← Bible Study Helper · Help page · Safe work abroad · For partners · Posters

Safeguarding

How the anti-trafficking tools on Bible Study Helper protect the people who use them.

Last updated 17 October 2026.

What the tools are for

Bible Study Helper is a free Bible question site in 150 languages, with no sign-up and no name needed, used mostly in South Asia, Africa, the Middle East and Southeast Asia. Its safety tools help people check a job offer before they pay or travel, ask for help privately, turn what they witnessed into a report for a hotline, keep emergency numbers with them abroad, and find hope after exploitation. We are a front door: we point people to police, emergency services, hotlines, embassies and aid organisations. We do not investigate, rescue, or monitor anyone.

Principles

  1. Safety before everything. In danger, people are told to call local emergency services first. A Quick exit button (and the Esc key) closes the tools and clears the screen; on help pages it leaves the site.
  2. Help never depends on faith. Faith is offered, never required. The private chat deliberately does not preach.
  3. The AI never gives a phone number or a link. Every number shown comes from our own directory, and our servers remove any number or link the AI writes, so a model cannot invent a number for someone in danger.
  4. The AI is honest about what it is. It says it is an AI, not a person, a hotline or an emergency service. It never calls a job "safe" — only "no warning signs found in what you shared", with advice to check independently.
  5. Nothing to trace. No sign-up, no name, no email, no phone number. We do not store IP addresses. Help pages use no cookies, no storage and no third-party scripts.
  6. We keep as little as possible — see below.
  7. No one is named or accused. Reports and warnings describe behaviour and patterns, never identifiable people.
  8. Warnings follow fixed rules and fixed words. Nobody monitors the site, so every warning is published automatically by a written rule: a disaster warning when the international disaster alert system (GDACS) issues an orange or red alert for a country, and a scam-wave warning when at least 12 people in one country answer "yes" to the same warning sign within 7 days. The words are fixed general safety advice, translated in advance and never written by the AI; each warning expires on its own after 7 to 21 days.

