VITON13 Research
Independent experiments, data and analysis on AI, search, design, commerce and the future of the web.
13 protocols are now in execution. Study 01 has frozen its 30-day evidence window and is awaiting independent editorial review; every unfinished result remains noindex until its review gate passes.
Publication rule
Evidence before volume.
Thirty URLs are not the goal. A study becomes indexable only after its method, first-party evidence, citations, limitations and quality review agree.
Protocols that still need internal data stay visibly labelled; experiments without results are never written as completed work.
Flagship report
The State of AI Search 2026
The State of AI Search 2026: How People Discover Information After Traditional Search
A 20-record evidence map of adoption, major systems, behavior, citations, publisher traffic and conversational commerce — with every denominator and limitation kept visible.
Read the annual report →AI Search
Crawling, indexing, citations and the evidence behind AI discovery.
We Tracked AI Crawlers for 30 Days: Here’s What Actually Visited Our Website
A reproducible, verification-aware VITON13 log study that separates claimed user agents from operator-verified crawler traffic.
Complete independent aggregate review and sign the publication decision.
We Launched a New Website From Zero: What Google Indexed in the First 30 Days
A day-by-day VITON13 timeline joining Search Console, sitemap discovery and verified crawler activity without inferring indexing from crawl logs.
Provide authenticated Search Console URL Inspection and Page Indexing exports for a frozen launch cohort.
ChatGPT vs Gemini vs Google AI Mode: Which One Finds New Websites Faster?
A timestamped, repeatable discovery test using new pages and a pre-registered winner criterion.
Acquire six eligible unused domains, publish 24 matched pages and start the registered 30-day observation schedule.
We Asked ChatGPT, Gemini and Google the Same 100 Questions — Here’s How Their Sources Differ
A public 100-query benchmark, normalized source dataset and transparent coding rubric.
Capture three fresh-session runs in each named consumer product and obtain two independent source labels.
The State of AI Search 2026: How People Discover Information After Traditional Search
An annually updated evidence map that distinguishes product capabilities, measured behavior and VITON13 analysis.
How to Get Your Website Cited by ChatGPT in 2026: What We Learned From Real Tests
Controlled tests reported as observations with confidence levels, negative results and unresolved questions.
Run the registered ChatGPT Search schedule with two independent labels per valid capture.
Google AI Overviews vs AI Mode vs Traditional Search: What Website Owners Need to Know in 2026
A source-first comparison that maps official product behavior to publisher decisions without promising traffic outcomes.
AI Search Optimization in 2026: What Actually Matters — and What Is Just Hype
A verdict matrix with evidence, counter-evidence and confidence for every disputed optimization tactic.
AI Development
Benchmarks, architecture, cost and bounded website agents.
We Built the Same Website With ChatGPT, Claude and Gemini — Here Are the Results
One frozen brief, identical environments, public scoring criteria and retained outputs for all models.
Freeze one implementation brief and obtain current ChatGPT, Claude and Gemini execution access under comparable plans.
Best AI Models for Web Development in 2026 — Tested on Real Projects
A non-affiliate task suite with weighted scores, failure cases and developer-intervention minutes.
Lock model versions and pricing, execute identical tasks and retain every prompt, output, test and intervention minute.
How Much Does It Really Cost to Build an AI-Powered Website in 2026?
A date-stamped cost model and calculator with transparent usage assumptions across four operating scenarios.
We Built an AI Assistant for a Website: Architecture, Cost, Memory and Mistakes
A redacted production architecture, measured latency and cost envelopes, threat boundaries and documented mistakes.
What Happens When You Let an AI Agent Control an Entire Website?
A sandboxed capability test with an explicit permission matrix, audit trail and no destructive access.
Clone a non-production sandbox, freeze the action matrix and execute only reversible tasks with approval and audit logs.
Future Web
What changes when software becomes a website’s reader and operator.
The Agentic Web Is Coming: How Websites Change When AI Becomes the User
A transition model linking human UX, machine-readable interfaces, identity, permissions and payments.
Do Websites Still Matter in the Age of AI Agents?
A two-sided argument that separates distribution, ownership, transaction and brand-experience functions.
Design
Measured interface quality, performance and premium perception.
We Analyzed 100 Premium Websites: What Actually Makes a Website Feel Expensive?
A coded 100-site dataset, factor definitions and inter-rater review rather than a taste-only list.
Freeze a defensible 100-site sample frame, archive every page and complete blind dual coding.
Apple-Like Web Design: 13 Principles You Can Use Without Copying Apple
A principle-to-implementation matrix grounded in official accessibility and platform guidance, with anti-copying constraints.
Why Most AI-Generated Websites Look the Same — and How to Make Yours Different
A repeatable visual-pattern audit and a divergence checklist tested against generated examples.
Web Design Trends 2027: What Will Survive — and What Will Disappear
A falsifiable forecast with confidence levels, observable 2026 signals and an annual scorecard.
Beautiful vs Fast: We Tested Whether Heavy Animation Actually Hurts a Website
Matched A/B builds measured on the same devices, networks and task flow, with raw Lighthouse and trace artifacts.
Build content-identical animation-on/off variants and collect repeated lab traces on the same device matrix.
We Removed Half the Interface From a Website — Here’s What Happened
A pre/post task inventory with test scripts, time-to-action and qualitative failure analysis.
Freeze full and reduced variants, recruit consented participants and run counterbalanced task sessions.
The Psychology of Premium Websites: Why Some Interfaces Feel Expensive
A strict split between peer-reviewed evidence, design interpretation and labelled VITON13 observation.
Agentic Commerce
Discovery, store data, payments and natural-language buying.
Agentic Commerce Explained: How AI Will Buy Products for You
A complete intent-to-fulfilment system map with trust, payment, fraud and merchant-data failure points.
We Asked AI to Find the Best Product Online — Here’s What It Chose and Why
A repeated, timestamped multi-system product test that records citations, availability and recommendation stability.
Freeze product categories and systems, then run repeated timestamped sessions with same-day availability verification.
How to Make Your Online Store Understandable to AI Shopping Agents
A machine-readiness checklist verified against visible HTML, structured data, feeds and accessible interactions.
Traditional Ecommerce vs Agentic Commerce: What Changes for Brands?
A stage-by-stage operating-model comparison covering data ownership, attribution and support after purchase.
We Built a Marketplace Where Customers Describe What They Want Instead of Searching for It
A non-promotional VIT MARKET case study with interface decisions, matching boundaries and documented product lessons.
Can AI Replace Search Filters in Ecommerce? We Built a Prototype to Find Out
A shared product corpus, matched user tasks and relevance rubric comparing filters with natural-language retrieval.
Freeze one product corpus and build filter and natural-language paths over exactly the same inventory before participant testing.
VITON13
Candid product, systems and build-in-public case studies.
How a Small Brand Can Build a Global Digital Ecosystem in 2026
A candid VITON13 system map separating shipped infrastructure, planned modules, trade-offs and operating constraints.
Building VITON13 in Public: Technology, Experiments, Failures and What We Learn
A dated, evidence-linked changelog of shipped systems, failed experiments, design decisions and next tests.