# Gemini Gems and Gemini Live: What They Can Do and the Limits Google Declares

> What Google's official materials say about Gemini: documented features, declared limits and what is still missing on Gems and Gemini Live.

Source: Giallo Studio (giallostudio.it) · Page: https://giallostudio.it/en/blog/gem-gemini-live-cosa-fanno-limiti · Published: 2026-11-09 · Updated: 2026-11-09

If you are wondering what Gemini Gems are and what you can do with Gemini Live, the honest answer is a partial one: the five official Google materials we read describe neither Gems nor Gemini Live. So we cannot tell you what they do, how they are set up or what limits they have, and we would rather not guess.

What the materials do document well is the rest of the Gemini ecosystem for work: Gemini Notebook, a research assistant that answers based on your sources, Gemini in Sheets for formulas and data, and custom agents in Google Workspace. A clear definition: these are generative AI tools, meaning systems that produce text and other content from a request, and for that reason they can make mistakes.

In this guide you will find the documented features, a work example for each, the limits Google declares and a path to try them out in your company. When official documentation on Gems and Gemini Live becomes available, the article will need to be updated.

<div class="key-points">
<p class="kp-title">In summary</p>
<ul>
<li>The official materials we were given contain no information on Gems and Gemini Live: you will find no claims about them here.</li>
<li>Gemini Notebook answers based on the sources you upload and shows inline citations so you can check every step.</li>
<li>Gemini in Sheets helps with formulas, visualizations and data completion, but its answers need to be verified.</li>
<li>Google states that Gemini apps can produce inaccurate information and that answers are not medical, legal or financial advice.</li>
<li>For business data, Google Workspace states that it is not used to train the models.</li>
</ul>
</div>

## What do we really know about Gems and Gemini Live?

Very little, and we say so explicitly. The texts we verified are the Gemini Notebook guide, two Gemini app guides on generative AI and on answers, the Gemini in Sheets page and the Workspace page on AI tools for businesses. Neither Gems nor Gemini Live appear in any of them.

For anyone making decisions in a company, this matters: a feature without official documentation has no declared limits, no known privacy terms and no activation requirements. Before adopting it you need a primary source. For the general picture of the product, you can start from our explanation of [what Gemini is](https://giallostudio.it/en/blog/what-is-gemini-ai), while for the Carbon model you will find what Google declares in the article on [Gemini Carbon and its availability](https://giallostudio.it/en/blog/gemini-4-carbon-what-it-is).

## Which Gemini features are documented for work?

The table summarizes what the official sources allow us to state, with the declared limit next to each feature.

| Tool | What it does according to Google | Limit or warning | Source |
|---|---|---|---|
| Gemini Notebook | Research assistant that organizes ideas and answers based on uploaded sources | Can make mistakes | [Gemini Notebook guide](https://support.google.com/gemininotebook/answer/16164461?hl=it) |
| Gemini in Sheets | Structures data, creates visualizations and formulas | Works on data you need to check | [Gemini in Sheets](https://workspace.google.com/intl/it/resources/spreadsheet-ai/) |
| Custom agents | Used with Docs, Gmail and more, without code | No specific limits are described in the materials | [Workspace AI](https://workspace.google.com/intl/it/solutions/ai/) |
| Gemini apps in general | Respond to requests in natural language | Answers are sometimes inaccurate | [Gemini answers guide](https://support.google.com/gemini/answer/16279220?hl=it) |

### Gemini Notebook: research on your sources

What it does: according to the guide, it is an AI-based research assistant designed to refine and organize ideas, and you can upload PDFs, websites, YouTube videos, audio files, Google Docs or Google Slides. Source: [Gemini Notebook guide](https://support.google.com/gemininotebook/answer/16164461?hl=it).

How to use it: you create a notebook, add sources and ask questions. The guide specifies that answers are based on the information in the uploaded sources, so anything that is not in the documents should not appear as fact.

Work example: a payroll consulting firm, as a hypothesis, uploads circulars and internal instructions and asks for a comparison of two procedures. The result serves as a draft to be reread.

Limits: Gemini Notebook can make mistakes and its answers do not reflect Google's point of view, as written in the [same guide](https://support.google.com/gemininotebook/answer/16164461?hl=it). For an operational path, see our article on [NotebookLM for company documents](https://giallostudio.it/en/blog/notebooklm-documenti-aziendali).

### Inline citations: checking where the answer comes from

What it does: the guide indicates that you can get information based on your sources with clear inline citations, to ensure accuracy, transparency and reliability. Source: [Gemini Notebook](https://support.google.com/gemininotebook/answer/16164461?hl=it).

