotes, testing notes, limitations, and conclusions remain open for inspection.
A Traditional Chinese entry point explained the research, collected anonymous student responses, and supported field recruitment across Taiwanese universities.
The extension inserted product analysis into Momo, close to a real purchase decision. It surfaced sustainability signals, company context, and alternative products without requiring the user to start from a separate research tool.
The web prototype allowed deeper questions after the first signal: ingredients, materials, carbon, labor, company behavior, alternatives, and the relationship between spending and investing.
Testing happened on participants’ own devices and connections, turning hardware, browser, and network constraints into product evidence. Image source: Green Filter testing archive.
Interaction design decisions
The AI mattered less than where, when, and how its judgment appeared.
The design challenge was not to generate more information. It was to fit trustworthy evidence into an attention environment already dominated by price, images, reviews, specifications, and promotional pressure.
01
Entry point
Meet the user inside the purchase, before attention disappears.
The research showed that students rarely begin shopping by searching for an ESG report. They begin with a product, a price, a picture, a need, or a discount. Green Filter therefore entered the existing commerce flow and treated sustainability as decision support, rather than a separate educational destination.
Place the signal near price, variants, and Add to Cart
Make the first insight understandable in one glance
Keep deeper evidence available without forcing it first
02
Information hierarchy
Translate evidence into a signal, then let people inspect the proof.
Raw carbon numbers, ESG categories, certifications, materials, factory claims, and controversies create cognitive overload. The interface needs a clear top layer, a short explanation of why, and visible sources underneath. Trust comes from being able to move between summary and evidence.
Use plain-language red, yellow, and green states
Explain what changed the score
Keep certifications and third-party sources attached
03
Actionability
A warning without an alternative only creates guilt.
Participants responded most strongly when the prototype suggested a comparable product or brand. The alternative still had to satisfy price, quality, size, safety, and availability. The design shifted from moral judgment toward practical agency.
Keep alternatives on the same page
Compare price and core specifications alongside impact
Use one persistent next action instead of a long report
04
AI role
Let AI do the synthesis without making chat the entire product.
The useful role for AI was to combine fragmented product, company, and environmental information, then explain it in language a student could use. Testing showed that few people wanted to begin with an open chat box. The interface should answer the obvious question first and make conversation an optional path for curiosity or uncertainty.
Generate a concise first analysis automatically
Offer suggested questions that match the current product
Show uncertainty and source quality instead of false certainty
05
Motivation
Health, safety, price, and durability can open the door to sustainability.
Food safety and personal risk were more immediate than abstract environmental language. That does not weaken the ecological purpose. It reveals how people understand consequences. Product health, worker treatment, origin, durability, and pollution can be presented as one connected quality story.
Use concrete human consequences before abstract categories
Connect lower impact with product quality and long-term value
Avoid assuming that every user arrives with the same motivation
06
Resilience
Design for the device and network people actually have.
Slow Wi-Fi, older laptops, browser restrictions, and dying batteries became research findings. A sustainability tool that requires unusually good infrastructure excludes the people it claims to help. The next product direction is lighter, mobile-first, and more tightly integrated with familiar shopping and payment tools.
Load the decision signal before secondary media
Design around smartphone use and platform restrictions
Preserve a useful fallback when AI or external data fails
What testing changed
A feature can be visible in code and invisible in behavior.
The sessions exposed gaps between the intended product and the experienced product. Some problems were conceptual, such as ESG jargon. Others were brutally ordinary: placement, loading time, unfamiliar navigation, platform restrictions, and an overlay that looked like an ad.
Observed
Price, photos, reviews, and specifications took attention first.
What it meant
Sustainability cannot expect a separate attention budget. It must work with the hierarchy of commerce.
Design response
Move the signal beside the price and connect greener choices to cost, quality, and availability.
Observed
The extension was sometimes missed or mistaken for an advertisement.
What it meant
An intervention can technically exist on the page while remaining behaviorally invisible.
Design response
Clarify ownership, use a recognizable entry cue, and make the first benefit visible without an extra click.
Observed
ESG labels were recognized but poorly understood.
What it meant
Institutional language can signal importance while failing to support a decision.
Design response
Replace jargon with a plain statement, a visual level, and a short explanation of the evidence.
Observed
Concrete alternatives were the most actionable feature.
What it meant
People can act when the product preserves choice instead of ending with a warning.
Design response
Keep one-click alternatives persistent and compare the practical attributes that already matter.
Observed
Carbon, labor, ingredients, and disposal information created surprise.
What it meant
Hidden product histories can change how familiar objects are understood.
Design response
Reveal one material fact at a time, then allow the user to inspect the full lifecycle and sources.
Outcome and limits
The outcome is a tested product direction and a public body of evidence.
Green Filter produced a working extension, an AI prototype, a large survey dataset, behavioral personas, documented usability findings, and a public research archive. It does not claim that one prototype solved sustainable consumption. It clarifies where a useful product can intervene and what must be solved next.
What the project established
Young adults do not need more guilt. They need better leverage.
Students already understood that money has environmental consequences. The missing layer was practical control: trustworthy signals, comparisons, alternatives, and an interface that arrives before the purchase is complete. The project reframed sustainability from a moral lecture into product intelligence that can support agency.
What remains open
The hard work is data quality, integration, and trust at scale.
A production system still needs reliable product passports, source provenance, data freshness, uncertainty handling, platform partnerships, mobile integration, and governance against greenwashing. These are product and institutional problems, not details an interface can hide.
I carried the question from campus fieldwork into a working interface.
Green Filter was Kris Haamer’s MA Interaction Design thesis at National Cheng Kung University in Tainan. The role covered research framing, literature review, expert interviews, survey design, on-campus recruitment, quantitative and qualitative analysis, clustering and personas, product strategy, interaction design, extension and prototype development, moderated testing, iteration, data visualization, and publication of the research archive.
