How to Create Charts with Copilot in Excel: The Complete Guide

Copilot in Excel can genuinely make charts now. That sentence would have been marketing fiction in 2024, generous in 2025, and is simply true in mid-2026. What Microsoft has not done is make it comprehensible: there are at least four distinct ways Copilot produces a chart in Excel — Copilot Chat, Edit with Copilot, Agent Mode, and Python-generated visuals — plus a =COPILOT() worksheet function, and Microsoft’s documentation treats the names as interchangeable when they absolutely are not.

This guide untangles all of it. You’ll get the honest map of which Copilot feature does what, prompt patterns that reliably produce usable charts, a clear-eyed list of what Copilot charts well and what it charts badly, and the situations where you should close the Copilot pane and build the chart yourself.

One position up front, because it frames everything that follows: Copilot is now the fastest way to produce a first draft of most standard Excel charts, and it is still the wrong tool for a final chart going into anything that matters. The gap between those two is smaller than it was a year ago. It is not closed.

First, the naming mess: which Copilot are we talking about?

If you’ve found conflicting tutorials online, it’s because Microsoft has shipped, renamed, deprecated, and re-shipped Excel AI features at a pace no publication has kept up with. Here’s the landscape as of July 2026:

Three dates worth knowing, because they explain why older tutorials mislead you. The old “advanced analysis” App Skills experience was deprecated and removed by late February 2026 — any tutorial referencing it is describing a dead feature. Agent Mode went generally available across Excel for web, Windows, and Mac, and in April 2026 Microsoft made agentic Copilot the default experience across Microsoft 365 subscriptions. And the April 2026 Excel update added the Chat/Edit switcher plus Python execution inside Edit with Copilot.

If you want the full head-to-head on the two most-confused features, we’ve written it: Excel Agent vs Copilot: What’s the Difference? The short version — Agent Mode plans and executes multi-step jobs autonomously; standard Copilot responds to one request at a time.

Before you prompt: the setup that determines everything

Copilot’s chart quality is downstream of your data structure, and this is not a minor point — it’s the single biggest predictor of whether you get a usable chart or garbage.

Format your data as an Excel Table first (select the range, Ctrl+T). Copilot reads Tables with named columns far more reliably than loose ranges. Headers become series names, and Copilot stops guessing where your data ends.

One header row, no merged cells, no subtotals mixed into the data. Copilot inherits every ambiguity in your layout. A human looks at a merged-cell header and understands it; Copilot looks at it and improvises.

Save to OneDrive or SharePoint. Agent Mode and several Copilot capabilities require it. Local-only files get you a degraded experience or none at all.

You’ll also need a qualifying license — Microsoft 365 Copilot (business) or Microsoft 365 Premium/Personal/Family plans carry the current feature set.

Making charts with prompts: the patterns that work

The difference between a frustrating Copilot session and a productive one is almost entirely prompt construction. The pattern that works is: data reference + chart type + comparison intent + any formatting constraint.

Goal	Weak prompt	Strong prompt
Trend over time	“Chart my sales”	“Create a line chart of monthly revenue from the Sales table, one line per region, for 2025–2026”
Category comparison	“Make a bar chart”	“Create a clustered column chart comparing Q1 vs Q2 units sold by product category, sorted descending by Q2”
Part-to-whole	“Show the breakdown”	“Create a 100% stacked column chart of expense share by department per quarter”
Target tracking	“Chart actual vs budget”	“Create a combo chart: monthly actuals as columns, budget as a line, from the Budget table”
Distribution	“Analyze the spread”	“Use Python to create a box plot of order values grouped by sales channel”
Correlation	“Do X and Y relate?”	“Create a scatter plot of ad spend vs revenue by campaign, and add a trendline”

Three rules underneath the table:

Name the chart type yourself. Copilot will recommend a chart type if you don’t, and its recommendations are conservative to a fault — it reaches for clustered columns the way a nervous cook reaches for salt. You should be deciding the chart type based on your data’s shape; if you’re not sure how, our chart decision guide exists for exactly this, and it’s a faster read than three rounds of Copilot revisions.

One chart per prompt in Chat/Edit. Compound requests (“make a line chart of revenue and also a pie of category share and format them nicely”) degrade output quality sharply. Batch work belongs in Agent Mode.

Iterate rather than restart. “Change the y-axis to start at zero,” “move the legend to the bottom,” “make the 2026 series orange” — follow-up refinements work well and preserve what’s already right. We maintain a full library of 40 ChatGPT prompts for better Excel charts; most translate directly to Copilot with minor wording changes.

