Skip to main content
Auto Tagger learns from your existing tagging patterns. The better your input data, the better the suggestions. Here are proven strategies for getting the most out of the plugin.

​
Building a Strong Foundation

​
Start with Manual Tagging

The plugin needs examples to learn from. Before expecting good suggestions:
1

Tag your existing notes

Go through your vault and add tags to at least 20-30 notes manually. Focus on notes you consider well-organized or representative of your content.
2

Use each tag multiple times

Remember: tags must appear in at least 2 documents to be suggested (see src/model.ts:119). Aim for 3-5 examples minimum for reliable suggestions.
3

Rescan after bulk tagging

Use the “Rescan vault” command (or button in settings) after tagging multiple notes so the model learns from your new patterns.
The plugin only learns from notes that already have tags. Untagged notes are counted in statistics but don’t influence the model.

​
Be Consistent

Consistency is more important than quantity.
Pick one tag format and stick with it:
#machine-learning  ✅ (always use hyphens)
#data-science      ✅
#python            ✅
Use tags for similar content:
  • Tag all Python tutorials with #python
  • Tag all project planning notes with #planning
  • Tag all meeting notes with #meeting
Tag names are case-sensitive and format-sensitive. #Python and #python are different tags to the plugin.

​
Optimizing Tag Quality

​
Tag Granularity

Balance between too broad and too specific: Too broad:
  • #notes (applies to everything)
  • #stuff
  • #important
Too specific:
  • #python-pandas-dataframe-merge-tutorial-2024
  • #meeting-with-john-about-project-x
Just right:
  • #python + #pandas
  • #meeting + #project-x
The TF-IDF algorithm automatically weights distinctive words higher (see src/model.ts:294-301). Broad tags that appear in many different contexts will naturally have lower, more diffuse vectors.

​
Use Tag Hierarchies

Obsidian supports nested tags with forward slashes. The plugin treats these as single tags: 
#programming/python
#programming/javascript
#work/projects/alpha
Benefits:
  • Organize related tags hierarchically
  • The plugin learns associations at each level
  • You can use Obsidian’s tag pane to collapse/expand categories
Co-occurrence boost example: If you consistently use #programming/python together with #data/science, the plugin will learn this pattern and boost #data/science when it sees #programming/python in a note (from src/model.ts:133-142).

​
Content Quality Matters

The plugin analyzes note content, not just titles:
Strong signals:
  • Descriptive body text (at least 50-100 words)
  • Technical terms and domain-specific vocabulary
  • Consistent terminology across similar notes
Weak signals:
  • Very short notes (< 20 words)
  • Notes with mostly links and no prose
  • Notes that are just lists of tasks
  • Generic placeholder text
Remember: The tokenizer filters out:
  • Stop words (“the”, “and”, “is”, etc.)
  • Words shorter than 3 characters
  • Pure numbers
  • Code blocks and inline code
  • Frontmatter (except tags)
See src/model.ts:197-217 for full tokenization logic.

​
Working with Suggestions

​
Understanding the Suggestion Notice

When a suggestion appears, you have several options (from src/main.ts:183-255):
1

Add

Accept the suggestion. The tag is inserted according to your “Tag placement” setting, and the next suggestion appears.
2

Skip

Reject this suggestion but continue to the next one. The plugin won’t suggest this tag again for the current note (even after edits).
3

Skip all

Dismiss all pending suggestions for this note. Only appears when there are multiple suggestions.
4

Show all

Open a modal with all suggestions (up to 2× your “Max suggestions” setting). You can select multiple tags to add at once. Top 3 are pre-selected.
Skipped suggestions are tracked per file. If you close and reopen the note, you’ll see suggestions again (from src/main.ts:38).

​
When to Manually Trigger Suggestions

Use the command palette command “Suggest tags for current note” when:
  • You’ve just written a lot of content and want to review all suggestions
  • You have “Auto-suggest while editing” disabled
  • You want to re-evaluate tags after skipping some
  • You want to see the full modal view immediately
This command bypasses the debounce delay and clears the skip cache (from src/main.ts:96-104).

