NotebookLM can turn PDFs, meeting recordings, and YouTube videos directly into slide presentations. The free tier supports it. CJK text distortion is the biggest problem, but a “discuss with an LLM first, then generate” workflow keeps the output usable.
What is NotebookLM?
NotebookLM is Google’s AI knowledge tool. Upload data (PDFs, Google Docs, web links, YouTube videos, audio files), and it analyzes the content using Gemini models to produce summaries, Q&A, podcast-style audio, and slide presentations.
The key difference from ChatGPT or Claude is that NotebookLM’s answers are strictly grounded in the uploaded source material. It won’t fabricate information; it pulls from what was provided. This matters for presentations: no AI hallucination ending up in your slides.
The most-used feature is the transcript output (in the Studio panel): drop in a meeting recording, get a detailed verbatim transcript, then pass it to Claude or ChatGPT for further analysis.
NotebookLM Presentation Plans
The Slide Deck feature works on all tiers, with quotas and limits varying:
| Item | Standard | Plus | Pro | Ultra |
|---|---|---|---|---|
| Sources per notebook | 50 | 100 | 300 | 600 |
| Slide Deck / revision quota | Limited | More limits | Higher limits | Highest limits |
| Export formats | Share link, PDF, PPTX | Share link, PDF, PPTX | Share link, PDF, PPTX | Share link, PDF, PPTX |
| Watermark removal | No | No | No | Yes (officially applies to Infographics and Slide Decks) |
Standard is enough for occasional use. Heavy users will want at least Pro for the source ceiling. Watermark removal requires Ultra; Pro doesn’t include it.
4 Steps to Generate a Presentation in NotebookLM
Step 1: Upload source documents
Open a new Notebook, add the materials. Supported formats: PDF, Google Docs, Google Slides, Google Sheets, YouTube links, web URLs, plain text notes, audio files.
Important: don’t upload everything. Pick 5-10 of the most relevant sources. Too many sources distract the AI; the resulting deck loses focus.
The best source quality comes from structured Google Docs (with headers and hierarchy). Well-structured PDFs also work well. YouTube and web pages depend on the content density.
Step 2: Choose Slide Deck in the Studio panel
Open the Studio panel on the right, select “Slide Deck”. Before generating, click the pencil icon for custom settings (many users skip this).
Two format options:
- Detailed Deck: more text, suited for follow-up reading or emailing to colleagues
- Presenter Slides: clean visuals, minimal text, suited for live presentation with verbal narration
For formal presentations, Presenter Slides usually looks better.
In the custom panel, a Prompt can be entered to guide style (“use professional blue tones, max 4 bullets per slide”). Style prompt adherence isn’t perfectly consistent.
Step 3: Wait for generation, then review
Generation takes 30 seconds to 3 minutes depending on source volume. The process runs in the background.
Once generated, review slide by slide. The February 2026 update added single-slide editing: click the pencil icon and instruct changes to a specific slide. This is much better than regenerating the entire deck.
Step 4: Export and refine
Three export options: share link, PDF, PPTX. To edit in Google Slides, the most reliable path is exporting PPTX first and uploading.
Standard refinement flow:
- Generate v1 in NotebookLM
- Export to PDF or PPTX
- Final CJK text, font, and layout fixes in Google Slides or PowerPoint
Advanced Workflow: Discuss with an LLM Before Generating
Throwing data directly into NotebookLM gets a usable but imprecise result. NotebookLM doesn’t strategically organize narratives, it summarizes the data faithfully but doesn’t think about “who is this presentation for, where should the focus go”.
A more reliable flow:
- Feed the data to ChatGPT or Claude first, discuss audience, core arguments, narrative structure
- Have the LLM produce a complete slide outline: per-slide titles, 3-4 key points, visual element suggestions
- Paste the outline into NotebookLM as a source document
- Generate the Slide Deck in the Studio panel, with style requirements added in the prompt
Output narrative logic and visual quality both improve. The LLM does the thinking, NotebookLM does the producing, each plays to strengths.
The same logic applies to meeting notes: iPhone recording → NotebookLM transcript → Claude analysis. NotebookLM’s transcript quality is high enough to compete with dedicated transcription services.
Colored-Pencil Style: Reducing the AI Aesthetic
A useful technique discovered through repeated experimentation: AI-generated images all share recognizable visual cues, over-polished gradients, unnatural lighting, eerily perfect composition.
Adding “colored pencil style, no gradients” to the prompt pushes images toward a hand-drawn feel. Rough paper texture, uneven color blocks, visible brush strokes appear more often. Output looks more like something hand-drawn, less obviously AI.
