A plain-English guide for anyone who does not know where to start
Artificial intelligence, usually shortened to AI, is a broad name for computer systems that can perform tasks we normally associate with human intelligence.
The type of AI discussed in this guide is generative AI. It can produce new text, images, audio, plans, summaries and other material in response to an instruction.
AI can sound remarkably human, but it is not a person. It does not know your community, care about a donor or understand the weight of a pastoral conversation. It works from patterns, instructions, information and tools. That can make it extremely useful, but human judgment remains part of the job.
An AI assistant is the website or app you talk to. You type or speak naturally, much as you would explain a job to a capable assistant. It can answer questions, help you think, work with files and produce drafts or other outputs.
ChatGPT, Claude and Gemini are three of the best-known AI assistants. They are made by different companies, but there is a great deal of overlap in what they can do.
| Assistant | Made by | The simplest explanation |
|---|---|---|
| ChatGPT | OpenAI | A general-purpose AI assistant with strong voice, file, research, image and productivity tools. |
| Claude | Anthropic | A general-purpose AI assistant that is particularly strong for writing, analysis and working through complex material. |
| Gemini | Google’s AI assistant, with close connections to Google products and Android. |
These products change constantly. One may be better at a particular job this month and another may catch up the next. A beginner does not need to monitor every model release or maintain three subscriptions.
Our current recommendation is ChatGPT or Claude.
If you want one simple default, begin with ChatGPT. It is a strong all-round assistant, works well on a phone and gives a beginner one place to learn the basics.
Claude is also excellent, and many people prefer the way it handles writing, long documents and extended thinking.
Gemini is useful, especially for someone who already works heavily inside Gmail, Google Drive and Android. For most shluchim beginning now, however, ChatGPT or Claude would be our first choice.
No. The free versions are enough to learn how AI works and can already be genuinely useful. You can chat, search, work with files and use a range of tools, although the exact limits and features differ between products.
We strongly recommend a paid plan for anyone who expects to use AI regularly in their shlichus.
Paid plans generally give you more usage, fewer interruptions, larger limits and fuller access to the strongest models and features. You do not need to pay before you understand the basics, but the paid version usually delivers far more value once AI becomes part of your work.
If a job involves words, information, ideas, files or planning, AI can probably help with at least part of it. It is especially useful for turning rough material into something organized and usable.
AI is often strongest at the first pass and the busywork. It can give you a starting structure, catch inconsistencies and save time. The final decision, relationship and approval still belong to a person.
Yes. Voice is one of the easiest ways to use AI, especially for someone who does not enjoy typing or writing prompts.
Use Voice mode and explain the situation as if you were leaving a detailed voice note for a capable assistant. Say what happened, who is involved, what you are trying to achieve and where you are stuck.
You can have a back-and-forth conversation, answer follow-up questions and ask it to turn the conversation into a draft, plan or summary.
Voice and dictation are related but different.
Voice lets you have a conversation with the AI. A dictation tool such as Wispr Flow turns your speech into written text inside WhatsApp, Gmail, Docs or another app.
Yes, in most cases.
AI assistants can work with PDFs, Word documents, presentations, spreadsheets, screenshots and photos. You can ask for a summary, find a particular detail, compare two documents, check the tone, examine spreadsheet data or explain what appears in an image.
Support for prerecorded voice notes varies by assistant, device and plan. The simplest option is usually to speak through the assistant’s Voice feature or use a transcription tool first. Some products and modes also accept audio files directly.
Uploading a file gives the AI real material to work from, which is usually more reliable than asking it to remember the contents.
There is no magic prompt. A good prompt is a clear explanation of the job.
Tell it what you have, what you need and what a good result should look like. “Make it better” forces the AI to guess what better means. A specific instruction gives it something concrete to work toward.
Instead of:
Try:
The most useful details are usually the audience, the goal, the facts that must be included, the tone, the limits and the format you need.
Give it the information a good staff member would need before starting.
A letter may require:
A program plan may require:
A simple way to structure a request:
Tell it plainly what you dislike.
AI tends to fill gaps with generic openings, exaggerated praise and polished sentences that say very little unless you give it a clearer standard.
You can say:
Examples of your own writing help even more. Give it one or two pieces that sound like you and ask it to match the level of formality, rhythm and directness without copying the wording.
No.
It may lack recent information, local knowledge, private information or the particular context that makes an answer right for your situation. It can also misunderstand a source or repeat information that was wrong to begin with.
A confident tone is not proof. AI can produce a polished answer even when the facts underneath it are weak.
ChatGPT, Claude and Gemini can search the internet, but you should not assume that every answer came from a live search.
Sometimes the assistant answers from its existing knowledge or from the material already in the conversation.
When current information matters, say:
A link is helpful, but it is not a guarantee. Open important sources and check that they actually support the claim being made.
AI generates the answer that appears most likely to fit your request.
When information is missing or uncertain, it may fill the gap with something that sounds believable but is not true. This is called a hallucination.
Hallucinations can include:
Giving the AI real source material reduces the amount it has to guess, but does not remove the need to check.
Check the parts that would matter if they were wrong:
A useful final question is:
Do not treat a general AI assistant as an authority on Torah.
It may invent a mareh makom, combine two sources, misquote the wording or give an explanation that sounds familiar but is not actually found in the sefer it names.
This is one of the reasons we built the Torah MCP.
An MCP is a way of connecting an AI assistant to outside tools and information. The Torah MCP gives the AI access to real Torah texts instead of asking it to rely only on what it remembers. That makes the answer more grounded and far easier to verify.
The Torah MCP does not turn AI into a rav, and it does not make every answer infallible. It gives the assistant better source material. Important sources, quotations and conclusions still need to be checked inside.
The same principle applies even without the Torah MCP. If you want AI to help explain a maamar, sicha or halachic source, give it the actual text.
Working from the real source will nearly always be more reliable than asking it to recall the source from memory.
It remembers what has been said within the current conversation.
Depending on the assistant, plan and settings, it may also use saved memories or information from previous chats to personalize later answers.
Do not rely on memory for an instruction or fact that really matters. Repeat the important detail in the current conversation or keep related work together in one project.
You can also review or turn off memory features in the assistant’s settings.
Before pasting or uploading information, ask whether you would be comfortable forwarding it to an outside service.
Use the smallest amount of information needed for the job, and anonymize or remove identifying details where possible.
As a safe default, do not share:
An approved business workspace may have different data controls from a personal account. Follow your organization’s policy and check the settings before relying on a general promise about privacy.
AI is useful for preparation, organization, drafting, sorting, translation, research and busywork.
Keep human responsibility for:
A donor message is a good example.
AI can turn real notes into a cleaner draft, but the appreciation, memory and personal line must come from you. Read the result before sending it and remove anything that overstates, invents warmth or sounds like a mass email.
You do not need to use all of this language yourself. These definitions are here so the common terms no longer feel mysterious.
The basics of how to get more out of AI. Learn the first 4 levels of the 7 levels of AI usage and discover many tips and tricks.