​
Tuning for Your Vault

​
Small Vaults (under 100 notes)

Challenges:
  • Limited training data
  • Fewer tag examples
  • More variance in suggestions
Recommended settings: 
Check delay: 1500-2000ms
Max suggestions: 3
Minimum confidence: 0.02-0.03
Tips:
  • Be extra consistent with tagging
  • Use fewer, more general tags
  • Manually tag at least 30-40 notes
  • Accept that suggestions will improve over time

​
Medium Vaults (100-500 notes)

Optimal conditions for the plugin:
  • Enough data to learn patterns
  • Not too large to slow down processing
Recommended settings: 
Check delay: 2000ms (default)
Max suggestions: 5 (default)
Minimum confidence: 0.01 (default)
Tips:
  • Experiment with co-occurrence by using related tags together
  • Develop a tag hierarchy for your main topics
  • Rescan after adding 20+ tagged notes

​
Large Vaults (500+ notes)

Considerations:
  • Initial scan takes longer (but only happens once)
  • More unique words and tags in the model
  • Very robust suggestions if tagging is consistent
Recommended settings: 
Check delay: 2500-3000ms
Max suggestions: 5-7
Minimum confidence: 0.005-0.01
Tips:
  • Consider disabling auto-suggest if you notice slowdowns
  • Use manual trigger instead for better control
  • Be patient during initial scan (may take 30-60 seconds)
  • Higher confidence thresholds can reduce noise

​
Specialized Vaults

Characteristics:
  • Technical terminology
  • Citation-heavy notes
  • Topic clusters
Tips:
  • Tag by methodology (#qualitative, #quantitative)
  • Tag by field/subfield
  • Use co-occurrence for paper-topic relationships
  • The plugin excels with technical jargon

​
Troubleshooting Poor Suggestions

​
No Suggestions Appearing

1

Check if model is ready

Look at the settings page. Do you see model statistics? If not, wait for the initial scan to complete or trigger a manual rescan.
2

Verify minimum requirements

  • Is your note longer than 3 words after tokenization?
  • Do you have at least 2 documents with some tags?
  • Are you below the minimum confidence threshold?
3

Check auto-suggest setting

Is “Auto-suggest while editing” enabled? If not, use the command palette to manually trigger.
4

Review existing tags

The plugin won’t suggest tags already in the note. Check your frontmatter and inline tags.

​
Suggestions Are Not Relevant

Possible causes and fixes:
Symptom: Random or unexpected suggestionsSolution: Review how you’ve tagged similar notes. Are you using the same tags for the same topics? Standardize your tagging and rescan.
Symptom: Many weak, barely-related suggestionsSolution: Increase “Minimum confidence” to 0.02-0.05 in settings. This filters out low-quality matches.
Symptom: Suggestions don’t match the specific topicSolution: Add more distinctive, topic-specific words to your note. Generic language produces generic suggestions.
Symptom: Old tags suggested even though you’ve changed your systemSolution: Click “Rescan vault” in settings to rebuild the model with current data.

​
Performance Issues

If you experience lag or freezing, try these steps:
  1. Increase check delay to 3000-5000ms
  2. Disable auto-suggest and use manual triggering instead
  3. Reduce max suggestions to 3
  4. Check vault size - initial scan of 1000+ notes may take time
The plugin processes files in batches and yields to the UI (from src/model.ts:63-76), but very large vaults may still experience brief pauses.

​
Advanced Patterns

​
Using Multiple Tag Strategies

Combine different tag types: 
---
tags:
  - type/article
  - status/published
---

#python #machine-learning #tutorial

Content here...

Section about pandas specifically. #pandas
  • Frontmatter: Metadata and status tags
  • First line: Primary topic tags
  • Inline: Contextual, paragraph-specific tags
You can change “Tag placement” in settings based on what you’re adding.

​
Leveraging Co-occurrence

Explicitly train tag relationships:
  1. Identify tags that should appear together (e.g., #project-x + #planning)
  2. Manually add both tags to several related notes
  3. Rescan vault
  4. The plugin will automatically boost #planning when it sees #project-x
The boost is proportional to how often they co-occur (from src/model.ts:140-141).

​
Periodic Maintenance

Set a reminder to:
  1. Monthly: Review and standardize tags using Obsidian’s tag pane
  2. After bulk imports: Always rescan
  3. Quarterly: Audit tag consistency across your most important notes
  4. When suggestions degrade: Rescan and check for tag drift

​
Getting Help

If suggestions still aren’t working well after following these tips:
  1. Check the console (Ctrl+Shift+I / Cmd+Opt+I) for error messages
  2. Verify your vault has at least 2 tagged documents with each tag
  3. Try the “Suggest tags for current note” command on a well-tagged note to verify the model is working
  4. Consider starting with default settings and adjusting one at a time
Remember: The plugin is only as good as your training data. Garbage in, garbage out. Consistent, meaningful tagging produces excellent suggestions.

Build docs developers (and LLMs) love

Get started for freeTalk to us