Skip to main content

Methodology

·858 words·5 mins
💬 Discussion ↓
Author
Maya Sasson
Editor of Asians in Israel. Writes about the Asian diaspora communities in Israel — Thai, Filipino, Chinese, Japanese, Korean, Vietnamese, Nepali — their workplaces, restaurants, embassies, and the practical mechanics of living here. Maya Sasson is the pseudonym used by the site’s editor; corrections and editorial correspondence go to editor@asiansinisrael.com.
Table of Contents

This page documents the practical mechanics behind the site — how a business gets into the directory, how a cluster guide is built, what the “last updated” badge actually means, and where AI sits in the workflow.

The directory
#

The business directory at /directory/ is the spine of the site. Every entry is structured data in data/businesses.yaml plus a per-entry markdown bundle in content/directory/{slug}/.

Sourcing. An entry enters the directory through one of four routes:

  1. Editor scouting — restaurants, shops and service providers found in person, through community channels, or by following Asian-cuisine and immigrant-business signals on Wolt, Google Maps, Instagram and local press.
  2. Community submission — readers and operators flag businesses we have missed. Submissions arrive via the contact page, the Telegram bot (@aainewbot), or the community forum.
  3. Reconciliation with public sources — we cross-check Wolt, Google Maps, official business pages and Instagram to confirm an operator is currently active before listing.
  4. Phone verification — for businesses where the public footprint is thin or inconsistent, we call before listing.

Verification status. Each entry carries a verification_status field — verified, tentative, or unverified — set according to how thoroughly we have confirmed the business is still operating at the address and phone number on file. Entries that fail verification (number disconnected, address vacant, Wolt entry removed) are either updated or unpublished; the directory will not knowingly carry pages for businesses that have closed.

last_reviewed. Each entry also carries a last_reviewed date. This date moves forward only when the editor has re-checked the underlying facts — still open, still at this address, hours and prices still roughly current. A last_reviewed badge renders on the public page so readers can judge how fresh the information is. Entries that go un-reviewed for too long get re-queued for a fresh pass; if a re-pass confirms a closure, the entry comes down.

What we do not list. Businesses with no connection to an Asian community in Israel; “Asian-fusion” venues run with no Asian community involvement or sourcing; businesses we cannot reach for verification after three attempts.

Cluster guides
#

Cluster guides (“Best Korean Restaurants in Israel”, “Asian Supermarkets City-by-City”) are roundup pages that funnel readers into the directory entries. Each guide pulls together a curated set of named businesses with a paragraph or two on each; every named place in a guide links to its full directory entry.

Selection criteria. A business gets named in a guide if it (a) has a directory entry that meets the verification standard above, and (b) earns inclusion on cuisine merit — the editor has eaten there, or the community-input signal is strong enough that the editor is confident naming it. Guides are not paid placements; the order is editorial and may shift between refreshes.

Refresh cadence. Guides are year-stamped ("…2026") and carry the same last_reviewed discipline as directory entries. We aim to refresh each guide at least once per year, sooner if a meaningful opening or closure shifts the landscape.

News, events, jobs
#

News posts are gated by the editorial-policy local_angle rule — see the editorial policy for what counts. Events are posted from organiser sources and verified against the source URL before publication. Jobs come from the hiring organisation or a verified community-member referrer.

We do not auto-syndicate from external feeds, and we do not republish other outlets’ stories without adding the local angle that justifies the post existing.

Languages
#

The site is bilingual by default: every article ships in English and Hebrew. The cultural-specific locales — Japanese, Korean, Simplified Chinese, Traditional Chinese — are added only when the content is specifically of that community. Korean embassy news goes into Korean; restaurant reviews of a Korean restaurant go into Korean. An English-news rewrite about Korea in general does not. Thai and Vietnamese locales were retracted on 2026-05-13 because the quality of community-specific content in those languages did not meet the bar.

Hebrew translation is the editor’s responsibility on every piece. The mechanics: an LLM produces a first-pass translation; the editor reviews, corrects and ships. The reverse direction (HE→EN) also happens occasionally for community-submitted pieces.

AI in the workflow
#

Where AI sits in this site is narrow and explicit: language drafting. An LLM helps produce the first-pass Hebrew translation, occasionally helps draft a section of English prose around a research-heavy topic, and helps audit copy for unidiomatic phrasing.

Where AI does not sit: choosing what to cover, judging whether a source is reliable, deciding whether a business deserves a write-up, ranking restaurants within a guide, deciding whether to retract a piece after a complaint, or interacting with readers and contributors. Those are editorial calls made by the editor.

This split is deliberate. Routing, tool selection, and judgement stay in human hands; language understanding is delegated to the model. If a piece on this site reads like it was written by a model end-to-end, the editor has failed at the job — and a correction request is the right response.

Errors and updates
#

When we get it wrong — wrong fact, wrong attribution, outdated information — the editorial policy sets out how to flag the issue and what we do in response.


Join the conversation

What do you think? Share your thoughts with the community