How to Categorize Customer Feedback in Google Sheets or Excel With AI
How to sort customer comments into categories with the =AI() function in Google Sheets, check every answer, and what to do in Excel now that its COPILOT function is gone.
Google Sheets can categorize customer feedback with its =AI() function: put the comments in a table, write the allowed categories with a one-line definition each into an =AI() prompt, fill the column, check every answer against the list with a COUNTIF, and count the results with a pivot table or QUERY. Excel no longer has a formula for this: Microsoft retired the COPILOT function on September 14, 2026, so in Excel you ask Copilot in the side pane or fall back to keyword formulas. This post is about doing the work inside the spreadsheet; sending rows from a sheet into an analytics tool is a different job, covered on our Google Sheets integration page. When the feedback outgrows a sheet (more rows than the quota, several channels, a report every month), Dimension Labs is the strongest next step: it reads every channel, enriches every comment with Dimensions built for your business, and builds the reports for you.
How do you categorize comments with the AI function in Google Sheets?
The AI function takes a prompt and, optionally, a range: =AI("prompt", B2). Google's own examples include categorizing customer inquiries and classifying sentiment. You can add it as an AI column in a table and fill it down, then select the cells and click Generate. The prompt is where the consistency comes from, so give the model a closed list, a definition per category and an instruction to reply with the category only:
=AI("Categorize this customer comment as exactly one of: Delivery, Billing,
Food quality, App, Other. Delivery = late, missing or wrong-address boxes.
Billing = charges, refunds, price. Food quality = freshness, taste, portions.
App = the app or website. Other = anything else. Reply with the category only.", B2)
Three habits keep the column usable:
Pass only the comment cell. The function reads the range you give it and nothing else in the file, so include any context (product, plan) in that range or in the prompt.
Check every answer against the list. A model can return "Delivery issue" instead of "Delivery", and a pivot table counts those as two categories. Put the allowed values in a small range and flag anything else:
D2: =IF(COUNTIF($H$2:$H$6, C2)=0, "NOT ON LIST", "") H2:H6 holds the five allowed categories: Delivery, Billing, Food quality, App, OtherDo not regenerate casually. There is no undo for an AI cell, only regenerate, and a regenerated cell may come back with a different answer. Once a column is checked, copy it and paste values only.
Need sentiment too? Add a second AI column with its own closed list (positive, negative, mixed). Keep one question per column; a prompt that asks for category and sentiment in one cell is harder to check and count.
What does a worked sheet look like?
Here is a small sheet laid out that way, with the category each comment should get and the result of a keyword formula next to it:
Sample data, written for this post
A feedback sheet with a category list, an AI column and a keyword column
Twelve comments about a fictional meal-kit service. Column C is the category you would give each one; it is what the =AI column should return and what you check it against. Column E is what the keyword formula returns. Bold: where the two differ.
| Row | B: Comment | C: Category (expected) | E: Keyword formula |
|---|---|---|---|
| 2 | Box arrived two days late and the ice packs had melted | Delivery | Delivery |
| 3 | Charged twice this week, still waiting on the refund | Billing | Billing |
| 4 | The salmon smelled off, threw it out | Food quality | Food quality |
| 5 | App logs me out every time I try to skip a week | App | App |
| 6 | Driver left the box at the wrong house | Delivery | Delivery |
| 7 | Portions got smaller but the price went up | Billing | Billing |
| 8 | Love the recipes, the chicken tacos were great | Food quality | Food quality |
| 9 | Couldn't skip a week because the button did nothing | App | App |
| 10 | Missing the sauce packet again | Delivery | Delivery |
| 11 | Cancelled, too expensive for what you get | Billing | Billing |
| 12 | Veggies were wilted on arrival | Food quality | Delivery |
| 13 | Customer service never answered my email | Other | Other |
The keyword column is the fallback for anyone without the AI function, and it shows the usual problem with keywords: "Veggies were wilted on arrival" is about food quality, but "arriv" is a delivery keyword and the delivery rule runs first. A model reads that sentence correctly; a formula needs another exception for every phrase like it. The formula itself:
=IFS(REGEXMATCH(LOWER(B2),"late|driver|arriv|missing|deliver"),"Delivery",
REGEXMATCH(LOWER(B2),"charge|refund|price|expensive|bill"),"Billing",
REGEXMATCH(LOWER(B2),"app|log|button|website"),"App",
REGEXMATCH(LOWER(B2),"salmon|chicken|veg|taste|fresh|recipe"),"Food quality",
TRUE,"Other")
REGEXMATCH is an ordinary Sheets function with no AI quota, and Excel for Microsoft 365 has REGEXTEST for the same job. To count the categories, use a pivot table, or one formula:
=QUERY(B1:C, "select C, count(B) where B is not null group by C order by count(B) desc", 1)
How many comments can a spreadsheet handle?
The AI function is capped per person. On Business Starter it is not available; on Business Standard, Business Plus, Enterprise Standard and Enterprise Plus each user gets 5,000 AI cells a month, and the AI Expanded Access add-on raises that to 25,000. Limits are per cell, per user, are not shared across a team, and do not roll over. Each Generate fills at most 350 selected cells.
