Methodology

Saigon Sentinel is a small US-based newsroom that pairs human editorial judgment with AI-assisted reporting. We’re open about that pairing. This page explains what AI does on our pipeline, what humans do, and the safeguards we use to keep coverage accurate and useful for Vietnamese-American readers.

What we publish

A few kinds of coverage, and the mix follows the news rather than a fixed quota. Increasingly our core is primary-source reports — plain-language write-ups built directly from official documents (USCIS and DHS notices, IRS guidance, Federal Register rules, state and city announcements): what changed, when it takes effect, who it affects, and what to do, with a link to the original. Alongside those we publish diaspora analysis — a short cited brief points to the original reporting and our piece is the analysis of what it means for Vietnamese Americans, with long-form deep dives on the biggest stories — explainers (“What to Know”) that walk general readers through a current story from at least three cited sources, plus data stories from public datasets with our own charts, a curated digest of links worth your time, and practical guides on immigration, taxes, healthcare, and community life.

Nearly all of these are original journalism we write ourselves — the primary reports, analysis, explainers, data stories, and guides. The digest is the sole exception: it is curation, a list of links to other outlets’ reporting with our one-line take, and we label it plainly as such.

We began guide-first and are steadily moving toward primary reporting built on government and public data — the information that most affects our readers’ lives, in official sources that are dense, hard to find, and that almost no one reads for leisure. Our job is to monitor, translate, and explain them in clear Vietnamese.

Where stories come from

Our reporting draws on two kinds of sources. For primary-source reports we read the official record directly — the Federal Register, USCIS and DHS notices, IRS guidance, and state and city announcements. For news coverage we follow established news organizations — including AP, Reuters, NPR, BBC, the Guardian, The New York Times, The Washington Post, Bloomberg, ProPublica, plus Vietnamese-language outlets (Tuổi Trẻ, VnExpress, BBC Vietnamese, RFA Vietnamese, VOA Vietnamese). We work only from each publisher’s own published reporting, and credit goes to the original reporter.

Each published article carries a “Sources” section linking back to every publisher whose reporting it draws on. If you spot a missing or wrong attribution, email us and we’ll correct it — see “Corrections” below.

What AI does — and what humans do

We use AI tools in four places on the pipeline: drafting a first pass of the Vietnamese text from the cited reporting, translating Vietnamese drafts into English, illustrating articles with stylized news graphics, and narrating the daily Vietnamese audio briefing.

Every article passes through the Editor-in-Chief before publication. The editor’s job is to confirm the story is real, the sourcing holds up, the diaspora framing isn’t fabricated, and the Vietnamese reads like Vietnamese. Articles that don’t clear that bar don’t run.

We do not pretend AI is doing original reporting. Where Saigon Sentinel adds value is in the analysis and diaspora-impact sections — original framing of a news story for a specific audience — and in the editorial gate.

Our safeguards against AI hallucination

AI text generation can invent facts. We treat that risk like a software bug: defense in depth.

1. Source grounding. Every draft is written against the published reporting we cite and link to, not from the model’s background memory. If the source doesn’t say it, the draft can’t say it.
2. No-fabrication rules in the diaspora section. When a story has no real Vietnamese-American connection, the article is required to say so explicitly rather than invent one. Pieces that fabricate a diaspora angle don’t pass the editorial gate.
3. Automated quality gates. Before publication every article is scored for originality against its sources (drafts that lean too closely on a source’s wording are rejected), checked for required citations, and spot-checked for fact grounding — concrete numbers, names, and dates must be supported by the cited material.
4. Human review. The Editor-in-Chief sets the standards and prompts behind every article, reviews each piece before publication, and corrects reader-reported errors within 24 hours. Nothing goes live until it has cleared both the automated checks above and that editorial review.

Illustrations and photos

Most articles run with a stylized AI-generated illustration — clearly a vector or paper-cutout aesthetic, never a photorealistic image meant to look like a news photo. The credit on every AI illustration reads “Illustration by Saigon Sentinel AI”.

When we have a real photo — provided by a stringer, sourced from a public-domain archive, or licensed — we mark it as such. The homepage lead slot only surfaces real photos, never AI illustrations.

Audio

The daily news briefing is voiced by our AI persona Hoài Trang, labeled as an AI host so there’s no ambiguity. Individual articles are read by a rotating selection of AI voices. Every article is available as audio — just press play.

Corrections

If you find an error in a Saigon Sentinel article — factual, attributional, or translational — email [email protected] and we’ll respond within 24 hours. Substantive corrections are appended in-place with a dated correction notice; minor fixes (typos, translation polish) are made silently.

More

See our editorial policy for the full standards, our newsroom page for who works on the publication, and about for our broader mission.