Select a product:
Introduction to LLM AI Charts and Maps for Excel
It is the same charting engine as the Power BI visual, hosted in Excel: you select a range, say what you would like to see — or say nothing — and a real, interactive chart is drawn in the worksheet. The chart is an object on the sheet, it follows your cells, and it is saved inside the workbook, so it redraws for anyone who opens the file.
This page is the deep dive: how a range becomes a chart, what the sample table and the example charts are for, what linking an account does, how your own AI model reaches Excel, and the chart catalog the engine draws from. The button-by-button reference lives in Help, and the short list of things the two hosts genuinely do differently is here.
Table of Contents
What it does, and where the chart lives
LLM AI Charts and Maps for Excel writes a chart for the data you select and draws it in the worksheet. It is a content add-in: it lives in an object on the sheet rather than in a side pane, which is why the chart can be positioned, resized and printed like anything else in the workbook — and why it is the chart itself you are looking at, not a picture of one pasted out of a chat window.
The chart is real code, saved inside the workbook. Three consequences are worth having up front, because they shape everything below:
- It redraws from your cells, instantly and offline. Edit the bound range and the chart repaints from the code already in the file — no call to us, nothing spent.
- It travels with the file. Send the workbook to a colleague and the chart draws for them with nothing installed and no sign-in.
- Nothing is a dead end. Every version is kept on your account, the code can be read and edited, and an earlier version can be pulled back into the workbook at any time.
Binding a range — and what “the used range” means
There are no data wells here. A worksheet has no model to fill them from, so the range is the data: point the add-in at a rectangle whose first row is the column headers, and every column under that header row becomes a field the chart can use.
Menu > Use selection binds whatever you have selected right now, and the range summary under the heading tells you what it read. Two habits save most of the trouble people hit here:
- Select the headers too. A selection that starts one row too low makes your first data row the column names.
- Or select nothing and let it take the used range. With no explicit selection the add-in reads the sheet's used range — the rectangle Excel considers occupied, from the first used cell to the last. That is the friendly default on a tidy sheet with one table on it, and the wrong answer on a sheet that also carries a title, a stray note three columns over, or two tables side by side. On a sheet like that, select the block you mean.
It charts what you can see. The read takes the visible cells of the range, so rows hidden by a filter or by hand, and hidden columns, are not in the chart — which is almost always what you wanted, and occasionally a surprise worth knowing about. The one case it refuses is a hidden header row: it says so rather than guessing at names.
Two more facts about raw rows, both of which the Power BI visual is spared because a semantic model has already answered them:
- Aggregation is yours to set. Nothing in a sheet has been summarised, so the add-in decides how to summarise a measure per group. The default sums amounts and averages rates; if a column is a percentage, a rate or an index, set Aggregation to Average or Median. Summing a percentage is the classic wrong answer.
- Dates have to be typed as dates. Excel hands dates over as serial numbers, and the add-in reads the cell formatting to tell a date from a number. A date column formatted as General arrives as a plain number and gets charted like one: format it as a date in the sheet and bind again.
Your rows never leave the workbook. A generation request carries the measured shape of the range — your column headers, the type inferred for each one, aggregate statistics and a few short category labels — plus the size of the object, your settings, your locale and anything you typed. See what's sent shows you the exact payload before anything is sent; the full disclosure is in the privacy policy for the add-in.
Seeing it work before you sign in
Three things work with no account at all, and between them they answer “is this any good?” before you commit to anything:
- Insert sample data writes a small demo table to a new sheet and binds to it, so there is something real in front of the add-in within a couple of clicks.
- Show an example chart draws a finished chart over that sample table. Nothing of yours is measured or sent — the data is ours — which is why it needs no acceptance and no sign-in. It is the product doing the actual thing, not a screenshot.
- What fits? reads the columns in your range and lists the chart types that suit them, ranked, with the ones that do not fit explained. It reaches no AI and costs nothing. Because it measures your columns, it asks you to accept the terms first.
And viewing is always free of charge. A workbook that already carries a chart draws it for anyone who opens it, with no add-in to set up and no account. What an active license decides is whether that render carries a watermark — the three cases are set out in Help. Rendering is never refused.
Generating, adjusting, and keeping versions
With data bound, the chart-type list opens by itself, ranked for your columns. Highlight a row to ask for that type, or leave every row unhighlighted and let the engine choose. The “What would you like to see?” box is optional: “revenue by region, as bars”, “dark theme, no gridlines, label the top five” — or nothing at all, which is a perfectly good way to work.
