You have a column of 300 customer comments and you need each one tagged happy, unhappy, or needs-follow-up. The old options were: do it by hand over two evenings, learn a formula you’ll forget, or install an add-on that wants $19 a month and access to your entire Google account.
There’s now a third option that’s been sitting inside Google Sheets, and most people haven’t noticed it. You drag the little square at the corner of a cell — the same handle you’ve used a thousand times to copy a formula down — and Sheets writes the whole column for you.
It’s called Fill with Gemini. It went out to more users on July 7, and if you’ve ever searched for “AI in Google Sheets,” here’s the thing worth knowing: almost every result on the first page is trying to sell you an add-on for a job your spreadsheet now does natively.
What it actually does
Fill with Gemini reads the context of your table and fills empty cells with written text. Four jobs, specifically: it writes (product blurbs, replies, descriptions), it summarises (a long paragraph down to a line), it categorises (300 expenses into clean buckets), and it tags sentiment (customer comments into happy / unhappy).
No formulas. No add-ons. It’s built into Sheets.
There are two ways to trigger it, and knowing both saves you a lot of confusion:
Drag-to-fill is the one that feels like magic. Fill in two or three rows the way you want them — “Refund request”, “Shipping question”, “Praise” — then grab the corner square and pull down. Sheets infers the pattern and does the other 297.
Prompt-to-fill is for when the column is completely empty. Select the blank cells, hit Fill, and write what you want in plain English: “Summarise the review in this row in under 10 words.”
Google says it’s 9x faster than typing it in yourself. That figure comes from Google’s own study of 95 people doing one 100-cell task — so treat it as a vendor number, not an independent finding. The direction is right. The precision isn’t the point.
The question nobody on page one answers: which plan do you need?
This is the single most common frustration with this feature, and the entire search results page dodges it. People open Sheets, look for the button, and it isn’t there.
Here’s the honest answer, straight from Google’s own rollout notes.
| If you have… | Do you get it? |
|---|---|
| Business Standard or Business Plus | Yes |
| Enterprise Standard or Enterprise Plus | Yes |
| Google AI Pro or Google AI Ultra (personal) | Yes |
| AI Expanded Access / AI Ultra Access add-on | Yes, with higher limits |
| Business Starter | No — not on the list |
| A free personal Gmail account | No — you need paid Google AI Pro or Ultra |
So the common shorthand — “it’s free inside your Google account” — is wrong, and it’s worth being straight about that. It’s included in some paid plans. A free Gmail account does not get it.
Two more things that quietly decide whether the feature shows up at all:
- Your admin can hide it. It rides on a setting called “smart features for Google Workspace.” If your IT admin has that switched off, the button doesn’t exist for you, no matter what you’re paying. It’s on by default for eligible accounts.
- It’s desktop only, English only, and text only. The enhanced Smart Fill behaviour needs at least three example rows to learn from, and it only triggers on text columns — not numbers, not dates. If you’re trying to get it to fill a column of prices, that’s why nothing is happening.
One dated detail worth knowing this week: the generous promotional usage limits run through July 15 — that’s tomorrow. After that, per-user limits apply, and the higher caps go to accounts with the AI add-on licences. Google hasn’t published the exact numbers. So if the feature works generously today and starts refusing large jobs later this week, that’s why — it isn’t broken.
Where it gets it wrong (and the 60-second check)
Here is the part that no page-one result will tell you, because most of them are advertisements.
This tool is confidently wrong sometimes. Not “occasionally glitchy” — confidently wrong, which is worse, because a wrong answer looks exactly like a right one when it’s sitting in a tidy spreadsheet cell.
The research is unusually clear about where it breaks. A 2026 benchmark ran the frontier models — including Gemini — against realistic spreadsheets. The best model scored 82.4% overall, which sounds fine until you translate it: roughly one wrong answer in every six. And accuracy is not evenly spread. It collapses as the job gets harder:
Two out of three complex aggregations come back wrong. On a large, messy sheet, average accuracy across models dropped to about 48.6%. A separate benchmark had GPT-4-class models at 28% on spreadsheet questions where humans scored 64%.
That is not an argument against using it. It’s an argument for knowing which jobs to hand it. Categorising, summarising and writing text — the things Fill with Gemini is actually built for — sit at the reliable end. Asking it to reason across several columns and do arithmetic is where it falls apart.
The specific failure modes to watch for, all documented:
- It reads the wrong column when your headers are similar. “Category” and “Sub-category” sitting next to each other is a classic trap.
- It invents categories. You asked for four buckets and now there’s an “Other” and a “General” and a “Misc” — and “Email” and “Email / newsletter” as two separate things.
- Merged cells and hidden headers break it. It reads your table as text, so the visual cues you rely on — bold headers, merged title rows — just confuse it.
The 60-second check. Don’t audit 300 rows. Do this instead:
- Filter the column you just filled and look at the list of unique values. If you asked for four categories and you’re looking at nine, you’ve found your problem in about eight seconds. Duplicates and invented buckets show up immediately.
- Random-sample ten rows. Add a helper column with
=RAND(), sort by it once, and read the top ten — the original text and the AI’s answer, side by side. Ten random rows will catch a systematic error every time. - