Answer-engine measurement
How AI assistants select, cite and rank sources—and how to measure it fairly.
Danhoangda is an independent AI lab in Hanoi. We build HowAI Suite—four systems that observe, govern, correct and act on what AI says about a company—Metal Mantra, an open evidence registry for AI agents, Dan Writer, a home for letters written by people, and Drint, a browser studio that takes packaging and print from die line to press.
00Why the lab exists
AI is becoming the interface between companies and the world. We build the instruments that make it observable, accountable and correctable—so every answer shows its evidence before anything acts on it.
01Lab console
Pick a buyer question and watch how HowAI Suite handles it—from the observed answer to an action a person can approve.
Which AI visibility platform should a B2B SaaS team choose?
Illustrative run with sample data for a fictional company. Not a live measurement.
Evidence before autonomy · Preserve the observed answer · Provenance before conclusions · Limits, made visible · People approve every spend · Recheck until it changes
02Systems
Each product has one hard job. HowAI Suite lets a company observe, govern, correct and act across the AI layer; Metal Mantra shows what AI agents reveal before you trust them; Dan Writer keeps the human voice in what people send each other; Drint makes what is designed on screen what comes off the press.
See who AI recommends—and win the answer back.
Measure real buyer questions across ChatGPT, Claude, Perplexity and Gemini. See the answer, the sources and the move most likely to change it.
Turn AI governance into living evidence.
One owned inventory of AI systems. Prioritize risk, close control gaps and keep a defensible history of every decision.
Find the false claim. Prove it changed.
Catch wrong prices, missing features and outdated facts, trace them to the source and recheck until the AI answer is accurate.
Pay only where the AI answer leaves you out.
An ad agent for AI chatbots that checks organic visibility first, works inside approved caps and keeps every spend change reviewable.
Trust the evidence, not the pitch.
An open registry that scans an AI agent's public source with a published five-pillar method and shows the findings, the scan scope and what stays unknown. Automated signals, not certification.
Letters written by people.
Write a letter on digital paper, sign it, seal it until a day you choose and share it by link. No AI ghostwriting, no ads and no feed: every letter is in its writer's own words.
From die line to finished print.
Design boxes, bags, labels, posters and menus on the real die at the real size. The box folds in 3D as you draw, clients approve from one link and the file goes to press as PDF/X-4.
03Inside the products
Every product is running in production at its own address. These are the current product pages.

See who AI recommends—and win the answer back.
Open HowAICite04Method
One rule runs through everything we build: keep the evidence before you draw a conclusion, and before any system acts.
Ask the real question, the way a buyer would, across the engines that answer it.
Keep the answer, its sources and the moment it was seen. Conclusions come later.
Rank what matters: the gap, the risk or the false claim most likely to cost you.
Fix the source, correct the claim or buy reach—only inside limits a person approved.
Measure again. A change counts only when the answer itself has changed.
05Research
Alongside the products, the lab studies how AI systems choose what to say, what makes a decision about AI defensible, and how agents can act only inside limits people set.
How AI assistants select, cite and rank sources—and how to measure it fairly.
What makes an AI decision defensible months after it was made.
Tracing a false claim to its origin and proving the correction landed.
Agents that publish, spend or change things only inside limits people set.
06Lab notes
07Questions
An independent AI lab in Hanoi, Vietnam. It owns and builds HowAI Suite, Metal Mantra, Dan Writer and Drint, and runs applied research on AI answer engines, governance and agent evaluation.
Danhoangda is founded and operated by Nguyễn Thành Dân. Product, support and security questions go directly to the person responsible for the system.
HowAICite observes what AI says, HowAISafe governs the AI a company uses, HowAITrust corrects false claims and HowAIAds buys reach only where the organic answer leaves you out.
An evidence-led registry for AI agents. It scans an agent's public repository with an open, versioned five-pillar method and publishes the findings, the scan scope and the limits. Its scores are automated signals, not security certification.
A place to write a letter on digital paper, sign it, seal it until a chosen day and share it by link. There is deliberately no AI ghostwriting: every letter is in its writer's own words, and only people with the link can read it.
A browser studio for packaging and print. Enter a size and the die builds itself; design on every panel, watch the box fold in 3D, send the client a review link and export a print-ready PDF/X-4 in CMYK. It is free to start.
Email the lab. Partnerships, research collaboration and product questions are all welcome.
08Danhoangda / Hanoi, Vietnam