How to use it: read the answer and open the citation next to each important statement, comparing it with the original document.

Work example: a real estate agency, as a hypothesis, asks for a summary of a condominium regulation and checks every point against the citation before replying to a client.

Limits: the citation makes checking easier, but does not replace it. For a verification method, use our [checklist for verifying AI answers](https://giallostudio.it/en/blog/verificare-risposte-ai-prima-di-inviarle).

### Turning sources into guides, briefings and mind maps

What it does: sources can be turned into accessible formats such as study guides, briefings, audio overviews and mind maps. Source: [Gemini Notebook guide](https://support.google.com/gemininotebook/answer/16164461?hl=it).

How to use it: after uploading the material, choose the format that suits whoever will read it, for example a briefing for people short on time.

Work example: an installation company, as a hypothesis, turns its internal manual into a briefing for new team members.

Limits: the new format inherits the possible errors of the sources and of the model, so a responsible person must review it before it is distributed. The same guide also notes that the service supports over 80 languages, useful for multilingual materials, [according to the documentation](https://support.google.com/gemininotebook/answer/16164461?hl=it).

### Gemini in Sheets: formulas, data and charts

What it does: according to the official page, Gemini helps structure data, create effective visualizations and formulas that simplify complex calculations, and can proactively suggest additional formulas. Source: [Gemini in Sheets](https://workspace.google.com/intl/it/resources/spreadsheet-ai/).

How to use it: describe in plain language the calculation you need and compare the proposed formula with a few rows whose result you already know.

Work example: an online store, as a hypothesis, asks for a formula to sum orders by customer and a chart of sales by category.

Limits: the page talks about reducing errors, but does not promise an absence of errors; the general guides remind us that Gemini apps [can produce inaccurate information](https://support.google.com/gemini/answer/16279220?hl=it). To go deeper, see our guide to [Gemini for Google Sheets](https://giallostudio.it/en/blog/gemini-for-google-sheets).

### Gemini in Sheets: data completion

What it does: it can recognize patterns as you enter data and automatically fill in the remaining entries, and it works with large data sets, with numeric values, text and other information in cells. Source: [Gemini in Sheets](https://workspace.google.com/intl/it/resources/spreadsheet-ai/).

How to use it: you start filling in a few rows and check the proposed entries before accepting them.

Work example: a craftsman, as a hypothesis, orders a list of materials whose codes follow a pattern and lets the tool complete the missing rows.

Limits: an autocomplete that looks consistent can still be wrong, so important rows should be spot-checked.

### Custom agents in Google Workspace

What it does: the Workspace page indicates that you can create custom agents and use them with Docs, Gmail and more, without writing code. Source: [AI tools for businesses](https://workspace.google.com/intl/it/solutions/ai/).

How to use it: define the agent's task and test it on non-sensitive documents before extending it.

Work example: a professional firm, as a hypothesis, prepares an agent that helps draft the first version of a reply starting from an internal record.

Limits: the materials contain no details on permissions, costs or requirements. The full documentation should be read before deciding. The topic is linked to our guide on [Gemini for Google Workspace](https://giallostudio.it/en/blog/gemini-for-google-workspace).

## Which work examples make sense for an SME?

The following examples are scenarios built to explain, not real client cases.

**An accounting firm.** It uploads the firm's operating instructions into a notebook and asks for a memo for a new team member. Inline citations make it possible to check every step, and the guide reminds us that answers [should not be considered advice](https://support.google.com/gemini/answer/16279220?hl=it) from a professional.

**A small e-commerce store.** It uses Gemini in Sheets to build formulas and charts on sales, then compares the results with its management system before making decisions on stock or prices. [The Gemini in Sheets page](https://workspace.google.com/intl/it/resources/spreadsheet-ai/) describes support for formulas and visualizations, not the replacement of human checks.

## What do the sources say about privacy and business data?

<div class="callout">
<p><strong>Business data.</strong> The Workspace page states that a company's data is strictly the company's and is not used to train the models. Source: <a href="https://workspace.google.com/intl/it/solutions/ai/">AI tools for businesses</a>.</p>
<p>For Gemini in Sheets, no content is shared with outside parties or used to train models outside your organization without authorization: <a href="https://workspace.google.com/intl/it/resources/spreadsheet-ai/">Gemini in Sheets page</a>.</p>
<p>For Gemini Notebook, with a Workspace account, uploads, queries and responses are not reviewed by human reviewers or used to train the models. With a managed work account, the administrator must enable access: <a href="https://support.google.com/gemininotebook/answer/16164461?hl=it">Gemini Notebook guide</a>.</p>
</div>

These are statements by the provider and do not replace an internal check. To set clear rules, also read how [business data privacy with AI](https://giallostudio.it/en/blog/ai-privacy-for-business) works and how to write a [corporate policy on the use of AI](https://giallostudio.it/en/blog/company-ai-use-policy).

