Back to blog

August 4, 2026

Why LLMs will not fix charting.

Native charting is built for the median user and AI charts die on generation. Why the model companies are unlikely to close that gap, and what we built instead.

Anyone who builds charts for a board deck, an investor update or a monthly report is choosing between two bad options, usually without noticing that a choice is being made. Native charting in Google Slides and PowerPoint gives you a chart that stays alive. You can click it, change the numbers, restyle it, and hand it to a colleague who can do the same. What it does not give you is much to work with. The chart types are basic and the formatting controls run out quickly, which is why so much of the work ends up being done by hand.

Asking an AI gives you the opposite trade. You describe the chart, something close to it appears in seconds, and it often looks better than what you would have built yourself. Then it stops. You cannot select a series and restyle it, you cannot set a brand default, and once the chart leaves the chat it is a picture. Next month the numbers change and you go back and ask again.

Nobody offers both, and it is tempting to read that as a gap the next model release will close. It will not. The shape of this problem comes from who each of these tools was built for.

Why native charting stays limited to basic charts

Native charting is not bad software. It is software built for the median user, and the median user needs a bar chart with a legend. That is a defensible decision, because every option a general purpose tool adds costs surface area, and surface area is the scarcest thing in a product used by hundreds of millions of people. Waterfalls, marimekkos, combo charts with a second axis and axis breaks are features most of those people will never open. Neither is the ability to set an organisation's palette and font as the default so that every chart matches without anyone thinking about it.

The people who need those things are a small share of users, and they were traded away deliberately rather than overlooked. That distinction matters, because it tells you what to expect. A gap created by an oversight closes as soon as someone notices it. A gap created by a decision does not, and as these tools mature they move further toward the middle rather than toward you.

Why AI charts stop being useful after generation

The reasonable hope was that AI would attack the problem from the other side. A model that can write working code should be able to produce a chart that survives a board pack. We went through every route a model has in How to make charts in Claude, and the finding was the same in all of them. What comes back is an output rather than an object.

That is the whole difference. An image, or code that renders an image, cannot be changed by the person holding it. They can only ask for another one. And on the occasions when the chart does land in PowerPoint as a native object, you are back where you started, holding something that is alive but limited.

Why the model companies will not close this gap

The first thing standing in the way is that the current arrangement suits them. A dead chart sends you back to the chat. You ask again with the segments reordered, again with your brand colours, again with the axis starting at zero, and every one of those rounds happens inside their product. An editable chart removes all of them, because you make the change yourself in a few seconds and the model is never involved. Nobody sets out to build the feature that ends their own loop.

The obvious objection is that these companies already work outside the chat window. Claude runs inside Excel and PowerPoint today, and there will be more of that. But those integrations are gateways to the same general model, placed in more locations. They are not a native experience that follows a single object across platforms, which is a different commitment. You have to build it on every surface and then keep it working there while the platforms shift underneath you. Nobody takes that on for a small group of power users.

There is a second reason as well, and it will make more sense once you have seen what we built.

What Chartbuddy builds instead

Our position is that the chart should be the thing that lasts, and that the chat window, the dashboard and the deck are destinations it travels to. That is what the four products are for.

Chartbuddy Embed puts real charts in your AI chat and in HTML. Ask for a waterfall and you get a Chartbuddy chart rather than a picture of one, so you can edit it where it sits and then drag it into your deck. Product teams can ship the same package inside their own applications and give their users the same object. Embed is free.

Chartbuddy Hub is the desktop editor, and it is where the chart lives instead of living inside whichever document it landed in. Generate it from a prompt, edit it, and push it out to Google Slides, PowerPoint, Slack or Notion. When something changes, pull it back from any of them and edit it again. Hub is free for the first thousand users in 2026.

Chartbuddy Slides is a native editor inside Google Slides, with fourteen chart types, full formatting control, and your colours and fonts set once. It is live today.

Chartbuddy PowerPoint brings the same editor to PowerPoint. It launches in September.

Which brings us back to the second reason. Embed is free and models reach for it on their own, so professional charts already arrive in chat without anyone at a model company writing a line of code. They have the capability, their users have the capability, and none of it costs them anything. A problem that is already solved for free does not get solved again.

Charts that stay alive and editable across surfaces

Charting will not be solved inside the chat window, and it will not be solved by native tools growing features they decided against years ago. It gets solved when the chart stops belonging to whatever made it and starts belonging to you.

Related reading

Create professional charts that stay alive across every surface.

Get started