How to Create Spreadsheets and Tables Using AI? Artificial Intelligence in Excel
Language models have been excellent at handling text from the very beginning. They write, summarize, translate, edit, and do so quickly, efficiently, and often surprisingly well. But whenever a request involved a data table, things became much more cumbersome. Recently, the Claude for Excel extension has changed things considerably. There is, however, a catch—the feature is available only on the paid plan, which costs $20 per month (plus VAT). There are, however, simpler ways to edit tables for free.
While text tables (in Markdown format) could be created in ChatGPT virtually from the very beginning of its existence, this is nevertheless a very basic solution and does not meet all needs. Table features in LLMs were and remain hidden—for example, many people learned about Markdown tables only several months or even several years after its launch.
The lack of artificial intelligence capable of handling spreadsheets is frustrating, because it does not seem like a difficult task. Copilot in Excel is particularly annoying here—it can analyze tables, but it cannot modify the contents of cells. If AI can create fairly complex programs, then why does it have trouble editing a table? Especially since many programs contain fairly complex tables. This intuition is correct—you do not need to create new language models (or, for example, new tabular models) for AI to be able to edit tables. The entire problem lies in the lack of suitable applications; quite simply, no one has bothered to create one. There are various add-ins, it is true, but this raises the issue of trusting third-party companies with our data.
However, you can easily vibe-code simple tools for your own use.
table of contents
-
Free Antigravity (a Google program that needs to be installed)
-
Antigravity, Codex, and other AI tools for writing programs.
Quick overview of the available options.
I discuss each of the solutions mentioned below in greater detail later in the article.
Claude (free)
In Claude, we can ask it to generate an Excel file (XLSX), which can then be downloaded. ChatGPT also has this function, but it works better in Claude—besides, we immediately get a table preview, familiar from Canvas mode. We cannot edit it manually, but the quick preview speeds up our work because we can immediately correct the table using commands. The paid version of Claude, costing PLN 99 per month, includes an Excel plugin (and this plugin is genuinely useful).
A Coded Simple Program in Canvas (Claude, Gemini, ChatGPT).
LLMs such as Claude, Gemini, or ChatGPT can write and run programs directly in a browser. This is a convenient option for getting started with vibe coding—just one prompt is enough to produce a simple working program. Of course, it has its limitations; we will not create The Witcher 3 this way, but it is possible to create some quite useful things. It turns out that spreadsheets are easy for LLMs to code. In 10–20 minutes, you can create a spreadsheet supporting cell merging, formatting, manual editing, copying and pasting, and import/export. It does not have to be a user-friendly editor, because an LLM can modify the program’s code in response to prompts concerning content changes. Once written, the program can be pasted into new threads.
Tables in Gemini Google Drive
Gemini in the paid version enables editing Google Sheets if we set the language to English (although sometimes this is more difficult than it might seem). Some people are unaware that they have the option to integrate Drive with Gemini, as it is added, for example, to the company’s Workspace package and other Google services. However, although this integration works, it does not work particularly well.
ChatGPT
ChatGPT’s paid version has interactive tables that can be opened in full-screen mode and modified manually. In addition, it can generate a downloadable file, while providing only a minimal preview (less convenient than what Claude offers).
While this article was being written, ChatGPT 5.4 was released. OpenAI’s latest model is reportedly optimized for working with spreadsheets. There is also an official plugin connecting ChatGPT with Excel, but it is currently in beta—we can assume that it will be made publicly available in the near future.
Free Antigravity (a Google program that needs to be installed)
Antigravity or Cursor is a program for creating other programs, or more precisely, a development environment. It enables language models to modify multiple files simultaneously. This tool is optimized for programming languages, but it can easily modify other files as well, including Excel spreadsheets. Installation takes a bit of effort, but is fairly simple. Its advantage is the ability to work with large files. Disadvantages: browsing the internet with this LLM is somewhat more difficult, although entirely possible, and operations are performed directly on our files, which is somewhat less secure. Advantages: in addition to working with large files and multiple files at once, it is also possible to create simple programs that make it easier to create spreadsheets.
We can use OpenAI’s Codex in a similar way.
Claude Plugin in Excel
The most professional solution. It is widely praised by people who work with spreadsheets. It requires installing an extension in Excel (which is simpler than installing a new program). In the Windows Store, it has around 2–3 stars (with many 5-star and 1-star ratings), because it requires a paid version costing approximately PLN 99 per month—so if someone found this add-in in the Windows Store, installed it, and was unaware of this requirement, they may be dissatisfied.
Other Plugins
In the world of professional Excel extensions, GPT for Excel (Talarian) leads the way, allowing users to seamlessly work with GPT, Claude, or Gemini models in a single spreadsheet. Its “Agent Mode” automates the creation of pivot tables and the extraction of insights, while its bulk processing feature (handling up to one million rows) combined with your own API key (BYOK) dramatically reduces costs for large-scale projects.