What we keep

WhatKept?
Text or screenshots put into the job-offer check, private chat or witness reportNot stored. Sent to our server and to our AI provider (OpenAI) only to write the answer. OpenAI's published API terms say API data is not used to train its models and may be retained for up to 30 days for abuse monitoring. (A few anonymous facts from job-offer checks are kept: see “Early-warning signals” below.)
Everyday Bible questions that seem to describe trafficking, abuse or self-harmThe words are not stored and never sent to analytics providers. Only an anonymous count remains.
Use of the toolsAnonymous counts only: which tool, language, country, risk level and types of warning sign. Since 19 October 2026 they also say which signs were marked in the self-checks (yes, no or not sure for each question), the country a person chose as where they come from or are going to (only if they chose one), and how they reached the site (for example an ad or a shared link). No visitor or session identifier, no text, and the time is kept only to the hour, so a count cannot be matched to a visit.
The private chat and questions that describe exploitation: the subject onlySince 19 October 2026 the main subject of each private-chat message is counted (for example pay, passport or wanting to go home), with whether the person seems to be planning to go abroad or is abroad now and the kind of work, if they said it. The words are never kept. Bible questions that seem to describe exploitation get the same count (their words are never stored either). The subject is labelled by a small AI model and can be wrong.
One quick question (optional)Shown once a day under an answer to visitors in countries where many people go abroad for work: one question with answer buttons and Skip, no free text, never after a question that described danger. Only the count of each answer is kept. The phone remembers which questions it was asked, so it is not asked again.
Job offers a person sends to our automatic check (optional; before 4 November 2026: "to warn others")Only with a ticked consent box that says what happens. No person reads the offer: rules pick out the recruiter's phone numbers, links or account names in it (never the sender's, never their name). One goes on our recruiter list only when offers or job checks from people in at least two countries show the same one (3 or more, 2 of them with serious warning signs). Numbers and sites from our help directory, government sites and our own are never listed; an identifier can be taken off for good after a complaint. The list may be given to the platform that hosts the account (WhatsApp, Facebook, TikTok...) and to anti-trafficking partners under a signed agreement, so the account can be stopped. Never published. The offer is kept for up to 12 months.
Safe Journey CardMade on the person's phone. Never sent to us.
Morning verse notification (optional)Only if someone taps "Send me a verse every morning" and allows notifications: the push address their browser gives us, their language and send hour. The notification itself carries no data; the phone fetches the day's verse. Turning notifications off for the site deletes it at the next send. No name, number or IP.
Daily verse on Telegram (optional)Only if someone subscribes: their Telegram chat number, language and delivery time, so the bot can send the verse. Sending /stop — or blocking the bot — deletes it. Questions asked in Telegram are not linked to it.
Weekly safety check-ins on Telegram (optional)Only if a person going abroad turns them on. To send the weekly verse and, if needed, an alert, we keep: their Telegram chat number and their family contact's (the one person they chose, who joins by opening a link), their languages, home and destination country, check-in dates, and the employer or agency details they chose to type (encrypted, and shown only to their family contact inside an alert). Replying 2, or no reply for two weeks (or the time they chose), sends that alert; after a 2 the person sees the same thank-you as after a 1, and if no family contact could be told, they are shown the emergency number instead. Sending /home deletes everything; the family contact can leave with /stopcontact; anything unused for 400 days is deleted. It is not a rescue service, and the messages say so.
Contract X-ray, screenshots and voice messages (website, Telegram, WhatsApp)Not stored. Sent to our server and our AI provider only to write the answer (voice messages are turned into text first). Anonymous facts are kept as described under “Early-warning signals”.
Job offers in community chat groups (Telegram group guardian)Only in groups that add the bot. It reads posts to spot job offers and replies only when it finds warning signs. Posts are never stored; we keep the group's number, its language and how many checks and warnings it had.
WhatsApp (when switched on)The sender's phone number is used only to reply, never stored or logged.
Early-warning signals from checksFrom every job-offer or contract check we keep anonymous facts only: destination, job type, the pay and fee written in the offer, visa type, the app it came through, the warning signs found, and keyed hashes of any phone numbers, links or agency names in it (a hash cannot be turned back into the number) plus a fingerprint of the wording. No text and nothing about the person. This lets us see that the same recruiter is reaching people in several countries. Kept 12 months.
"What did you pay to get a job abroad?" (optional)Asked as one of the quick questions under answers (and at the end of the fake-job game): who paid (you or your family) and, since 19 October 2026, one tap for roughly how many months of pay abroad it cost (in bands) and one for the region of the job. No identifier at all; the time is kept only to the hour. (Before 19 October 2026 the survey asked for amounts, currencies and the destination country.)
"What will you do now?" after a risky result (optional)Since 19 October 2026, when a job-offer check, a contract check or an agency licence check shows warning signs, one optional question with answer buttons asks what the person will do (not pay or go, ask more questions, talk to family, still go, not sure). Only an anonymous count of the answer is kept, with the tool and the result; nothing is stored on the phone. Whatever the answer, the next safe step is offered (for someone still going: the Safe Journey Card, help numbers and weekly check-ins).
"What happened later?" and "Did you get through?" (optional, since 8 November 2026)After a risky job check, only if a person taps "Yes, ask me later", their phone keeps a small note (which check, the result, the day and the country of the job; no words, no name) so the site can ask on a later visit what happened; the answer is an anonymous count and the note is deleted. After a tap on a help number in our tools, the site may ask whether the call got through (nothing is stored). Never on help or child-protection pages. Links to partner organisations carry a neutral tag (never our name) so they can count visits we send.
Weekly check-in and weekly warning on notifications (optional, since 4 November 2026)Only if someone turns them on (the button says to do so only if no one else reads the phone): the notification address, language, send hour and what they asked for. The notification carries no message; the phone shows neutral words ("Are you safe this week?"). "I'm OK" and "I need help" are counted with no address or id. "I need help" opens the help numbers and a private talk. Never offered on help or child pages.
Warnings shown on the site (since 4 November 2026)Disaster warnings go live automatically (general safety advice in fixed words, one per country, 21 days). Wave warnings appear when at least 12 people in one country tell the 6-question job check the same warning sign in a week: fixed general words, never a name, a place below country level or an employer.
Public signals page (since 4 November 2026)Monthly totals by country and route at /signals/, from the anonymous counts. Every count under 10 is hidden and a share is shown only when both numbers are 10 or more.
Spot the fake job (game), church postersAnonymous counts only (rounds played and answers, poster scans per poster code). The phone itself remembers the last play date.
Freedom ambassadors on Telegram (optional)Only if someone signs up to forward one warning a week: their Telegram chat number and language, and the week of the last message. Sending /stop — or blocking the bot — deletes it.
Agency licence checkSearches our copy of the government's own public list of licensed recruitment agencies (Nepal, India, Sri Lanka, the Philippines), on the phone. What a person types is not sent to us; only an anonymous count of the result. We copy agency names, licence numbers, towns and licence status, never owners' names, phone numbers or e-mails, and always link to the official register, which can be newer than our copy.
Protect children tools (since 28 October 2026)Checks for parents (before a child leaves home; a marriage for someone under 18), the warning pictures and the child safety card work on the phone; we keep only anonymous counts of the signs marked and the result (the family phone number typed for the card never leaves the phone). "Check a message" sends only the pasted words to our server and our AI provider to look for grooming or sextortion; they are not stored and no person sees them; photos are never asked for or accepted. When the danger screen notices a child in danger in a question or chat message, only the kind of situation is counted. The guide pages /protect-children/ keep no id. Child helplines come from Child Helpline International's member list (unverified until called). Our child safeguarding policy and our procedure for child sexual abuse material are available to partners on request.
Comments under our own Facebook posts and ads (when switched on)Comments that look like job offers moving the talk to WhatsApp or a phone number, or that show serious warning signs, are hidden (never deleted) so no one is lured from our page. Only daily counts are kept; never the comment, the name or the account.

