Green
Help with limited tasks
Let AI summarise a text you provide yourself, explain a concept or draw up discussion questions.
AI stands for Artificial Intelligence
Read with attention.
Ask with insight.
Use AI tools such as ChatGPT, Claude, Gemini and Copilot as a study aid, without handing your own reading, judgement and faith over to a machine.
You don’t need to register. Your choices stay on your own device.
How it works
Set up just once.
Ask every time.
Your project instructions tell AI how you read the Bible and how it should help. You set them up just once. Inside the project, AI reads them automatically with every question, so you don’t need to repeat them and your questions can stay short and specific.
Does your tool not have projects? Then paste the instructions as the first message of every new chat.
Just once
My Bible study project
Project instructionsTranslation, reading principles, way of working
Just once- + your instructionsHelp me read Mark 4:1–20 attentively.
- + your instructionsWhat does “justified” mean in Romans 5:1?
- + your instructionsGive three discussion questions on Psalm 23.
A traffic light
Where does AI help, and where does it stop?
Amber
Check before you trust it
Verses, quotations, history and the original languages: open the source yourself.
Red
Life decisions stay yours
AI is not your spiritual authority or decision-maker. Big matters you discuss with people.
Our starting point
Scripture leads.
The tool serves.
In the Reformation heritage of the authority of Scripture, without a church label.
Scripture above every voice
An AI answer is an aid to test, not a new authority.
2 Timothy 3:16–17 ↗Examine whether it is so
Test everything; hold on to what is good.
Acts 17:11 ↗ 1 Thessalonians 5:21 ↗Created in God’s image
We remain responsible; we do not hand our judgement over to a machine.
Genesis 1:27 ↗ Romans 14:12 ↗To know God himself
AI does not replace your relationship with Jesus. Salvation is by grace, through faith.
John 17:3 ↗ Ephesians 2:8–9 ↗Reading principles
Get to know the lenses.
Eight big questions. 72 named principles. A framework you put together yourself.
The concepts do different work: some settle authority, others help you read, and others form a larger theological lens. Read each explanation before you choose it.
Where would you like to begin?
Choose a branch to open itThe historical roots: before, during and after the Reformation
Early starting points
Rules of faith, the rule of love and different layers of meaning already played a role among early and medieval interpreters.
Reformation renewal
Luther, Melanchthon, Calvin and Bullinger emphasise, among other things, the authority of Scripture, the original languages, the intention of the text and the relationship between faith and reading.
Further formulation
Confessions state principles such as sufficiency, clarity and comparing Scripture with Scripture more systematically. The Reformation was not one uniform method.
Methods take new forms
The historical-grammatical label, modern language study and inductive ways of studying develop further. Not every present-day name is a sixteenth-century slogan.
The five solae are a useful summary; their ideas and the later grouping of the slogans have distinguishable histories. Each principle’s notes give the relevant context and sources.
Project instructions · just once
Your study. Your framework.
Write down just once how you read the Bible and how AI should help. Paste it into a project, and it applies to every question you ask there. You don’t need to repeat it with every question.
01
What do you want to work with?
This applies to every chat in the project. So keep it general, unless the whole project is about one book or theme.
02
Choose your reading principles
No theological lens has been chosen for you in advance.
Start with six reading skills
Context, literary form, word usage, historical-grammatical interpretation, observation–interpretation–application, and testing conclusions.
03
Add your own preferences
Anything else AI should know about your study.
Use it
Ask a good question.
Read for yourself.
Just once: a project with your instructions. After that, with every question: read the passage, ask good questions and test the answers. Here are example questions that can help in everyday situations.
Just once
Paste your instructions into a project
Copy the text from Build your instructions. Create a project for your Bible study and paste it into the project’s instructions. From now on AI reads it with every chat in that project: you don’t paste it again. The exact place differs between tools: see Privacy and memory for ChatGPT, Claude, Gemini and Copilot. No projects? Paste the instructions as the first message of every new chat.
Every time
First read the passage
Read it in your Bible, together with the surrounding paragraphs. Write down what you notice, what surprises you and what you are unsure about.
Start a new chat in the project
Open your project and start a new chat there. Name the passage and what you want to know. You don’t need to repeat your translation and reading principles: they are already in the instructions. Use the opening question alongside and adapt it.
Test the answer against the text
Check quotations and sources yourself. Ask: “Which words in the passage support this conclusion?” and “What is a reasonable alternative interpretation?” AI can make convincing mistakes. Test and check shows how.
Sometimes
Refine your instructions
Do you repeat something in every question? Then it belongs in your instructions. Update them in Build your instructions and replace the old text in your project. A change here does not automatically update your AI project.
Where does it belong?
What always applies,
and what applies only today.
Your instructions carry the fixed things. Your question only needs to say what is new today.
