Something I've noticed about AI slide tools: they all lead with the same pitch design from scratch / pick a template / make deck beautiful. The hero moment is always "look, a gorgeous slide appeared."

Fair enough. But for analysts, RevOps, and BI folks, the real grind is recurring: new numbers com in, reuse or refer to existing deck style, same charts types you just need data refreshed and insights rewritten across multiple slides. Low tolerance for error. You're not reinventing the deck; you're trying not to break it.

For Google Slides, most people still bounce between Sheets and Slides, re-wiring charts and fixing drifted formatting. The category barely talks about this maybe because "generate a stunning deck" demos better in 30 seconds than "update 12 slides without breaking chart formatting."

Why do AI-generated slides still feel worse than AI-generated web pages?

I've been thinking about this a lot while building in the presentation space.

When I ask an LLM to generate a landing page, the output is usually structurally coherent: headings, sections, hierarchy, readable typography. HTML/DOM gives the model something semantic to work with.

Slides are different. Under the hood, a slide deck is often a bag of objects placed at absolute coordinates text boxes, shapes, charts with weak structural semantics. Many AI slide tools respond by exporting images or rebuilding whole pages. That can look fine in a screenshot, but you lose editability and the result often falls apart when you need precise changes.