If, on the other hand, your priority is research and working with text, Numerous.ai will be the better choice. This tool excels at extracting specific information from unstructured data, analyzing the sentiment of thousands of reviews in a flash, and generating ready-to-use content—such as SEO descriptions or email replies—directly in Excel cells.
Both of these plugins are paid, although unlike Claude for Excel, they allow you to test their features for free.
Applications of Artificial Intelligence in Excel
General Applications of Spreadsheets
To answer this question, one must first understand the general capabilities of spreadsheets in management, data analysis, and learning. Many people are unaware of the possibilities that ordinary tables offer. We can distinguish a visual aspect (the structure of a table is simply useful when we have many items—especially when each row is tall and the cells contain images; tables can also serve a function similar to mind maps) and a programming aspect—Excel can perform complex calculations.
Combining these two functions produces very good results. For example, if we have around 100 objects (products, companies, concepts, etc.), we can keep them in a single table and conveniently filter them as needed—so there is no need to create a dozen versions of the same file.
Programming functions are particularly useful for calculating costs, as tools developed for internal use and as support in managing small teams. Their main advantage is that the spreadsheets can be easily modified, and no specialist is needed for this. This eliminates the problem of testing new tools and migrating to new solutions when the old one turns out to have shortcomings.
A spreadsheet that is essentially a simple program can be created over the course of a weekend (assuming no prior knowledge of spreadsheets), and the finished document can also be easily modified by people without any knowledge of Excel.
AI in spreadsheets
Artificial intelligence significantly simplifies the process of learning Excel. Previously, you had to learn every formula and solution on your own. There was a wealth of material available online, but figuring out why something did not work took some time.
AI can write formulas for us and suggest why something isn’t working, so we can focus on creating the table’s logic without worrying about semicolons, parentheses, or colons.
Uploading table screenshots, pasting formulas, or uploading a file to the model and talking to AI about tables is useful in itself. However, it would be nice to speed up this process and skip the copying step.
AI supporting data analysis in spreadsheets.
In this case, artificial intelligence does not modify the file but reads it and draws useful conclusions from it. For example, AI can identify different customer segments or assess the cost of tasks in relation to their results. Additionally, artificial intelligence can suggest how to sort the table so that we can independently arrive at valuable analytical conclusions. It is worth mentioning that using language models to analyze data is a provisional solution, useful as support in small companies. A more robust solution is to use machine learning. It requires less computing power than typical LLMs, so such analyses can be carried out on ordinary computers—the costs start at several thousand zlotys.
The Solution I’m Looking For
I would like a tool that can automatically generate tables based on text and fill existing tables with data. This is the most tedious part of working with tables. Assistance in coming up with a table layout would also be very useful. It is important that the artificial intelligence be able to handle somewhat more complicated layouts, such as merged cells, columns and rows grouped into larger sets, and so on. The LLM handling the table does not need to be a master at writing text, reasoning, and collecting data—we have a range of other tools for that. A mode in which I create data-filled text and then paste it into a spreadsheet, with the AI creating the layout itself, would already be a major step forward. Skipping the step of pasting a large file would be convenient, but it is not essential.
Claude AI – the current king of tables.
In the paid version (costing PLN 99 per month), you get access to a plugin integrated with Excel.
This plugin works incredibly well. It does exactly what artificial intelligence should do. Claude can work with merged cells (many language models have trouble with this because they are programmed to assume that a table consists of simple columns and rows, with the top row serving as the column header).
A practical example. First, I pasted in a raw table, and Claude had no trouble formatting it nicely; additionally, it filled it with content from the specified page.
But that’s not all. The first table contained everything: individual lessons and surfing camps. I decided it would be better to split it into two tables—the upper one focuses on individual lessons, with quick information on whether camps are available for a given discipline, while the lower table focuses on camps and trips.
Claude did a fairly good job of figuring out what columns the bottom table should have. This may not be entirely visible in the screenshot (especially if you’re reading this blog on your phone). In the original table, six columns concerned the number of people and the duration of the lessons—in the camp table, these columns are not included; instead, we have columns such as duration, meals, and accommodation.
Tables as a powerful tool in research.
For many topics, it is useful to create a list of similar items (e.g., concepts, characters, companies, etc.). A table is a better form of list because it allows you to sort objects according to various parameters. However, creating tables manually is time-consuming, so people often skip this stage of the research if it is not absolutely necessary. Sometimes, pasting simple tables generated in ChatGPT was a solution, but this manual clicking still took a considerable amount of time. With Claude, you can easily generate a table and then adapt it to your needs. Unfortunately, Claude cannot download images, but it can prepare links to the relevant locations.
Summary: Is Claude Worth Buying for Spreadsheets?
