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

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.

Our observation: the gap isn't "slides vs. web" as media—it's structured vs. unstructured editing surfaces.

That's why we've been building SEWPOINT (sewpoint.app)—a web app where you paste a Google Slides link and edit native objects (text, tables, charts, shapes) with plain-language instructions. No plugin. Changes stay editable in Slides afterward. We're in early Beta; new signups get 200 free credits, then pay-as-you-go.

One workflow we've been testing: taking a reference slide and mimicking its structure/style onto new content—without flattening everything to images. Example here: Reference slide mimic.

Questions for this community:

• Have you seen AI slide tools that preserve native editability and visual quality? What worked?

• Is the "generate a new deck from scratch" workflow actually what teams need—or is in-place editing on existing decks the bigger pain?

• What would convince you to trust an AI with a deck you'll present to executives or investors?

Curious what's worked (or failed) for you.
admin@sewpoint.app ｜ linkedin.com/company/sewpoint | https://youtu.be/G6JLHsBqlH4