Why AI presentation tools export flat images, and how to get editable slides

Before any of the theory, run this on a deck you already have. It settles the question faster than any feature page will.

1. 1

   Open the exported .pptx in PowerPoint or Keynote

   Not in a preview pane, not in the tool that made it. The whole question is what a different program sees when it opens the file.

2. 2

   Click once on a headline

   Then look at what got selected. If a box appears tightly around the words, and clicking again puts a blinking cursor in the text, that is a real text shape. If instead a box appears around the entire slide, edge to edge, you have selected one picture that happens to have words printed on it.

3. 3

   Press Ctrl+A on a slide, then look at the status bar

   A native slide reports several objects selected. A flattened slide reports one.

4. 4

   Try the outline view

   View, then Outline View on Windows. Real slides show their titles and bullet text as an editable outline. A flattened deck shows a list of empty slides, because as far as PowerPoint is concerned there is no text in the file at all.

A .pptx is a zip archive of XML files. Each slide is described in its own XML document as a tree of shapes, and every shape carries its own position, size, fill, and, if it holds text, the actual characters along with the formatting runs that style them. Any program can write that XML. What differs is how much of the work a tool is willing to do.

The reason the first two exist is not laziness. A tool that designs slides in a browser has already solved layout in CSS, where text wraps according to a font stack, a line-height and a container width. PowerPoint's text layout engine is a different engine with different metrics. Hand the same paragraph to both and the line breaks will not land in the same place. Flattening makes that problem disappear, because a picture of a correct layout is always a correct layout.

Emitting a native shape tree means committing to solve that properly: measuring the text where it was rendered, converting those measurements into PowerPoint's coordinate system, and accepting that if the maths is wrong, the customer sees an overlapping headline rather than a slightly soft image.

Sources: [1] [2]

The cost is invisible at the moment of export, which is exactly why it catches people. Everything below is a consequence of the same fact: there is no text in the file.

You cannot fix a typo

Not with an edit, anyway. The fix is to go back to the generating tool, correct it there, re-export, and redo any change you made in PowerPoint since. If the tool has since changed a layout, the rest of the deck moves too.

Find and replace does nothing

Rename a product, update a date across 40 slides, swap a client's name in a proposal template: none of it is possible. There is nothing to find.

Spell check stays silent

A flattened deck always passes, which is worse than failing. Nobody catches the typo in the 60pt headline until it is on a projector.

Screen readers get nothing

Microsoft's own accessibility guidance is explicit that important information should not live only inside an image, and that text belongs in real placeholders so reading order works. A flattened slide fails both. If you present in a public sector, education or enterprise context with an accessibility policy, this can be disqualifying rather than merely awkward.

Your brand team cannot use it

Corporate template compliance, theme colours, the approved font: all of it is applied to shapes. A picture ignores every one of them.

Translation is off the table

Both PowerPoint's own translation and every third-party localisation workflow read text runs. Images are skipped in silence.

The file is bigger, often much bigger

A slide's worth of text and vector shapes is a few kilobytes of XML. A full-bleed rendering of the same slide at a resolution high enough to survive a projector is measured in megabytes. Multiply by 30 slides.

Sources: [3] [4]

Plenty of tools land between the extremes, and some document it honestly. Gamma's export help page, for example, sets out which effects fall back in a .pptx and notes a caveat about italics, which is more candour than most feature pages offer. Treat that kind of documentation as a good sign rather than a red flag: a vendor who tells you where the export degrades has thought about the export.

Where partial output is genuinely fine: the text is native, and only decorative elements, a background treatment or an unusual gradient, come through as images. You can still fix a number, translate the deck and pass the accessibility checker.

Where it is not fine: the charts are pictures. Charts are the thing most likely to need a last-minute update, and a chart that arrives as an image has also lost its underlying data, so nobody downstream can check the numbers or re-cut them.

• Native text, image backgrounds: usually acceptable.
• Native text, image charts: workable, but budget time to rebuild any chart that changes.
• Image text: treat the export as a PDF that happens to end in .pptx.

Sources: [6]

Marketing copy is not a reliable signal here, because "fully editable PowerPoint export" is written by tools at all three points on the spectrum. Check the artefact instead.

1. 1

   Export a throwaway deck on day one

   Before you build anything real. Generate five slides including a chart and a table, export, and run the test at the top of this page.

2. 2

   Open it somewhere the vendor did not choose

   PowerPoint desktop, PowerPoint for the web, Keynote and Google Slides all behave differently. If your audience opens decks in Google Slides, test there.

3. 3

   Check the fonts survived

   Open the deck on a machine that does not have your brand font installed. Substitution is the second most common way a deck breaks in transit, and it is a separate problem from flattening.

4. 4

   Look at the file size

   A 30 slide text-and-chart deck that weighs 60 MB is telling you what is inside it.

5. 5

   Run the accessibility checker

   Review, then Check Accessibility. "Missing alt text" on every slide, and nothing else flagged, is the signature of a flattened deck.

Sources: [3]

If you are choosing between tools, it helps to know what the harder path actually requires, because it tells you what to ask about.

• Geometry that comes from measurement, not estimation. Where a text block ends up depends on the font's metrics at that size in that width. A tool that guesses will overlap eventually.
• A coordinate conversion that does not drift. PowerPoint positions everything in EMUs, integers at 914,400 to the inch. Rounding in the wrong place accumulates across a slide.
• Real tables and real charts, not pictures of them. This is the part most often quietly skipped.
• Fonts embedded in the file, so the deck survives the trip to a machine that does not have them, plus a sensible fallback for the environments that ignore embedded fonts.
• A verification pass. Once slides are generated rather than hand-placed, the only way to know a headline did not collide with an image is to check the output and regenerate what fails.

That last point is the difference between a generator and a generator you can trust with a deck you have not looked at. Flattening removes the need for it, which is precisely why flattening is so common.

Sources: [2] [5]