In my opinion, absolutely, if you already use traditional spreadsheets. It is definitely worth testing this feature. If we cancel the subscription within 14 days, we can get a refund (although I haven’t checked how this works in practice), but even spending PLN 100 on testing Claude is a worthwhile investment. Faster table creation and editing means I will use this feature more often. In many cases, I can see that a table containing specific data would be very useful—for example, when I write about companies developing AI in Poland, a table with such data would be useful for many of my articles. Generally, though, creating a useful table takes 2–6 hours. Using web scraping tools or traditional LLMs speeds up the process, but a lot of the formatting still has to be done manually. With Claude, creating a quick table with 15 columns and 40 entries takes a maximum of 30 minutes.
Editing Tables Using ChatGPT 5.4.W
The latest model from OpenAI, namely ChatGPT 5.4 is optimized for working with tables. Ultimately, a plugin enabling integration with Excel is expected to be made available, functioning similarly to the one for Claude, although it is not yet widely accessible.
Paid Gemini Workspace – you need to set English as the language in Google; it is useful, but does not always work well.
The Google Drive ecosystem integrates with Gemini. So, if you have a paid version, you may already have access to AI that edits spreadsheets. Many people are unaware of this because they use Google as a drive or as a corporate email client. Another pitfall awaits us here: the spreadsheet editing feature only works when English is set as the language (without it, you can only discuss spreadsheets, ask for analyses, etc.). We therefore need to change the account settings to English.
When it finally works, Gemini’s performance is underwhelming. The LLM cannot modify an existing table; it can only insert new interactive tables.
How to change the language in Google?
Contrary to appearances, it is not as simple as it might seem. The easiest way is to click this link: https://myaccount.google.com/language
Alternatively, you can click through everything: click our avatar icon in the top-right corner, then select: Manage your Google Account.
Next, we select personal data, and then select: language.
And to make things even less straightforward, it’s not enough to set English as the primary language—we also need to remove Polish (if language detection is enabled, Google will add it again anyway). Changing the language takes a moment.
How Do AI Tables Work in Google Spreadsheets?
Essentially, Gemini’s capabilities in spreadsheets are currently limited to data analysis and inserting interactive tables. The only advantage is that we don’t have to paste data from the table into the language model for Gemini to do something similar.
Vibe-Coded Applications
If you don’t want to pay PLN 99 per month for Claude, an alternative is to build simple applications that run directly within the language model in code interpreter mode (Canvas). You can use ChatGPT, Claude, or Gemini for this. In an application built this way, the language model can generate tables and modify their contents, which we can then paste into Google Sheets or Excel.
A few commands are enough for this; doing everything in a single command doesn’t work very well.
The process for creating such an application is as follows:
- Write a program that runs in Canvas in Gemini; it should be a simple spreadsheet
- Add cell-merging options
- Add table import and export options
- Add options for copying and pasting selected groups of cells
- Add an option to export as an Excel file
In this way, completely free of charge and without the need to add API keys or anything like that, we obtain the ability to manipulate tables using a language model.
The advantage of this solution is its good integration with the internet and all other language model functions. In addition, we can easily add new features, such as viewing a single row in the form of a form.
Antigravity, Codex, and Other AI Tools for Writing Programs.
One provisional solution is to use AI tools designed for writing code. Such programs can modify files on our disk, and since they can change code files and modify programming data stores (e.g., JSON files), they can also modify tables.
This is somewhat more difficult than creating a simple program with browser-based language models, and setting it up will take a little more time, but in this case, the AI works with files that are immediately available on our disk. Good results can be achieved by writing instructions both for the agent directly in the Antigravity program and instructions in the folder where the AI creates tables. Additionally, we can create database and programming files, such as JSON files, from which the AI agent then creates tables.
AI Agent in the Browser.
Another way to edit a table is to use AI agents operating in the browser. Such an agent does not operate on the backend; instead, it clicks and enters text in the same way a human would.
The drawback of this solution is slower performance; it took the agent approximately 10 minutes to create such a simple table. The advantage of agents is that, in theory, they can work with any program.
Access to agents handling the “screen, mouse, and keyboard” can be obtained at:
- the paid version of Claude
- the paid version of ChatGPT: directly in the program, the GPT model launches a virtual machine, as well as within the Atlas browser
- in the free version of Antigravity
- Programs such as Manus, Clavbot, Moltobot, OpenClav.
If we add the appropriate instructions, such an agent can work better; however, the topic of AI agents is a separate issue.
Access to agents operating the screen, mouse, and keyboard can be obtained at:
- the paid version of Claude,
- the paid version of ChatGPT: directly within the program, the GPT model launches a virtual machine, as well as within the Atlas browser,
- the free version of Antigravity,
- programs such as Manus, Clavbot, Moltobot, and OpenClav.
If we add the appropriate instructions, such an agent may perform better; however, the topic of AI agents is a separate matter.