Walkthroughs by chart type: what to expect from each

The table gives you the patterns; here’s how each of the major chart requests actually plays out in practice, because Copilot’s competence varies more by chart type than Microsoft admits.

Column and bar charts are Copilot’s home turf. From a clean Table, “create a clustered column chart of revenue by region for each quarter” lands correctly first time in the overwhelming majority of runs. Where it stumbles is sorting — it defaults to source order, so append “sorted descending by total” to the prompt rather than fixing it afterward. Horizontal bars need to be requested explicitly (“bar chart,” and if it gives you columns anyway, “make it horizontal”); Copilot shares the general population’s bias toward vertical columns even when your category labels are three words long.

Line charts work well with one wrinkle: dates. If your date column is text-formatted (a depressingly common state of affairs), Copilot charts the labels as categories, evenly spaced, without warning you — the classic date-axis failure. Fix the source column to real dates before prompting and the problem disappears.

Combo charts are a pleasant surprise: “monthly actuals as columns, budget as a line” is understood reliably, and Copilot will put a percent series on a secondary axis when units differ. What it won’t do is warn you when a secondary axis is a bad idea — that judgment stays with you.

Pie charts — Copilot will build them on request without pushback, eleven slices and all. It is an obedient tool, not a design conscience. Decide slice count yourself before prompting.

Waterfalls and funnels are where output gets inconsistent. Sometimes you get the native chart type, sometimes an improvised stacked-column approximation, and the difference between runs is not predictable. For these, our step-by-step tutorials remain faster than prompt roulette.

Statistical charts (box plots, histograms with fitted curves, heatmap matrices, correlation pairplots) route through Python, and here Copilot genuinely extends Excel rather than just accelerating it. Say “use Python” explicitly and review the generated code preview — not because it’s usually wrong, but because reading three lines of pandas is how you catch the one time it grouped by the wrong column.

PivotCharts deserve a special mention: ask Copilot to “summarize units sold by state and chart it” on raw transaction data and it will propose a sensible PivotTable-plus-chart combination, fields laid out correctly. For exploratory summarization this is arguably Copilot’s single best charting trick.

When Copilot’s chart goes wrong: a short troubleshooting order

Before assuming the AI is broken, run this sequence — it resolves the large majority of failures:

1. Wrong data charted? Your range isn’t a Table, or has blanks/merged cells splitting it. Ctrl+T the range, re-prompt naming the Table.
2. Series missing? Numbers stored as text don’t chart. Check for the telltale left-aligned “numbers” in the source column.
3. Dates spaced wrongly? Text dates, as above — convert the column, don’t fight the chart.
4. Chart type ignored? Name it more precisely (“100% stacked column,” not “stacked chart”) and issue it as a fresh prompt rather than a correction.
5. Formatting request refused or mangled? You’ve likely hit the 15% Copilot can’t reach — do that step manually. Two minutes in Format Chart Area beats ten minutes of re-prompting.

Agent Mode is a different animal. You give it an outcome, not an instruction — “Build a sales dashboard from the Data sheet with a monthly revenue trend, top-10 customer bar chart, and regional share breakdown, with slicers for year and region” — and it plans the steps, executes them across the workbook, checks its own output, and iterates. You watch the step list in the pane and can redirect mid-run. Microsoft has even added a model switcher (OpenAI’s and Anthropic’s latest models, as of the 2026 GA releases) for different reasoning styles.

Our verdict after months of real use: Agent Mode is the right tool when charts are components of a larger deliverable, and the wrong tool for a single chart. For one chart, its planning overhead is pure waste — Edit with Copilot is faster. For a dashboard scaffold, a multi-sheet report, or a “turn this data dump into something presentable” job, it’s the best thing Microsoft has shipped in years.

Two cautions. Agent Mode makes direct changes without a preview gate, so run it on a copy of anything shared or important — the step list makes changes reviewable, not reversible-by-magic. And it only sees the open workbook; it won’t pull context from your other files.

Python charts: the quiet headline feature

Since the April 2026 update, Copilot can write and execute Python during edits — and this is how it produces chart types Excel simply doesn’t have natively: box plots, violin plots, heatmap matrices, pair plots, network graphs. Ask for one of these (or explicitly say “use Python for this”), and Copilot generates pandas/matplotlib code on a dedicated analysis sheet, shows you the code, and renders the result — your source data untouched.