## Why can generative AI get things wrong?

Google's guide on generative AI explains that it can make up things that are not true and that this phenomenon is called hallucination. The cause, according to the same guide, is that unlike Google Search, LLMs do not gather information from the web. Source: [Learn more about generative AI](https://support.google.com/gemini/answer/13954172?hl=it).

The guide also encourages you to think critically about answers and to use Google and other resources to verify information presented as factual. It also specifies that generative code features are experimental and that the user is responsible for how the suggested code is used: [same guide](https://support.google.com/gemini/answer/13954172?hl=it). To explore the phenomenon, read [AI hallucinations and how to avoid them](https://giallostudio.it/en/blog/ai-hallucinations).

## Which mistakes should you avoid?

- **Assuming Gems work in a certain way.** It happens because the name is everywhere. Avoid it by looking for the official documentation before promising features to clients or colleagues.
- **Treating answers as professional advice.** The [Gemini app guides](https://support.google.com/gemini/answer/16279220?hl=it) say not to consider them medical, legal, financial or other advice. Avoid it by always keeping a responsible professional in the loop.
- **Trusting citations without opening them.** Citations help with checking, but only if someone actually reads them against the original document.
- **Uploading sensitive material without rules.** Even with Workspace's privacy statements, you need an internal policy on what to upload and who can do it.
- **Forgetting the administrator.** With a managed work account, Gemini Notebook requires activation by the administrator, [as the guide indicates](https://support.google.com/gemininotebook/answer/16164461?hl=it). Ask for it to be enabled before planning a pilot.

## How do you apply it in your company?

Prerequisites to have ready before you start:

<ul class="checklist">
<li>A Google Workspace account with access enabled by the administrator.</li>
<li>A small set of non-sensitive documents to test on.</li>
<li>A person responsible for rereading every result.</li>
<li>A written rule on which data can be uploaded.</li>
</ul>

The steps for a first test:

<ol class="steps">
<li>Choose a repetitive, well-defined task, for example an internal summary or a recurring formula.</li>
<li>Prepare the sources and ask the administrator to enable the tool.</li>
<li>Ask the same questions to Gemini and to an experienced person, and compare the answers.</li>
<li>Note the errors and decide whether the tool is useful as it is or only with a review.</li>
<li>Extend its use only after writing rules on data and responsibility.</li>
</ol>

What can go wrong: plausible but wrong answers, formulas that do not account for a special case, unverified citations, access that has not been enabled. To set up a pilot, also see the guide to the [AI pilot project](https://giallostudio.it/en/blog/ai-pilot-project).

## What can a company do today?

With the verified facts, a company can organize documents with Gemini Notebook, build formulas and charts in Sheets and evaluate custom agents in Workspace, always with human review. On Gems and Gemini Live, it is best to wait for official documentation before building a process on them.

If you want to understand which of these tools makes sense in your processes and how to control it, you can start with an [AI consulting](https://giallostudio.it/en/ai-consulting) engagement with Giallo Studio.

## Frequently asked questions

### What are Gemini Gems?

The official Google materials we verified contain no description of Gems. For this reason we cannot say what they do or what limits they have. It is best to consult Google's official documentation before using them in your company.

### What can you do with Gemini Live?

The materials we were given do not mention Gemini Live, so we cannot list its features. We prefer not to attribute undocumented capabilities to it. We will update the article when an official source becomes available.

### Does Gemini Notebook answer only based on my documents?

According to the official guide, it is designed to answer based on the information in the uploaded sources and offers inline citations. It can still make mistakes. Citations should therefore be checked against the original document.

### Is my company's data used to train the models?

The Workspace page states that company data is not used to train the models. Similar statements are also reported for Gemini in Sheets and for Gemini Notebook with Workspace accounts. These are claims by the provider, to be paired with an internal policy.

### Can I use Gemini's answers as legal or financial advice?

No. The Gemini app guides say not to treat answers as equivalent to medical, legal, financial or other advice. Verification and a responsible professional are always needed.