Generate is the step that spends a credit; everything before it is free of charge. Progress streams in the pane, Stop is there while it runs, and the chart draws in the object — usually inside a minute.
After that, the two buttons that matter are Adjust… and Different take. Adjust takes a sentence and keeps your chart while it applies the change, rather than starting over; Different take keeps the data and the chart type and asks for a visibly different rendering of it, for when the type is right and the execution is not. Change type goes back to the list with your data still bound, and spends nothing until you generate.
Every generation is a version held on your account. Menu > Edit the code opens that chart on your account page in the browser, where the code can be read, edited and saved as a new version; typing a version number into the Version setting brings that version back into the workbook. Fetching stored code never calls a model and never mints a new version, so moving between versions costs nothing.
The add-in has no format pane — a content add-in gets none — so the knobs live in a Settings overlay, and they travel with the chart inside the workbook. The distinction to learn first is that a few settings apply now and repaint the chart on screen, while the rest steer what the engine makes and therefore take effect on your next Generate. The full table is in Help.
Signing in, and the link your workbook holds
Generating needs a linked account; viewing never does. Link your account opens our own sign-in page in a dialog — the same sign-in as the rest of the site, multi-factor and all — and the pane comes back naming who it is linked to and how many credits are left.
There is no license key to type into the add-in, on purpose. A typed key is a permanent credential sitting in a file that gets emailed around. What the workbook stores instead is a short-lived link: it is scoped to this one mechanism, it expires on its own rather than sliding forward while you work, and revoking it leaves your license untouched. Open the workbook a week later and Generate will ask you to re-link — two clicks, and it needs your account credentials, which is exactly the control: whoever has the file does not have your account.
None of that touches rendering. A chart draws from the code in the workbook whether the link is live, expired or was never there. If you are ever unsure which state you are in, the pane says so in words rather than leaving you to infer it from a failure.
Licenses are per organization, not per named seat, and one license covers both hosts — the Excel add-in and the Power BI visual, on one account, drawing on the same credits. Licenses and credits are bought on our site rather than through Microsoft, and the current figures are there and on the price list.
Bringing your own AI model
If your organization wants generations to run through its AI endpoint — your Azure OpenAI deployment, your OpenAI or Anthropic key — that works in Excel, and it is set up once, on your account: not in the add-in. Connect the key on your account page (Account → LLM AI Charts and Maps), switch the provider on, and pin it as your account's model.
From then on every generation the account makes — from Excel or from the Power BI visual — is routed to your endpoint, because the override is applied on our server rather than chosen by the client. The Excel pane has no key field and simply shows the default model, which is cosmetic: the routing is already decided by the account it is linked to.
The reasons to do it are the usual two: the model call goes to an endpoint your organization already sanctions, and you pay your AI provider directly, so a generation costs a low fixed service fee instead of the usual credit cost. The setup itself is described step by step in the Power BI section — Setup Bring Your Own Model — and the account it configures is the same account Excel links to.
Chart Catalog
The engine can render every chart type in its catalog, and the add-in draws from the same catalog and the same picker as the Power BI visual. You don't pick from this list directly — you describe what you want (or simply let the AI decide) and it chooses the best fit for your data's shape. This catalog is generated live from the engine's current chart metadata, so it always reflects what's actually available in the deployed build.
One honest caveat for Excel: the add-in renders the D3 lane, so a type that only exists as a Python, Plotly or Vega rendering in the Power BI visual is not offered here. The chart-type list inside the add-in is always the authority for a given range — it ranks what fits your columns, in this host, before anything is generated.
Every type is there with a picture, what it's good for and the data it works best with, and you can search and filter the list.
What differs from the Power BI visual
One engine, one account, one license — but the two hosts are not the same place, and pretending otherwise only costs you time looking for things that are not there. The differences are set out, line by line, in what is different in Excel: where the data comes from, why Aggregation matters here, what clicking a mark does, where the settings live, and what a colleague sees when you send them the file.
The short version: in Excel you select a range and get a chart on the sheet that travels with the workbook; in Power BI you fill data wells and get a visual that cross-filters the report and travels with the .pbix. The chart catalog, the picker, the plain-English refinement, the version history and the account are the same on both sides.
Where to go next: the button-by-button reference is in Help, the product page is here, and the Power BI visual's own deep dive is here.