When a generated slide comes out badly, people tend to reach for one of two explanations. The first is “Copilot isn’t good enough.” The second is “the user wrote a bad prompt.” Sometimes one of them is right. Often neither is, and the real cause — a missing example slide, a placeholder sized wrong, a rule nobody wrote down — goes unfixed and repeats in thousands of decks.
Diagnosis matters because each cause has a different fix, owned by a different person. This post gives a practical way to sort failures by cause.
The categories
We sort generated-slide failures into nine types. Each has a letter so they can be tallied quickly during review.
| Code | Failure | Typical symptom |
|---|---|---|
| T | Technical | Text overflows, image cropped, content in the wrong placeholder |
| B | Brand | Wrong colors, fonts, imagery or visual language |
| H | Hierarchy | The important thing isn’t the most visible thing |
| S | Structure | Wrong way to communicate the content — e.g. bullets where a comparison was needed |
| D | Density | Too much or too little for the layout |
| V | Visualization | Poor chart choice or poorly presented data |
| I | Intent | Deck doesn’t achieve what it was meant to |
| P | Prompt | The instruction was insufficient or unclear |
| A | AI constraint | A limit of the current model that no template change fixes |
Where each usually comes from
Each failure type tends to point to a particular part of the system.
T — Technical → layouts and placeholders. Microsoft’s guidance says Copilot uses placeholder type, position and size to decide what goes where. Overflow usually means a placeholder sized for idealized content; cropping usually means proportions or overlapping decoration; content in the wrong slot often means text boxes where placeholders should be. Owner: template builder.
B — Brand → theme, styles or Brand Kit. Default colors on a new chart point to a theme that doesn’t hold the brand. Wrong images point to the Brand Kit or asset library. Wrong fonts point to theme fonts. Owner: template builder and brand manager.
H — Hierarchy → example slides. Copilot learns emphasis from examples. If generated slides consistently bury the key point, the examples may not show strong hierarchy — or may not exist for that slide type. Owner: designers.
S — Structure → missing layouts or examples, sometimes rules. Content forced into the nearest available layout suggests the right layout doesn’t exist, or has no example showing its purpose. With strict brand adherence on, Copilot can’t create a new layout to compensate. Owner: template builder and designers.
D — Density → examples and rules. If the template shows only one density per slide type, Copilot applies it to everything. Microsoft recommends examples at light, medium and heavy. Note instructions (“Keep the text minimal”) and Brand Skills can reinforce limits. Owner: designers and brand manager.
V — Visualization → theme, styles and data examples. Chart colors come from theme accents; chart appearance from styles; chart choices and takeaway titles from data examples and rules. Owner: designers.
I — Intent → prompt, source or person. A deck that is well-formed but misses the point usually lacked information about audience and purpose, or its outline wasn’t reviewed before slides were generated. This is rarely a template problem. Owner: the person creating the deck.
P — Prompt → training and rules. If the same prompt weakness recurs across many people, it’s worth addressing in guidance for employees — and sometimes in a Brand Skill that sets sensible defaults so a thin prompt still produces a reasonable structure. Owner: brand manager and whoever runs enablement.
A — AI constraint → acceptance, review, or escalation. Some failures persist whatever you change. Microsoft is clear that Copilot’s output should be reviewed by a person. Recognizing a genuine model limit prevents endless template tinkering that won’t help. Owner: review process; revisit when Microsoft updates the product.
A diagnostic routine
For each failed slide:
- Name the symptom — what exactly is wrong?
- Assign a code. If more than one applies, pick the most fundamental. A structure failure often causes density and hierarchy failures downstream.
- Check the prompt and source first. Rerun the same request with a clearer prompt. If that fixes it, it’s P or I, not the template.
- Check the example set. Is there a finished example for this kind of slide at this density? If not, that’s the likely cause of H, S, D or V.
- Check the layout. Is there a layout for this job, with correctly typed and sized placeholders? If not, it’s S or T.
- Check the theme and kit. For any B or V failure involving colors, fonts or images.
- Try the smallest system change that would prevent the failure, and rerun the full prompt set.
- If nothing helps, classify as A and make sure review catches it.
Count the failures
The real value comes from tallying. After reviewing a meaningful batch of output — a full run of your test prompt set, or a sample of real generated decks — count failures by code. The distribution tells you where to invest:
- Mostly T, S and D? The template’s structure and examples need work. This is usually the highest-leverage fix, because it improves every deck.
- Mostly B and V? Theme, styles and Brand Kit.
- Mostly P and I? Employee guidance, outline review habits and sensible defaults in Brand Skills.
- A significant share of A? Set expectations, strengthen review, and retest when Microsoft ships changes.
We don’t publish a “typical” distribution, because it depends heavily on the template and the organization — and because the point is to learn yours. Count, fix the largest category, rerun, count again.
Key takeaways
- Different causes need different fixes, owned by different people; diagnose before fixing.
- Sort failures into nine types: technical, brand, hierarchy, structure, density, visualization, intent, prompt and AI constraint.
- Check the prompt first, then examples, then layouts, then theme and kit.
- Tally failures by type to see where to invest, and fix the system rather than the slide.
- Recognize genuine model limits, and make sure review catches them.
Want to know what’s behind your generated decks’ problems? Have your template scored.**
Related reading
- How to test a Copilot template before your company uses it
- Where Copilot reads your brand from: a map of the five inputs
- Clean up a bloated template before Copilot learns from it
Sources
- Microsoft Support, Keep your presentation on-brand with Copilot (July 2026)
- Microsoft Support, Create a new presentation with Copilot in PowerPoint (June 2026)
- Microsoft Support, Manage template settings in your Brand Kit for Copilot in PowerPoint (August 2026)