Files for your first read

First Read builds a report, dashboard or deck from a file you upload while you sign up. It is looser than a Data Sources upload: there is no column mapping and your columns can have any names. You only pick which column holds the text people wrote. No file of your own yet? Download a sample dataset and use that.

The data step, in five moves

1. Add your file. Drop it on the box or select Browse. CSV, JSON, JSONL or Parquet; for an Excel file, save it as CSV first. You can add several files, each with a different name.
2. Check what was read. Each file opens as a card with its size, rows and columns. The file is read on your computer: nothing is uploaded until you select Build my First Read on the last step.
3. Pick the text column. Under Which column holds what customers wrote?, choose the column of free text. Use the table below if you’re not sure which one. That file is marked Main.
4. Read the signal score. The Signal assessment on the right scores the column out of 100, from No signal yet to Strong signal, and shows whether it found text, timestamps and a customer segment. If the score is low, try a longer text column or a file with fuller answers.
5. Select Next. Any other files you added travel with the main one as context.

Which column to pick

Your data	One row is	Pick as the text column	Columns worth keeping
Reviews	One review	The review body, not the title	Rating, date, store or location, source
Survey answers	One response	The open-ended answer, such as “Why did you give that score?”. If there are several, pick the one with the longest answers	Score (NPS, CSAT), date, plan or segment
Support tickets and email	One ticket, or one message	The customer’s message, or the whole thread	Channel, date, plan, status, outcome
Chat and chatbot conversations	One conversation, or one message	The transcript or message text	Channel, date, whether it was resolved
Call transcripts	One call, or one speaker turn	The transcript text	Agent, date, call reason
Social posts and community threads	One post	The post text	Platform, date, topic

When a conversation takes several rows (one per message), keep a column that groups them, such as a conversation or ticket id, and one that says who wrote each message.

Lay it out like this

• One row per thing someone wrote: one review, one survey response, one ticket or chat. A whole thread can sit in one cell, or take one row per message as above.
• Longer text gives a better read. Setup scores the text column on how much was written and how many rows have any. Full sentences score well; a column that averages a couple of sentences or more scores strongest.
• A date column, if you have one, so the first read can show what changed over time.
• Columns that say who or where: a rating or score, a plan, a location, a channel. They let the first read split and compare what people said.
• Leave out rows with no text rather than filling them with placeholders such as N/A.

Sample datasets to download

Pick one, download it and upload it as it is. Each dataset has Dimensions already filled in as extra columns (sentiment, topics, signals such as churn), so you can also open it to see what enrichment adds to text.

• 01 · Retail locations Public reviews Home Depot Google reviews of 24 Atlanta-area stores, Jan 2025 to Jun 2026 Use caseCompare locations: find what separates the best-reviewed stores from the worst under the same brand. 2,504 reviews1.8 MBText column: Message ↓ Download CSV
• 02 · Consumer app Public reviews Duolingo App Store reviews from English-speaking markets, Jan 2025 to May 2026 Use caseTrace a product decision: see how one change to the app shows up in ratings, by app version and country. 5,672 reviews4.2 MBText column: Message ↓ Download CSV
• 03 · Telecom support Synthetic sample T-Mobile 520 support calls written as if to T-Mobile, one row per turn Use caseRank support issues by what the customers are worth: every call carries the customer’s monthly revenue. 8,120 turns3.5 MBText column: text ↓ Download CSV
• 04 · Enterprise SaaS Synthetic sample Zendesk 782 tickets, survey answers and account notes written as if for Zendesk Use caseFind churn early: what customers wrote, joined to whether each account renewed or churned. 1,830 messages1.4 MBText column: text ↓ Download CSV

In the two support datasets a conversation spans several rows: sessionId groups them and incoming is true for what the customer wrote.

After your first read

The first read works from your file as it is. To get Dimensions on the same data, add it under Data Sources: see Upload a file. Your project’s Overview tab walks through the steps.