Just once · in the instructions
- Your Bible translation
- Your chosen reading principles
- How AI should help: one step at a time, or an overview first
- At what level it should explain
- Fixed rules, such as “name the verses” and “do not invent sources”
Every time · in the question
- Today’s passage or theme
- What you want to know
- Who it is for, and why
- What the answer should look like
- Which text AI may use
Everyday questions, asked well
What would you like to do today?
Choose a situation. Each example shows the question, why it is put that way, and what you should check yourself afterwards.
Build your own question
Five things make an
answer checkable.
A question like “Tell me everything about Romans” leaves AI guessing what you want. Rather, answer five short questions. Your question is put together below as ordinary sentences, with numbers that show which part does which job.
A clear question helps, but does not guarantee a correct answer. The test remains the text in front of you.
Your question
Short glossary
Big words, clear meanings.
Before you begin
Your chats.
Your choices.
Three separate settings decide what happens to your chats. Know the difference before you begin a Bible study.
Model training
May they learn from my chats?
Decides whether the company may use your new chats to improve future models. Switching it off deletes nothing and does not change what AI remembers about you.
Memory
What does AI remember about me?
Decides whether AI uses details from earlier chats in new answers. Memory changes how AI answers you, including the assumptions it makes about you.
History
Is the chat stored?
Decides whether a chat stays in your list. Deleting a chat is not the same as switching training off, and it does not always delete a stored memory either.
These three are separate switches. If you switch one off, the others do not change.
What I don’t enter
Protect other people
and yourself.
Use public or fictional examples. A setting that switches training off does not yet make it safe to share confidential information.
- Pastoral and confidential matters of others. Conversations with church members, counselling details and prayer requests stay with the people they were meant for. Merely removing a name does not necessarily make someone unrecognisable.
- Your own sensitive information. Health, finances, passwords, identity numbers and addresses do not belong in a Bible study chat.
- Children’s information. Don’t enter names, photos or details of children.
- Material you may not share. Upload only texts you may share. Ask permission first before you upload a congregation’s internal documents or someone else’s unpublished work.
- Feedback on private chats. With some tools, a thumbs up or down can make the whole chat available for training, even when training is switched off.
Step by step
Where are the settings?
Checked against the official help pages on 20 October 2026. Button names are given as they appear in the apps. Settings change often and can differ by plan, country and device.
ChatGPT OpenAI
- Model training
Settings → Data controls → Improve the model for everyone → off → Done
New chats are then not used to train models. Your chats stay in your history. On a phone: open the sidebar, choose your profile icon, then Data controls. If you rate an answer with a thumbs up or down, the whole chat can still be used for training.
- Temporary chat
Temporary chat, at the top of a new chat
Does not appear in your history, creates or changes no memory and is not used for training. OpenAI may keep it for up to 30 days for safety purposes. Choose Unpersonalized before you begin if the chat should not use your memory and instructions. If you save it, it becomes an ordinary chat.
- Memory
Settings → Personalization → Memory
Depending on your account you will see Memory, Reference saved memories, Reference chat history or a Memory summary. You can also ask: “What information about me is stored in memory?” If you delete a chat, a separate saved memory is not necessarily deleted too: remove both.
- Project
New project → ••• → Project settings
Paste your instructions here. Inside the project they apply instead of your general custom instructions. Under Memory you choose Default memory or Project-only memory; the latter keeps the project’s context inside the project. Shared projects always use project memory.
- Family and account
Settings → Parental controls · Settings → Security
Parents can link a teenager’s account and manage certain features, including memory. Use a passkey or two-step verification (MFA) for your own account.
Official: Data controls ↗ Memory ↗ Projects ↗
Claude Anthropic
- Model training
Settings → Privacy → Help improve our AI models → off
New chats are then not used for future training. Data already in a training run stays there. Chats flagged by safety systems may still be used for safety work. On the phone: the same path under Settings.
- Incognito chat
Ghost icon at the top right of a new chat outside a project
The chat is not saved in your history, not added to memory and not used when Claude searches past chats.
- Memory
Settings → Memory → Generate memory from chats
Choose Pause memory to stop memory temporarily or Reset memory to delete everything permanently. Searching past chats (“Search and reference chats”) has its own switch. For one chat without memory: click “+” in the message field and switch Memory off before your first message.
- Project
Projects → your project → Memory
Each project has its own memory. Switch off “Use account memory” if your Bible study project should stay completely separate from your other chats.
Official: Model improvement ↗ Chat search and memory ↗
Gemini Google
- Activity and training
Gemini Apps Activity → Keep Activity
If Keep Activity is on, your chats are stored and used to improve Google’s services, including training models, with the help of human reviewers. They are deleted automatically after 18 months; you can choose 3 or 36 months. If it is off, future chats are kept with your account for 72 hours and not used for model improvement, unless you give feedback.
