Analyzing metadata

Metadata analysis identifies which metadata fields meaningfully segment your qualitative data. For example, it can reveal whether participants from different age groups, regions, or roles emphasise different themes in their responses.

Running metadata analysis

To run a metadata analysis:

1. Navigate to the categories view of your project.
2. Click Analyze metadata.
3. Skimle examines all your metadata fields and determines which ones produce meaningful differences across your insight categories.

The analysis runs in the background and typically takes a few minutes. Results appear in the categories view once complete.

How it works

Skimle combines two complementary approaches to find significant patterns:

• Semantic analysis — compares the meaning of insights across metadata groups using text embeddings. If participants with different metadata values discuss genuinely different things, this signal is detected.
• Distribution analysis — uses statistical tests to check whether certain metadata values cluster in specific subcategories more than would be expected by chance.

A metadata field is considered significant only when both signals agree, reducing the risk of spurious findings.

Reading the results

For each insight category, the categories view shows:

• Which metadata fields produce meaningful segmentation, ranked by significance.
• Group descriptions — for each significant field, Skimle describes what characterises each metadata group's responses. For example, under an "Age group" field, you might see that younger participants focused on usability while older participants prioritised reliability.
• Temporal comparison — if your documents have creation dates, Skimle automatically compares recent and earlier time periods to highlight how themes have shifted over time.

Requirements

For metadata analysis to produce useful results, your project needs:

• At least 15 insights in the categories being analysed.
• At least two distinct values for the metadata field being tested.
• At least two insights per metadata value.

Fields that do not meet these thresholds are skipped automatically.

Tips for effective metadata analysis

• Use descriptive metadata values. "Senior developer" is more useful than "Role A" because it helps the AI write more meaningful group descriptions.
• Keep the number of unique values manageable. Fields with dozens of unique values are harder to interpret. Consider grouping values (e.g. age ranges instead of exact ages).
• Add metadata before running analysis. Metadata analysis uses whatever fields exist at the time it runs. If you add new fields later, run the analysis again to include them.