Help numbers

Numbers come only from our directory of 141 countries (emergency services, national anti-trafficking and child helplines where one clearly exists, official recruitment-licence registries and embassy pages). Where a language is spoken across many countries, we do not guess one country's hotline; we show the local emergency number and the Global Modern Slavery Directory instead. Each entry is marked unverified until a person confirms it: emergency numbers are checked against official government sources (never test-called); helplines are called to confirm the number, hours and languages; links are opened on a phone. Confirmed entries show a ✓; a helpline that fails the check is removed until fixed. Links are re-checked monthly.

Status: 314 entries, none confirmed by a person yet (the call sheet is ready). Until an entry is verified, the tools always add: "If a number doesn't work, call your local emergency number or use the global directory."

How we check accuracy

We test the tools against written cases before and after every change, using our own evaluation sets (kept in the code repository). Latest results: job-offer check 26 September 2026, question screening 8 October 2026.

TestResult
Job-offer check: realistic scam offers in 8 languages (fees to the worker, personal accounts, kept passports, visit visas, scam-compound jobs, sexual exploitation, child recruitment…)30 of 30 caught, all rated high risk
Job-offer check: fair, legal job offers0 of 16 wrongly flagged
Job-offer check: warning signs a careful reader would expect59 of 64 found
Question screening: people describing exploitation, abuse or self-harm in 12 languages23 of 23 caught
Question screening: ordinary questions, including questions about slavery, trafficking and suicide, writing requests, grief and money or work troubles0 of 26 flagged
Question screening on 300 real, recent questions from the site0 of 300 flagged (none described exploitation or self-harm; the first run wrongly flagged 2 questions about depression — fixed)

Each test found something we then fixed: rules that ignored "no placement fee" or "you keep your passport", a keyword net that flagged general questions about suicide or trafficking, and questions about depression receiving a message meant for exploitation (people who mention self-harm now receive their own message).

Children

Recruiting children, or asking someone to lie about their age, is always treated as a serious warning sign. Child helplines are listed where a national one clearly exists. We never ask for age or name, and never for an exact location: a tool asks only which country, where it needs that.

Automatic checks and partners

Nobody reads the offers people send: they go to an automatic check. A recruiter's phone number, link or account leaves the platform (only to the platforms themselves and to partners under a signed agreement) when people in two or more countries report or check the same one; numbers and sites from our own help directory are never listed, and a business that shows a number is its own can have it blocked for good. We are seeking partnerships with established anti-trafficking organisations and hotlines so that reports and patterns reach people who can act on them, under written data-sharing agreements.

Partners may receive totals from the anonymous counts, by country and month, never single records, with every number under 10 left out, only under a written agreement (template). The risks of these counts and how we reduce them are set out in our data protection impact assessment.

Languages

The tools are available in 150 languages. Their text was translated with AI help and checked automatically (wrong letters, numbers, placeholders, leftover English); it has not been reviewed by native speakers.

Concerns and contact

If you see a mistake — a wrong number, a harmful answer, or a translation problem — please tell us: [email protected]. We correct safety issues first. See also our Terms and Privacy Policy.