Improving Auto-Tagging Accuracy in Dovetail for SME Insights

Julia K. · 2025-09-17 17:42:59.898752+00

Hi fellow Dovetailers, We use Dovetail to derive insights from SME interviews during a large conversion project. We've set up a solid tagging structure (following Dovetail's ) and are exploring the auto-tagging feature to save time. Our first attempt yielded about 75% accuracy. We're considering manually tagging a few interviews to see if this helps improve auto-tagging accuracy. Has anyone tried this approach, or have tips on getting better results? Appreciate your input! Julia

Dovetail Champions

Improving Auto-Tagging Accuracy in Dovetail for SME Insights | Dovetail Champions

9 comments

· Sorted by Oldest

Jazmin
·
·
Hi Julia, Thanks for reaching out. 👋 Happy to help out! We recommend creating tags and their descriptions in the project before requesting suggested highlights—this ensures they can be applied more accurately. Suggested highlights work best when there’s a smaller set of clearly defined tags, since the system can focus more accurately rather than trying to cover an extensive tag board. At the project level, you can choose which tag groups within a tag board should be used for suggested highlights (see screenshots below).
(89 kB)
For the best results, I recommend enabling only a few key tag groups so the suggested tags remain accurate and relevant. You can also refer to this help center article: https://docs.dovetail.com/academy/capture-themes-with-tags#making-the-most-of-tags-with-descriptions
Julia K.
·
·
thank you, Jazmin! We followed Dovetail recommendations on creating tags pretty closely, and i think it makes a difference. My question was more regarding using manual tagging as a way to "train" the system to apply the tags in a way we envisioned it, and end up with better, focused highlights in the end. Would love to hear from someone who followed a very specific Dovetail recommendation from the paper above (see screenshot), and noticed improvement in the quality of the highlighted and tagged data. Sincerely - Julia
Loading attachments
Jazmin
·
·
Hi Julia, That's correct, you can improve the quality of your magic highlights by: ** highlighting and tagging some examples of the kind of content you want to be magically highlighted/tagged ** writing good tag descriptions to help guide what you are looking to find. Over time, it should improve as it will learn based on the accepted highlights and your own manual highlights. 😊
Julia K.
·
·
Ah! That sounds right - i think i saw it in one of the documents/guides. My questions is - is tagging of my data allows system to gain a very specific learning about me as a client, or contributes into the pool of general information relevant to all Dovetail customers? Perhaps, in other words, Is the conversation between me and Dovetail mediated by AI channeled by "client id" (me), or "provider id" (Dovetail)?
Jazmin
·
·
We are not using any customer data to train any of our AI models. You can read about our product-specific terms in our help center at the link here. Our MSA is also located in our help center here for your team's review. Lastly, all security, privacy, and compliance policies and documentation are listed on our Trust Center.
Julia K.
·
·
Thanks! I might be confused with what "Provide examples Manually tagging a few examples gives the AI a reference point to learn from" means. Does it mean if i create tags some interviews and run "suggest highlights" process on others the AI will use examples of tagging from tagged interviews and apply those to untagged ones?
Jazmin
·
·
Hey Julia, To clarify: ** We do not use any customer data to train our AI models. Nothing you tag or highlight is used to retrain or update the underlying model. ** What happens instead is that, when you run Suggested highlights inside a project, the system looks at the highlights and tags that already exist in that project only as reference points. This gives the AI more context so it can suggest highlights that are consistent with the examples you’ve already created. Think of it less like the AI “learning” permanently, and more like it’s “looking at your examples” in real time to guide the suggestions. Does that make sense?
Julia K.
·
·
Thank you Jazmin, That certainly does. That in particular "when you run Suggested highlights inside a project, the system looks at the highlights and tags that already exist in that project only as reference points". Appreciate your input!
Jazmin
·
·
No worries! Happy to help out 😊 Have a great weekend.

Jazmin
·
·
Hi Julia, Thanks for reaching out. 👋 Happy to help out! We recommend creating tags and their descriptions in the project before requesting suggested highlights—this ensures they can be applied more accurately. Suggested highlights work best when there’s a smaller set of clearly defined tags, since the system can focus more accurately rather than trying to cover an extensive tag board. At the project level, you can choose which tag groups within a tag board should be used for suggested highlights (see screenshots below).
(89 kB)
For the best results, I recommend enabling only a few key tag groups so the suggested tags remain accurate and relevant. You can also refer to this help center article: https://docs.dovetail.com/academy/capture-themes-with-tags#making-the-most-of-tags-with-descriptions
Julia K.
·
·
thank you, Jazmin! We followed Dovetail recommendations on creating tags pretty closely, and i think it makes a difference. My question was more regarding using manual tagging as a way to "train" the system to apply the tags in a way we envisioned it, and end up with better, focused highlights in the end. Would love to hear from someone who followed a very specific Dovetail recommendation from the paper above (see screenshot), and noticed improvement in the quality of the highlighted and tagged data. Sincerely - Julia
Loading attachments
Jazmin
·
·
Hi Julia, That's correct, you can improve the quality of your magic highlights by: ** highlighting and tagging some examples of the kind of content you want to be magically highlighted/tagged ** writing good tag descriptions to help guide what you are looking to find. Over time, it should improve as it will learn based on the accepted highlights and your own manual highlights. 😊
Julia K.
·
·
Ah! That sounds right - i think i saw it in one of the documents/guides. My questions is - is tagging of my data allows system to gain a very specific learning about me as a client, or contributes into the pool of general information relevant to all Dovetail customers? Perhaps, in other words, Is the conversation between me and Dovetail mediated by AI channeled by "client id" (me), or "provider id" (Dovetail)?
Jazmin
·
·
We are not using any customer data to train any of our AI models. You can read about our product-specific terms in our help center at the link here. Our MSA is also located in our help center here for your team's review. Lastly, all security, privacy, and compliance policies and documentation are listed on our Trust Center.
Julia K.
·
·
Thanks! I might be confused with what "Provide examples Manually tagging a few examples gives the AI a reference point to learn from" means. Does it mean if i create tags some interviews and run "suggest highlights" process on others the AI will use examples of tagging from tagged interviews and apply those to untagged ones?
Jazmin
·
·
Hey Julia, To clarify: ** We do not use any customer data to train our AI models. Nothing you tag or highlight is used to retrain or update the underlying model. ** What happens instead is that, when you run Suggested highlights inside a project, the system looks at the highlights and tags that already exist in that project only as reference points. This gives the AI more context so it can suggest highlights that are consistent with the examples you’ve already created. Think of it less like the AI “learning” permanently, and more like it’s “looking at your examples” in real time to guide the suggestions. Does that make sense?
Julia K.
·
·
Thank you Jazmin, That certainly does. That in particular "when you run Suggested highlights inside a project, the system looks at the highlights and tags that already exist in that project only as reference points". Appreciate your input!
Jazmin
·
·
No worries! Happy to help out 😊 Have a great weekend.