How to set up AI tag matching

AI tag matching automatically assigns existing tags from your taxonomy to quotes when you import data through Data Hub. Instead of tagging every quote by hand, the AI reads each quote and applies the tags from your taxonomy that best match its content. Use this article to set it up and get the most accurate results from it.

Before you start

• AI tag matching only works with tags that already exist in your workspace. It never creates new tags.
• Each tag group needs a description before AI matching can be enabled for it. The AI uses that description to understand what the group is about, so without one, matching can't be turned on for that group.

Steps

1. Go to Taxonomy > Tag groups.
2. Turn on Enable AI to auto-assign existing tags at the top of the page.
3. Open the tag group you want the AI to match and add a description. Describe what the group represents and what each tag value means. For example: "Classifies the type of friction a customer encountered. Use 'Onboarding' for issues during initial setup, 'Billing' for payment and pricing complaints, 'Performance' for speed or reliability issues, and 'Support' for problems reaching the team."
4. Once you save a description, check Assign tags with AI next to that tag group.
5. Go to Data Hub and import your file as usual. Any columns you leave unmapped during the CSV mapping step are ignored automatically, so you don't need to set them to Ignore one by one.
6. When the import finishes, AI tag matching runs on the new quotes and assigns tags from any tag groups you enabled.

Tips

What makes a tag group a good fit

• Keep tags semantically clear. A good tag is one you could guess from reading a quote with no extra context, like Sentiment (Positive / Neutral / Negative), Product area (Search / Onboarding / Reporting / Integrations), or Feedback type (Feature request / Bug / Compliment / Confusion).
• Keep tag groups focused and small. Fewer tag options means better accuracy. Choosing between 5 tags is a different task for the AI than choosing between 50.
• Use human-readable tag names. The AI only works with the text of your quotes and your tag names. It has no access to workspace history or internal systems, so tags like "BRAIN-2" or internal ticket codes won't match reliably.

What to avoid

• Tags based on who said something rather than what they said, such as Stakeholder: Employee / Customer / SMB / Enterprise. The AI has to infer this from the quote text and will often guess the most common-sounding option.
• Ambiguous or overlapping tag values, especially where telling two tags apart needs internal context the AI doesn't have.
• Large tag groups. Accuracy drops once a group passes about 20 tags. If a group keeps producing wrong tags, try splitting it into smaller, narrower groups.

Getting better results

• Describe tag values, not just the group. "Use 'Search' for quotes about finding content, 'Onboarding' for first-run experience, 'Reporting' for dashboards and exports" works better than a bare group name like "Product area."
• Test on a small import first. Upload a sample of 20 to 30 quotes and check the results in the Quotes table before running a full dataset.
• Only enable AI matching where semantic matching genuinely makes sense. Quality matters more than coverage.

Good to know

• AI tag matching runs on new quotes at import time. It won't change or overwrite tags you've already assigned manually.
• If nothing in a tag group is a good fit for a quote, the AI leaves that quote untagged for that group rather than guessing.
• If AI matching keeps applying the same tag to everything, check whether the tag group has a description, whether its tag values overlap, or whether the tags depend on information that isn't present in the quote text, like a speaker's role.
• AI-assigned tags currently look the same as manually assigned tags on a quote. Distinguishing between them isn't available yet.