Why it works: NotebookLM defaults toward clean, corporate, stock-photo aesthetics. Hand-drawn style breaks that polished feel. Combined with “no gradients”, the most telltale AI signs are reduced.
For finer control:
Soft colored pencil on textured paper, no gradients, no digital effectsHand-drawn illustration style, imperfect lines, warm muted tonesSketch note style with colored pencil accents
These style prompts aren’t always fully executed in NotebookLM’s native interface. For precise control, generate images separately and paste them back into the deck.
The CJK Text Distortion Problem
The biggest pain point for non-Latin script presentations in NotebookLM. Generated slides often show CJK characters with missing strokes, fused components, distortions to the point of being unreadable. The longer and more structurally complex the text, the more likely the error.
Workarounds
- Use the “discuss first, then generate” workflow to keep text fed to NotebookLM as short as possible
- Manually correct after export: download PPTX, replace distorted characters with correct text in PowerPoint or Google Slides
- Switch to Google Slides + Gemini sidebar: if text accuracy is the top priority, abandon NotebookLM’s one-click generation and design in Google Slides with Gemini sidebar instead, eliminating the rendering issue
- Consider OCR tools: tools like PDNob Editor can batch-recognize CJK text in images and convert to editable text
Google hasn’t published an official fix timeline for this issue, nor disclosed the underlying image model version powering NotebookLM Slide Decks. Treat any specific model name or percentage claims circulating in communities with caution. This is observation, not an official spec.
NotebookLM vs Gamma vs Google Slides: Quick Comparison
| Item | NotebookLM | Gamma | Google Slides + Gemini |
|---|---|---|---|
| Core strength | One-click generation from existing data | High design freedom, rich templates | Fully editable, accurate CJK rendering |
| CJK support | Poor (text distorts) | Good (text is editable element) | Good (native font rendering) |
| Visual quality | High (polished images) | Medium-high (template-dependent) | Medium (requires manual design) |
| Edit flexibility | Low (exported as images) | High (every element editable) | Highest |
| Best for | Fast research-to-deck conversion | Designing decks from scratch | Precise control over every detail |
| Price | Standard free, Plus/Pro/Ultra by Google AI tier | Free / Plus annual ~$9/mo / Pro $18/mo / Ultra $90/mo | Free / via Google Workspace or Google AI |
A blended approach is most practical. Use NotebookLM to quickly turn large amounts of data into draft decks. For client-facing or accuracy-critical presentations, generate the draft in NotebookLM, then fix slide-by-slide in Google Slides. Gamma comes in for visual-heavy designs.
For all-English decks, NotebookLM’s narrative logic and image selection hold up. For CJK content, the rendered glyph distortion is severe enough to disqualify the deck from formal use without a heavy rebuild in Google Slides or Canva.
FAQ
Can NotebookLM make presentations?
Yes. NotebookLM launched Slide Deck in November 2025. Free tier supports it.
Does NotebookLM mangle CJK text?
Yes. Generated slides and images often show distorted CJK characters. Workaround: outline with an LLM first, fix text in Google Slides after export.
Is NotebookLM free?
Free Standard tier, plus paid Plus/Pro/Ultra. Source limits: 50/100/300/600 by tier.
NotebookLM vs Gamma?
Depends on need. NotebookLM excels at generating from existing materials with high visual quality but CJK distortion. Gamma offers more design freedom and editability. Use NotebookLM for extraction, Gamma or Google Slides for refinement.
Penchan’s Experience
NotebookLM is part of Penchan’s daily workflow, primarily for transcript output. Meeting recordings → upload → transcript → pass to Claude / ChatGPT for analysis. This pipeline runs multiple times a week.
The Slide Deck feature was tested but doesn’t currently fit production work because the Traditional Chinese text distortion is severe enough to break formal presentations. Image selection and narrative logic on the deck side are excellent (a step above Gamma), but the rendering issue forces a Google Slides redo step.
The workaround that holds up: NotebookLM for outline and content extraction only, all visual design redone in Google Slides or Canva. Multi-step but each tool handles its strongest part.
The “colored pencil + no gradient” prompt convention came from iterative testing on cover image generation. Adding it cuts AI-aesthetic considerably for image outputs, even if NotebookLM’s adherence is inconsistent.
Further Reading
- 2026 AI Presentation Tools Comparison: Gamma, NotebookLM, Copilot Hands-On
- Gamma AI Presentation Tutorial
- AI Presentation Prompt Tips
- NotebookLM official site
— Penchan