Skip to main content
This guide covers how your team can use Channels — Dovetail’s home for high-volume, continuous feedback. It walks through what data belongs in a channel versus a project, how to create one and bring data in, how to write context so Dovetail organizes the feedback the way you need, how to read trends, and how to analyze with chat, agents, and dashboards.

​
1. What belongs in a channel (vs. a project)

To put it simply, a project holds rich, qualitative data tied to a specific topic or initiative — interviews, customer calls, a one-off survey. A channel holds the always-on feedback that keeps arriving day over day, week over week, month over month, and lets you watch how the themes change over time. The test for a channel is continuous, high-volume feedback on the same subject. If the same kind of data keeps coming in and you care about how it trends over time, it’s a channel. Some examples: NPS/CSAT, support tickets, app and store reviews, etc. A few guidelines:
  • You need at least 200 data points before a channel will generate themes. A low-volume source (say, ten responses a month) is too small — that data belongs in a project.
  • One channel per unique data source. Don’t create three channels for the same app reviews, and don’t mix fundamentally different motions (sales calls and support tickets) in one channel. Dovetail has no way to know two channels are duplicates, so overlapping data inflates your counts and undermines the single source of truth. It also keeps each channel’s context and themes refined to one kind of data. If you need different filtered views of the same channel, use a dashboard rather than a second channel.
  • Surveys fit only if they recur. A post-enrollment or NPS survey that runs continuously on the same questions is a great channel. A one-time survey is not — the whole value of a channel is seeing change over time. And the questions have to stay consistent; a survey that asks about something completely different each period won’t work.
A note on quantitative data: channels analyze written, qualitative text. A row with only a numeric score and no comment can’t be classified.

​
2. Creating a channel and bringing data in

You can create a channel from the main menu (New → Channel); set the destination folder while you’re there, since it’s easier than moving it later.

​
Import options

  • CSV upload — always works, and it’s a perfectly good way to start. Many teams begin with a CSV and connect a live integration on top of the same channel later.
  • Native integrations — connect a source so data flows in automatically (for example, support-ticket tools). Note that some integrations are scoped by design — a support integration may pull only closed tickets, and a survey tool’s integration may cover tickets but not its surveys.
  • Zapier — a connector that bridges tools without a native integration (for example, a survey platform): “when there’s a new response, add a data point to this channel.” It’s a paid third-party tool.
  • Open API — build and manage the connection yourself.

​
CSV example

After you import a CSV, you tell Dovetail two essential things:
  1. Which column to analyze — the single open-text column (the comment, the “why did you give that score,” the ticket body). Note that channels can only analyze one qualitative column. If you have two free-text columns you want included, combine them into one column before uploading, and point Dovetail at the merged column.
  2. Which column is the date — this is what lets Dovetail plot everything over time.
Every other column can optionally become a field (metadata you filter by) via per-column toggles. You don’t have to import every column — leave out anything you don’t need. For native integrations, refer to the specific integration documentation for guidance.

​
Field best practices

Fields are what make filtering useful later, so set them up deliberately:
  • Use dropdown fields, not free text or raw numbers when you can. Free-text fields make filtering unreliable because values vary by spelling and abbreviation (“English” versus “ENG” won’t match). A dropdown gives you the actual options to pick from.
  • For ratings, use the dedicated NPS or CSAT field types so they power score-over-time views — don’t leave a score as a plain number field, or you won’t be able to filter on it cleanly.
  • Map only the fields that matter. Extra fields won’t hurt analysis, but they clutter your filters with options no one uses.
Field types can now be changed after import, and the fix aligns across all the data — but it’s still best to get content-versus-field and field types configured at setup.

​
Import best practices for CSV-based channels

  • Seed with a backlog, but not too far back (6 months or less).
  • Combine the backlog into one large first upload rather than month-by-month, so topics and themes form on broad patterns instead of a single month’s.
  • If using a CSV, cut uploads cleanly by period (full days or months) to avoid duplicates, which aren’t auto-removed.
  • Name each upload clearly (for example, “June 2026”) so it’s obvious what’s been ingested. To add data later, open the channel, go to its source, and add a new source — they stack on top of each other.

