Skip to main content
There are three ways to change the questions your brand is measured on. For most jobs the AI chat is the simplest and safest. Use CSV only when you’re adding dozens of prompts at once or rebuilding the panel.

​
Before you start

  • You need access to the Prompts screen. Any workspace member has it.
  • For CSV you need a spreadsheet program.
  • You can’t go over your plan cap. Check the Prompts (12/500) count next to the title first.
Wording edits are staged whichever route you use, and apply when you save a version. Adding and deleting prompts takes effect straight away.

​
1. Choosing the route

The reason not to reach for CSV over a single wording change is simple: the moment you open the file you risk blanking a cell or dropping a column, and those mistakes go through silently.

​
2. Editing with the AI chat

AI Agent at the top right of the Prompts screen opens the chat. Just say what you want in plain language. Anything that changes data appears first as a Needs approval card showing the before and after, and nothing happens until you press Approve.
If you don’t like the wording it proposes, press Edit on the approval card, rewrite it, then approve.
Some things the chat can’t do: save a version, run a CSV import, change your plan limit, or change classification values like scope (prompt_tier) and question shape (structure). Ask us if you need those changed. You can also edit directly on the screen: expand a topic, hover a prompt, click the pencil, and change the wording in place.

​
3. The CSV procedure

Follow the order. If you skip step 1 and upload an old or hand-built file, existing prompts get recreated as new ones or deleted.
1

Download the current panel

Press Download CSV at the top right of the Prompts screen. Edit that file. The value that identifies your existing prompts (prompt_id) only exists in the downloaded file.
2

Edit in a spreadsheet

Add or remove rows and change only the cells you need. Don’t delete or rename any columns.
3

Import, then pick Add or Replace

Press Import and choose your file. It’s checked first, then the Import topics and prompts dialog opens in Add mode. Switch to Replace if that’s what you want; switching re-checks the file.
4

Check the preview

You’ll see the topic and prompt counts and what will change (50 new · 12 updated · 3 removed · 0 skipped). Make sure the numbers match what you expect. Problems show up here, and nothing is saved yet.
5

Save a version

An upload does not create a version by itself. Click the Version N chip in the header and press Save version so your wording changes take effect.

​
4. Add vs Replace

If you pick Replace, the file must contain every prompt you want to keep. Any row you leave out is silently deleted and stops being measured. If nothing needs deleting, there’s no reason to use Replace. In both modes, wording changes to rows that already exist are staged. Only new rows and removals are live immediately. Deleted prompts keep their past measurements, and saved versions stay intact.

​
5. prompt_id rules

prompt_id is the prompt’s identity. Keeping it and changing the wording rewrites that prompt; it does not create a new one.

​
6. The 16 columns

You can reorder them, but don’t rename or delete any. Required Fixed values only Anything outside those values is saved as an empty cell without an error, and only shows up in the warnings list. Free text language, target_country and product_name cannot be changed on a row that already has a prompt_id.

​
Troubleshooting

When it’s rejected, nothing is saved. Each reason is listed with the row number.
  • prompt_id values belong to another workspace → Clear prompt_id on the rows you copied in and upload again.
  • Duplicate prompt_id values are not allowed → Clear prompt_id on the duplicated rows.
  • Duplicate prompt templates are not allowed → Delete one of the identical rows.
  • prompt_id values use a template belonging to another prompt → Download a fresh file and start over.
  • Existing prompt scope cannot be changed by CSV import → Put language, target_country and product_name back as they were on existing rows.
  • Prompt classification conflicts must be fixed before import → For an explicit brand-versus-competitor question use torso or tail in prompt_tier, or remove the comparison wording.
Some cells could not be read doesn’t block the import. Only the affected cells come through empty.
  • prompt_tier “Broad” is not a value this column accepts → The screen shows Broad / Specific / Niche, but the file needs head / torso / tail.
  • estimated_volume “5,640” is not a whole number → Write 5640, with no separators.
  • cluster_name is empty, so this prompt was not imported → Fill in the topic name and upload again.
These go through with no warning at all.
  • Blanking a cell overwrites the whole row, wiping the old value. Only touch the cells you’re changing.
  • Deleting a column wipes that value on every row.
  • A row missing in Replace mode deletes that prompt and stops its measurement.
  • A row with an empty cluster_name is ignored.
  • A value outside the accepted list is saved as empty.
  • A misspelt product_name doesn’t link.
  • Changing wording but leaving cluster_name keeps the old group name.
If the preview counts don’t match what you expect, it’s one of these. Check the file before confirming.
Save as comma-separated CSV in UTF-8. You can reorder columns and case doesn’t matter, but don’t rename or delete one. Don’t put line breaks inside a question. Commas and quotation marks are fine. The blank lines between topics are normal — leave them.

​
Finish by saving a version

When you’re done, click the Version N chip in the header and press Save version. That records the wording and panel membership at this moment as one version, and applies the wording edits you had staged. The Pending changes card shows what’s outstanding: adds and deletes as Already applied to the live panel, wording edits as N prompt edits. To throw the staged edits away instead, press Discard. Saving takes a moment — large panels can take up to a minute, so keep the window open. Earlier versions stay as they were, and the Version picker on the analytics screens reads your data against any of them.

Prompts

What’s on the screen and how to read the numbers

Visibility

What your edited panel measures