From Google's and Microsoft's help pages
Sheets has a working AI function with a monthly cap; Excel's COPILOT function is gone
What each option allows, as each vendor's own pages describe it.
| Option | Who has it | Limits that matter for feedback |
|---|---|---|
| Google Sheets =AI() or =Gemini() | Business Starter: no access. Business Standard/Plus, Enterprise Standard/Plus: 5,000 cells a month per user. AI Expanded Access: 25,000 | 350 selected cells per Generate; cannot sit inside IF or another formula; no undo, only regenerate; reads only the range you pass it |
| Excel COPILOT() function | Retired September 14, 2026 | Cells that recalculate return #NAME?; values already calculated stay as cached values |
| Copilot in Excel (the pane) | Depends on your Microsoft 365 subscription and admin settings | You ask in words, for example to classify a range into a new column; no formula left in the cell |
| REGEXMATCH (Sheets) or REGEXTEST (Excel 365) | Ordinary formulas, no AI quota | Matches words, not meaning: fast and free, but every phrase needs a rule |
A category column and a sentiment column use two cells per comment, so the cap arrives quickly:
Arithmetic on Google's published limits
At 5,000 AI cells a month, 20,000 comments with two AI columns take eight months of one person's quota
Months of one user's AI function allowance (Business Standard/Plus or Enterprise Standard/Plus) needed to fill a category column and a sentiment column once. Each bar also shows how many times you press Generate at 350 cells a time.
A thousand comments a quarter fit comfortably. Twenty thousand, which a busy support inbox or a multi-location review program can produce, would take eight months of one person's allowance and 115 rounds of selecting and generating, before any checking.
How do you categorize feedback in Excel now?
Since September 14, 2026 the COPILOT function returns #NAME? when a cell recalculates; answers it had already produced stay in the workbook as values, so copy them to values before anything recalculates. Microsoft's suggested replacement is to ask Copilot in Excel, in words, to classify a range and add the result in a new column. Whether you see Copilot depends on your Microsoft 365 subscription and your organization's settings. Without it, the keyword approach above works with REGEXTEST in place of REGEXMATCH.
Whichever route you use, export only the columns the analysis needs before the text goes near a model; what IT teams check before customer feedback goes into an AI tool lists the questions to answer first.
When does customer feedback outgrow a spreadsheet?
Usually at one of four points: the comments exceed the quota, the same categories have to run every month, the feedback comes from more than one place (reviews, tickets, surveys, calls), or someone needs to know which category actually moves ratings or revenue rather than which is mentioned most. A file of feedback can also go straight to an analytics tool instead; how to analyze a CSV of customer feedback covers the layout. If you write code, how to classify customer reviews with an LLM in Python does the same categorizing in a script, with checks and a graded sample, and how to tell if your AI tagging is accurate covers checking labels from any tool.
Dimension Labs is an omni-channel analytics agent for customer data. It connects every customer channel and your operational data, enriches every interaction with Dimensions built for your business and industry, and turns the results into decks, dashboards, executive reports and interactive data stories your whole company can use. Plans start at $299 a month with no demo. For a team that started in a spreadsheet, that means:
Every channel in one place: the survey answers in your sheet next to reviews, tickets, chats and calls, all with the same categories.
Easy integration: connect the sheet itself with the Google Sheets integration, or upload the file, and join the comments to orders, accounts or locations.
Enrichment without the prompt-per-cell work: your category list becomes a Dimension, a fixed list with a quote beside each label, tested on a sample before it runs on every comment. The pre-built library comes with Explore at $299 a month; your own categories start on Discover at $999, and joining them to revenue data is on Prove.
A visual analytics layer: the agent builds the monthly report, deck or dashboard, saves it to the Final Reports folder, exports PDF or PPTX, hands back the underlying rows as a CSV if you still want them in a sheet, and reruns the plan each month.
The NBA made the same move from static reporting: it now analyzes 100,000+ open-ended fan survey responses within hours instead of months, and its insight review cycle went from about 60 days to under 6 hours.
FAQ
What is the best way to categorize customer feedback in Google Sheets?
For a few hundred comments, the =AI() function with a closed category list in the prompt and a COUNTIF check on the answers. For feedback that arrives every month from several channels and needs reports other teams read, Dimension Labs, where the same categories run on every source and new rows from the sheet arrive through the Google Sheets integration, sent by a scheduled job on your side.
Does Excel have an AI function like Google Sheets?
Not any more. The COPILOT function was retired on September 14, 2026; use Copilot in Excel from the pane, or keyword formulas with REGEXTEST.
How many rows can the Google Sheets AI function process?
350 selected cells per Generate, and 5,000 cells a month per user on Business Standard/Plus and Enterprise Standard/Plus, or 25,000 with AI Expanded Access. Business Starter has no access.
Can I use the AI function inside IF or another formula?
No. Google's help page says embedded AI functions are not supported, so keep the AI column separate and build checks and counts in other columns.
If your feedback has outgrown the sheet, start with one file: upload the export and get the categories, the counts and the first report back.