- Human review
A share of chats is read by trained people. These chats are disconnected from your account and kept for up to three years. So enter nothing you would not show a reviewer.
- Temporary chat
Temporary chat
Does not appear in your recent chats or activity and is not used for personalisation or training. It is kept for up to 72 hours to answer you.
- Memory
Settings → Memory · Instructions for Gemini
With Memory on, Gemini can use past chats to personalise answers. To keep one chat out of it, delete that chat. Instructions you save for Gemini (in some regions “Saved info”) stay until you remove them.
Official: Gemini Apps Privacy Hub ↗
Copilot Microsoft
Microsoft released a new Copilot app on 18 August 2026. The settings differ between the new and the older version.
- Memory (new app)
Settings → Personalization → Saved memories
Switch it off to stop Copilot saving new memories. Existing memories are then not deleted: choose Manage → Delete all memories.
- Shared experience
Settings → Personalization → One shared experience
Decides whether your Copilot chats and your use of Bing, Edge and MSN personalise each other. Personalised ads have their own switch: Allow ads personalization.
- Model training
Older app: profile → Privacy → Training on conversation activity
Also switch off Training on voice conversations. Microsoft’s page for the new app does not mention a separate training switch; check your app’s privacy settings and Microsoft’s privacy dashboard.
- Instructions
Paste your project instructions as the first message of a new chat if your version has no place for standing instructions.
Official: Privacy controls, new app ↗ Older app ↗
Using a different tool? Look for the same three things in its settings: model training, memory and history. If you cannot find them, treat every chat as possibly stored and read.
Memory and your Bible study
Memory is convenient.
It can also flatter.
An AI that remembers what you believe can tend to tell you what you want to hear. That makes memory a matter of careful study, not only of privacy.
Keep your study in a project
Instructions, sources and chats stay together. Where possible, keep the project’s memory separate from your other chats.
Test without memory
Also ask an important question in a temporary or unpersonalised chat. Is the answer different? Ask why.
Look at what is remembered
Ask regularly: “What information about me is stored in memory?” Correct or remove what is wrong or does not belong there.
Instructions are not memory
You write and read your project instructions yourself. Memory is put together by the system and is not always fully visible.
My checklist
Is my tool ready?
Your ticks are stored only on this device.
And this website?
Bible Machine sends nothing to a server, AI service or analytics service and uses no tracking cookies. Your study choices and checklists are stored only on your own device. “Start over” in the builder deletes your study choices, and “Clear the list” deletes a checklist. Links to external sources open those websites, with their own policies.
Test and check
Test the answer.
Open the source.
AI writes convincingly, even when it is wrong. Learn to recognise the three most common mistakes, ask good follow-up questions and check every kind of reference yourself. How to put a good question is in Use it.
The people of Berea examined the Scriptures every day to see whether what they heard was so (Acts 17:11). The same attitude fits every answer a language model gives.
Three mistakes to recognise
Try it yourself.
Copy a test question into your AI tool and compare the answer with the text.
01 · Hallucination
A fluent answer can invent facts or sources.
The model predicts likely words. It can create a verse, quotation, book or date that sounds right but does not exist. How often this happens varies by model and task.
Test question
Give the exact text of Matthew 42:7 and explain it.
What a good answer does: it says that Matthew has only 28 chapters. Also try: “How old was Methuselah when he died?” and open Genesis 5:27 yourself: 969 years. A correct answer today does not mean the model will never invent something.
02 · Flattery
The answer confirms your assumption rather than testing it.
AI is trained to be helpful and agreeable. As a result it can tend to agree with the way you put the question, especially if it remembers what you believe. Researchers call this sycophancy.
Test question
I think Romans 3:28 means that good works are unnecessary for Christians. Is that right?
Follow-up
Don’t just agree with me. Read Romans 3:21–31 together with James 2:14–26 and Ephesians 2:8–10. Show what my statement overlooks.
Why this example: the relationship between faith and works was a core question of the Reformation. A good answer lets each passage’s own argument speak and does not resolve the tension with a slogan.
03 · Bias
The question, the training or the product steers the answer.
Training material, hidden instructions and your own wording can push the answer in one direction. A loaded question draws out a loaded answer.
Loaded question
Prove from Romans 5 that peace with God means a believer no longer suffers.
Neutral question
What does Romans 5:1–5 say about peace with God and about suffering? Distinguish the text from the interpretation.
Compare: verses 3–5 speak precisely about suffering, perseverance and hope. Does the first answer correct the question’s assumption, or does it follow it?
Follow-up questions
Questions that test the answer.
Which exact verse or section supports each point? Show how the conclusion follows from the context.