​
3. Context, topics, and themes

A channel organizes itself in three layers. Context is what you write when you set up the channel. It tells Dovetail who you are, what the data is, and what to pay attention to — i.e. the overall direction for synthesizing your data. Topics are the big umbrellas — broad groupings like “app performance” or “user interface.” Dovetail generates a starting set with descriptions; you can rename them, rewrite their descriptions, delete them, or add your own. Learn more about how to write Context and Topic Descriptions. Themes live inside topics and are narrower and more action-oriented — the specific things people actually raise, like “slow loading times” or “playback controls disappearing.” This is where data points collect. You can edit theme titles, create your own, or delete existing themes at any time. A few behaviors worth knowing:
  • One data point can belong to many themes. A single review or ticket that raises five different issues lands in five themes — Dovetail never forces it into just the best match.
  • Always give a new topic a description. That’s how Dovetail knows what belongs in it.
  • Editing applies going forward; creating new applies retroactively. If you edit a topic or theme’s title or description, the change affects future data only. If you create a brand-new one, Dovetail looks back across existing data as well as forward.
  • Deleting a theme doesn’t delete the data. The underlying data points simply become unclassified (and reappear if you recreate the theme). Dovetail can also create new topics and themes as new data is imported.

​
4. Reading and using a channel

The heart of a channel is trends over time. Open a theme and you get a chart of its mentions over time; switch between bar and line, and adjust the interval — a wider interval like by-month is usually easier to read than by-week. Gaps simply mean there are no matching data points in that window. Filter by your fields to scope the view to a specific segment — only promoters, only a region, only a customer type. As you filter, the theme summaries regenerate to describe just that subset, and the counts update so you can compare magnitudes (a heavily mentioned pain point versus a minor one). Click any individual data point to see its full record — all of its fields and every theme it belongs to. And from a theme or data point you can send to Linear or Jira to open a ticket, carrying the evidence with it: “this is now a product request because this many people are asking.”

​
5. Analyzing with chat and agents

Two layers of AI float with you across Dovetail, and both work on channels. Chat lets you ask questions of the data conversationally. The default context follows where you are — in a channel, you’re chatting with that channel. You can also remove the default and set your own context, mixing and matching multiple channels, projects, or folders. That’s how you compare across sources. Agents are chat with an action layer: anything chat can do, an agent can do on a schedule and then deliver. Every agent needs two ingredients (both shown in blue in the prompt): the context (the data to use) and the action (what to do and where to send it).

​
6. Dashboards

Dashboards turn a channel into a clean, high-level view stakeholders can actually read — a landing page instead of a wall of charts.
Dashboards are currently in beta, so if you don’t see the option, enable Dashboards under More → Settings → Beta.
When you create a new dashboard, Dovetail will create a starter dashboard for you so you’re not staring at a blank page. From there, you can edit the prepopulated widgets or add a new one. For each widget you choose the data source, set the time period and interval, and apply filters. A few things to know:
  • Each widget pulls from one channel — a widget never aggregates across channels, by design. You can, however, place widgets from several channels on one dashboard, and several widgets from the same channel (for example, overall NPS plus NPS by segment).
  • The keyword widget is the exception — it searches across the entire workspace, counting how often a word appears anywhere it’s said, whether or not anything’s been tagged. Useful for tracking competitor or topic mentions.
  • Theme-trend widgets default to the top four themes, which update dynamically as volume shifts. You can instead pick specific themes, but a manual selection stays static.
  • Rename a widget’s title whenever you filter it so it’s clear what someone’s looking at (for example, “Sentiment from NPS — promoters”).
  • Stakeholders can chat with a dashboard and click any widget to drill into the underlying channel.

​
Where to next

Analysis in projects

See how analysis works for deep, qualitative studies.

Channels

Learn how to create, configure, and manage channels.

Context and topic descriptions

Write context and topics that organize feedback the way you need.

Dashboards

Build a high-level view stakeholders can read at a glance.