This matters more than Microsoft’s marketing suggests, because it changes the answer to “can Excel make a ___ chart?” for a whole category of statistical visuals. The trade-off is real, though: Python charts are rendered images, not native Excel chart objects. They don’t restyle with your theme, they don’t respond to slicers, and updating them means re-running the analysis. We cover when that trade is worth making — and when it isn’t — in the full Copilot + Python chart deep dive.

The =COPILOT() function: useful, but not for charts

The =COPILOT() worksheet function puts natural language inside a formula — =COPILOT("classify these product reviews as positive or negative", A2:A200) returns a live, spilled result. It’s genuinely useful for the data preparation stage of charting: classification, extraction, text summarization feeding a range you then chart normally.

It does not create charts, and you shouldn’t try to bend it into doing so. Its role in your charting workflow is upstream: it turns unchartable text into chartable categories.

What Copilot charts well — and what it charts badly

Time to take positions.

Copilot is good at: standard business charts from clean Tables (column, bar, line, combo — reliably correct on the first or second prompt); actual-vs-budget combos; quick exploratory charts where you’re feeling out the data; Python statistical visuals that would otherwise require real coding; and PivotChart suggestions, where its field-layout instincts are better than most intermediate users’.

Copilot is bad at: precise formatting control — it gets you 85% and fights you on the last 15% (axis number formats, exact series colors matching brand palettes, label positioning); charts from messy, un-Tabled data, where it guesses silently; anything requiring judgment about honesty — it will happily truncate a y-axis or use a misleading dual axis if the data shape nudges it that way; and specialty business charts like waterfalls and funnels, where output is inconsistent enough that our step-by-step tutorials remain the faster path to a correct result.

Copilot should never be trusted alone with: final charts for board packs, investor materials, or anything published. Not because it fails often, but because when it fails, it fails plausibly — a wrong series mapping looks exactly like a right one until someone reads the numbers. Review every chart against the source data. Every time.

Limitations Microsoft won’t lead with

• Consistency between sessions is poor outside the Skills framework. The same prompt tomorrow can produce a differently formatted chart. For recurring reports, this is disqualifying on its own — it’s the specific problem the new finance Skills were built to solve.
• Copilot inherits your data’s problems silently. Blank rows truncate series; text-formatted numbers vanish from charts; it rarely warns you.
• Requires connectivity and a saved-to-cloud file. No offline mode.
• Feature names churn quarterly. What this guide calls Edit with Copilot and Agent Mode has already survived several rebrandings (“Allow editing” appears in some builds’ UI). We review this page quarterly; our quarterly charting roundup tracks each shift as it lands.

The bottom line

Use Copilot as your chart drafter: strong prompts on clean Tables, Agent Mode for multi-chart builds, Python for statistical visuals Excel can’t draw. Keep the chart decisions — type, message, honesty — for yourself, and keep final quality control manual. That division of labor is where the real productivity gain lives in 2026, and pretending the AI can own the whole pipeline is how bad charts end up in front of your CFO.

FAQ

Can Copilot in Excel create charts automatically?

Yes. Edit with Copilot inserts native Excel charts from natural-language prompts, Agent Mode builds charts as part of multi-step tasks, and Python in Copilot generates statistical chart types Excel lacks natively. Copilot Chat can only suggest charts for you to insert.

Why won’t Copilot make a chart from my data?

Almost always a data-structure problem: format the range as an Excel Table (Ctrl+T) with a single clean header row, remove merged cells and embedded subtotals, and make sure the file is saved to OneDrive or SharePoint. Then re-prompt with the Table name explicitly.

What’s the difference between Copilot and Agent Mode for charts?

Standard Copilot handles one request at a time — ideal for a single chart. Agent Mode autonomously plans and executes multi-step jobs — ideal when charts are part of a dashboard or full report build. Full comparison: Excel Agent vs Copilot.

Can Copilot make chart types Excel doesn’t have?

Yes, via Python: box plots, violin plots, heatmaps, pair plots, network graphs and more, rendered from matplotlib on a dedicated analysis sheet. The trade-off is they’re static images, not live Excel chart objects.

Should I let Copilot choose my chart type?

No. Its suggestions are serviceable but conservative, and chart choice is the one decision that most determines whether your chart communicates. Decide the type yourself — our chart decision guide makes it a two-minute call — then have Copilot build it.

Related reading

• Which Chart Should You Use? A Decision Guide by Data Type
• Copilot + Python in Excel: AI Chart Generation Deep Dive
• Excel Agent vs Copilot in Excel: What’s the Difference?
• 40 ChatGPT Prompts for Better Excel Charts