What does the text say explicitly, what is inferred from it, and what is application?
Are there any assumptions in my question that could influence the AI answer?
Check Scripture carefully for any passages that contradict this interpretation. Give an alternative grounded in Scripture.
Which claims in the answer could not be checked? Mark them clearly.
For each source, give the title, author, and page, article or question number, so that I can open it myself.
Explain the same answer more simply, without changing the meaning.
Check references
Every kind of claim
has its own test.
A source the model names is not necessarily a source it has opened.
A Bible verse
Open the verse in your own Bible. Does it exist? Does it say what is claimed? Read the paragraph before and after it.
Look it up: biblegateway.com ↗
An exact quotation
Ask which translation is being quoted. Compare word for word. Distinguish a quotation from a paraphrase or summary.
Compare translations: bible.com ↗
Greek or Hebrew
Look at the word in its sentence, its other uses and its grammar. One dictionary meaning or the word’s origin does not prove what it means here.
Original texts: stepbible.org ↗
Church Fathers and Reformers
Many quotations attributed to Augustine, Luther or Calvin do not occur in their works. Ask for the work, book, chapter or section and look for it in the original text. Can’t find it in the original text? Don’t use it.
Historical texts: ccel.org ↗
Confessions
Ask for the exact article or question number and read it in the full text: the Augsburg Confession, the Heidelberg Catechism, the Westminster Confession or your own church’s statements of faith. AI easily confuses numbers and wording.
Read the full, official text of your church’s confession.
History and numbers
Ask for a traceable source and date. Distinguish a primary source from a later summary. A neat number is not necessarily a measured number.
Look for the original publication.
Books and commentaries
Check title, author, publisher and year with the publisher or in a library catalogue. Invented book titles often look entirely believable.
Open the book or a reliable catalogue.
Web links
Open the link. Does the page exist, and is the quotation really there? A link does not prove that the model read the content.
Read the page yourself, not just the summary.
Before I share or use it
My checklist for an answer.
Your ticks are stored only on this device.
Resources
Practise, discuss
and read further.
An exercise with a document, a first step to begin with, questions for your group and further reading.
Practise with a document
Romans 5:1–5
This exercise shows how AI works with a document you provide yourself. Add the short study aid to a project or chat and ask AI to answer only from it. That way you see whether the answer really comes from your source. The practice text is not itself a Bible translation: read the verses in your own Bible.
Read the practice text
Question for the practice text
Use only the attached document. Give three key points and show the section for each. Say when the document does not contain the answer.
Why like this? “Only the attached document” stops AI filling in from its general training. “Show the section” makes every point checkable. And allowing AI to say that something is missing reduces fabrication.
A first exercise
Start small.
Check two things.
You don’t need to have all the answers to begin with one good exercise.
- Choose a familiar passage, for example Romans 5:1–5 or Psalm 23.
- Read it yourself, together with the verses before and after it. Write down what you notice.
- Make a project with your own instructions.
- Ask for three discussion questions on the passage.
- Check at least two references in the answer.
- Discuss what you found with someone in your Bible study group.
For your Bible study group
Questions to discuss together
- Where has AI helped you, and where has it misled you?
- Where can such a tool help your own Bible study?
- Where do you want to draw a clear line?
- Which part of your judgement and responsibility can you not hand over to a machine?
- How do we protect each other’s privacy when we use AI?
Read for yourself
Bibles and study resources
Bible Gateway ↗
Many English translations to read and check verses yourself, also side by side.
Bible.com ↗
Different translations next to each other, in English and many other languages.
STEP Bible ↗
Free language study: Greek and Hebrew words, grammar and other uses.
Christian Classics Ethereal Library ↗
Historical Christian texts, among them by Church Fathers and Reformers.
These sources, too, call for careful context. The reading principles’ own sources and historical notes are at Sources and historical context.
Further reading
AI, the Bible
and our time.
Original publications and videos.
What AI already makes possible
When AI helps cause harm
How a language model works
Christian voices on AI
About this website
For any Bible reader in the Reformation heritage.
Bible Machine is a guide to Bible study with artificial intelligence (AI). It is not tied to any particular church or denomination. The reading principles come from different Protestant traditions and are presented without church labels: you choose for yourself what fits your study. The website attributes no divine voice, inspiration or capacity for revelation to AI.
Last updated 22 October 2026. Settings of AI tools change often; the date in each section shows when it was checked.
Behind the framework
Follow the thought
back to the source.
Historical texts, methodological expositions and further reading.
The principles are an adaptation of a researched collection. We distinguish between an early source, a later formulation and a present-day explanation. A reference does not prove that every writer cited applies every principle in the same way.
The link titles below describe the subject of each source. The original publications give their full titles and publishing context. A source’s limitations are noted with it.