Key takeaways
- Gems are custom versions of Gemini with saved instructions and files for tasks you repeat.
- Good Gems describe a role, goal, rules, output format and reference material, then get tested and refined.
- Deep Research plans, searches many sources, reads them and writes a cited report over several minutes.
- Lag and crashes usually come from long chats, large files, heavy modes, demand or browser issues.
- Gemini means twins in Latin; Google's name nods to the merger of its AI teams and NASA's Gemini programme.
Gemini Gems are custom versions of Google's Gemini assistant that you set up once with instructions, and optionally files, so it behaves the same way every time for a task you repeat, such as a study coach, writing editor or interview practice partner. They save you re-explaining context in every chat. I design system prompts for AI products, which is essentially what a Gem is, so this page covers Gems, how Gemini and Deep Research work, NotebookLM, lag, crashes and hallucinations, and what "Gemini" means.
What are Gemini Gems?
Gems are personalised Gemini assistants saved with your own instructions and, on supported plans, reference files. Open a Gem and Gemini already knows its role, audience, rules and format, so you can start working immediately.
Also asked as: how gemini works
Google provides premade Gems, such as a learning coach and writing editor, and you can create your own from the Gems manager in the Gemini app. Gems can be shared with others in some plans, which helps teams use consistent prompts. They are similar in idea to custom GPTs in ChatGPT and projects with instructions in Claude: reusable context for repeated work.
How do Gemini Gems work?
When you open a Gem, Gemini adds your saved instructions, and any attached files, to the conversation as context before your messages. The model then follows those instructions, like a system prompt, for every response in that Gem.
To build a good Gem: describe the role, who it helps, the goal, rules to follow, the output format, and what to do when information is missing. Attach reference files, such as a style guide or syllabus, if your plan supports it. Test it with several real requests and refine the instructions where it goes wrong. Clear, specific instructions with reasons work better than long lists of vague rules.
What are some example Gemini Gems?
Useful examples include a study coach that quizzes you on uploaded notes, an email editor that rewrites drafts in your tone, an interview practice partner that asks role-specific questions, a code reviewer for your stack, and a content planner that follows your brand guidelines.
Example instructions for a study Gem: "You are a patient tutor for a second-year BCA student. Explain concepts simply with one example each, then ask one practice question. If I answer wrongly, explain why and give a hint before the answer. Use the attached syllabus to stay on topic." Specific roles, audiences and behaviours make Gems far more useful.
How does Gemini work?
Gemini is a family of large multimodal models from Google. They are transformer networks trained on text, code, images, audio and video, then tuned to follow instructions. When you send a message, the model predicts a response token by token, using the conversation and any tools.
The Gemini app adds features around the model: Google Search for current information, image generation and editing, Canvas for documents and code, Deep Research, Gems, and connections to Google apps. Different model tiers balance speed and capability, and thinking modes reason step by step before answering. For developers, the same models are available through the Gemini API.
How does Gemini Deep Research work?
Deep Research turns a question into a research plan, which you can edit, then searches the web extensively, reads and analyses many sources, and writes a structured report with citations, typically over several minutes.
Also asked as: how gemini deep research works
It works like a research agent: it iterates between searching, reading and deciding what is missing, rather than answering from one search. Reports can be exported to Google Docs. Use it for multi-source questions such as market overviews or literature summaries, and give it a clear scope, audience and format. Verify key claims in the cited sources, since summaries can misread or overgeneralise.
How does a Gemini notebook work?
Google's notebook tool is NotebookLM, which uses Gemini to work only with sources you upload, such as PDFs, documents, slides, websites and videos. It answers questions with citations to your sources, and creates summaries, study guides and audio overviews.
Also asked as: how gemini notebook works
It is a form of retrieval-augmented generation: your sources form the knowledge base, and answers are grounded in them with clickable citations. That makes it useful for studying course material, researching from specific papers or onboarding with company documents. Keep notebooks focused on one topic, and check citations for important claims. Gemini can also connect to NotebookLM content in some setups.
Why is Gemini lagging so much?
Gemini lags when using thinking or Deep Research modes, processing long conversations or large files, generating images or video, during peak demand, or because of slow networks, heavy browser extensions or low device memory.
Fixes:
- Switch to a faster model for simple questions.
- Start a new chat when a conversation gets long.
- Upload only the pages or files you need.
- Close other heavy tabs and disable unnecessary extensions.
- Update the app or browser, or try the mobile app.
- Check your internet connection.
If many users report problems at once, it is likely a service issue that resolves on its own.
Why does Gemini crash so much?
Crashes or failed responses usually come from very long conversations, large uploads, unstable connections, outdated apps or browsers, conflicting extensions, or temporary service problems. On phones, low memory and battery-saving settings can also close the app.
Try clearing the app cache or browser cache, updating the app, signing out and back in, and testing in a private browser window to rule out extensions. Break large tasks into smaller requests. If errors mention capacity or limits, wait or switch models. Persistent crashes on one device but not others point to a device or app issue rather than Gemini itself.
Why does Gemini hallucinate so much?
Like all language models, Gemini predicts plausible text and can produce confident errors when it lacks information, misreads sources or is pushed to answer beyond its knowledge. Questions about niche facts, recent events without search, exact numbers and citations are most error-prone.
Reduce hallucinations by asking it to search and cite sources, providing documents to answer from, asking it to say when it is unsure, and verifying important facts. Grounded tools like NotebookLM, which answer from your sources, hallucinate less on those sources. In my RAG systems, retrieval with citations and a confidence threshold that refuses weak answers serve the same purpose.
What does Gemini mean?
"Gemini" is Latin for "twins". It is a constellation and a zodiac sign associated with the twins Castor and Pollux. NASA's Project Gemini in the 1960s, the two-person spacecraft programme between Mercury and Apollo, used the name too.
Google's Gemini name has been explained by its leaders as a nod to twins, reflecting the merger of Google's two AI teams, Google Brain and DeepMind, into Google DeepMind, and to NASA's Gemini programme as a step towards ambitious goals. In astrology, Gemini is associated with communication and curiosity, but that has no connection with the AI model.
Sources
GenAI engineer and AI product builder with 2+ years shipping production-grade AI systems: RAG pipelines, fine-tuned LLMs, hybrid retrieval and multi-modal apps across vision, speech and OCR, architected end to end from ingestion to deployment. Leads engineering for SCULT INDIA's 14-member team, founded the 500+ member TechVerse Enclave community and has mentored 200+ students. Three hackathon first prizes: Changethon 2025 (IIT Roorkee), Product Genesis at Vividhotsava 2025 (CSJMU) and BYTEBATTLE (MeetKats).
Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at colleges. For campus sessions the inviting institution covers the visit end to end: travel (a First AC train ticket for journeys up to five hours, a flight for anything longer), accommodation, meals, and pickup and drop; the honorarium and workshop fees are negotiable on WhatsApp or on the call, and requests are welcome from institutions that can cover the visit this way. Email pranjulrathour41@gmail.com.
