Build apps · Your App Cloud
Build an app that makes sense of feedback
Bring survey answers, reviews and support comments into one place. Have a language model suggest which of your categories each comment fits, then confirm the tags and read the quotes behind a theme before you report it.
Built around a purpose
From a pile of comments to themes you can trace.
For people who must say what customers are asking for, from more comments than they can read twice. That might be a product manager with a quarter of survey answers, a support lead with ticket comments, or an owner with a season of reviews.
For illustration: 1,200 answers to the survey question "What is one thing we should change?" from a café group with three California locations, each with a date and a location, plus 300 online reviews pasted into a second sheet. Add the decision the work feeds, such as which three changes to make before the spring menu.
A comment library where every theme shows how many confirmed comments it holds, how that number moves by month and location, and the customers' own words behind it.
Plan this on Pro. Suggesting a category for each comment runs a language model inside the app, and saved tags and reviewer sign-in need a database and authentication. Free publishes a static frontend only, with no database to keep a confirmed tag. Suggestions for all 1,500 comments are model usage, metered separately, so try your categories on a sample first.
Name the decision first
Write down what the analysis is for and what finding would change your plan. Keep only the fields that serve it: the comment, its date, its location, and whether it came from the survey or a review.
Clear out noise before tagging
Remove exact duplicates, blank answers and replies such as "n/a" or "nothing". Mask email addresses, phone numbers and staff names. Apart from that masking, keep each customer's wording as written, in a field of its own.
Draft categories from a first read
Read 100 comments picked at random and group them by hand. Turn the groups into eight to twelve categories, each with a one-sentence definition and an example comment. Add Other and Unclear at the end.
Ask for a suggestion and a reason
Describe in Studio what a suggestion should contain: one category from your list and a one-line reason. A person confirms or changes each one and moves anything that fits nowhere to Unclear for a second look.
Count, quote, then report
Show confirmed counts by month and by location, and let each count open the comments behind it. Choose report quotes from inside the theme, so each one is typical of it and not just the most vivid.
Scope the first version
What a first version includes.
Comment records
One record per comment with the original text, the source (survey or review), date, location and, where the export has one, a respondent ID so repeat writers stand out.
Your category list
Each category with a one-sentence definition, an example comment and a note on what it leaves out. Other and Unclear sit at the end of the list.
Suggested and confirmed tags
Each comment shows the suggested category, the reason given, and the tag a person confirmed. Only confirmed tags reach the counts.
The Unclear queue
Comments that fit none of your categories, or two of them, waiting for a person. A queue that keeps growing is a sign a category is missing or two definitions overlap.
Theme counts
Confirmed comments per category by month and by location, where each number opens the list of comments behind it.
Theme notes
A short description of each theme, drafted by the model and edited by you, with the quotes you chose linked to their full comments.
Tagged comments to take elsewhere
An export of the app's data to send with the report, so the confirmed tags and their comments travel with it. Compare the exported counts with the ones on screen before you send it.
Test it in the preview
Before you share it.
- Before you look at any suggestions, tag 50 comments yourself, then compare. Where you and the app disagree, reread the comment. If the definition fits both tags, rewrite it. If it is clear, keep your tag and note the miss.
- Open the largest theme and read every comment in it against the definition. If several don't belong, the definition is too broad. Tighten it and ask for suggestions again.
- Find a review that was posted on two sites and search for it in the app. If both copies are there, delete one and confirm the theme count drops by one.
- Search the comment text for @ signs, phone numbers and staff first names. Anything you leave can be read by everyone with access to the comments, so mask it before reviewers sign in.
- Signed out, open the link to a theme. Customer comments should not load until someone signs in.
- Change one category's definition and ask for suggestions again on a sample. Tags a person already confirmed should stay as they were.
Together in your app cloud
The pieces behind the possibility.
Choose the models and services that support the experience you want to create.
Language models
Reason, write, and understand
Data models
Turn information into useful structure
Database
Keep data and context for apps and agents
Authentication
Give the right people access
Shape the details
Make it work for your world.
- Count people as well as comments where you can. One regular who writes in every week can look like a trend, so keep the respondent ID if your export has one and check how many people stand behind a theme.
- Size isn't importance. Four comments about a billing error or a safety problem can matter more than 200 about the music, so read the small categories before you rank anything.
- Say where the comments came from. A survey hears only from the people who answered and reviews lean toward strong opinions, so put the sources and dates beside every count you share.
- Decide who needs the raw comments. Taggers need the full text, but people who only read the findings may need just the counts and chosen quotes. Write that down as the access you describe for each collection (Pro), then sign in as a tagger and as a reader and compare what each can open.
A little more context
Good questions.
Can I choose my own categories?
Yes, and you should. Write them from a first read of your own comments rather than a generic list, give each a one-sentence definition, and keep Other and Unclear. When Other keeps growing, read it: those comments often point to a category you haven't named yet. When the same mistake repeats, change the definition, not just the tag.
Can I analyze survey exports?
Yes, once each answer sits on its own row with its date and the question it answers. The CSV and Excel guides below cover preparing the file. After import, check that the comment count matches the export. An upload is a snapshot, so later answers need a deliberate re-import or entry in the app. Keep the question text, because "What should we change?" and "What did you like?" need different categories.
How do I know the themes are right?
Treat them as a draft until two people agree on them. Ask a colleague to tag 30 comments using only your written definitions, without seeing your tags. Where you disagree, rewrite the definition until you would both choose the same category. Only then is it fair to judge the model's suggestions against it.
What happens to comments that arrive later?
Ask for suggestions on them when you're ready, or on a schedule you set (Pro). If you schedule it, have the task cover only comments added since its last run and list them for review. Keep the counts to confirmed tags, so a number changes only after a person has looked. If comments reach the app by upload, plan the re-import as well.
Explore Pricing for plans, included capabilities, and usage details.
Start from what you have
Bring the file you already use.
Keep exploring.
Your ideas, up and running
Your next idea has a home.
Build an app. Run an agent. Make it yours.
Start building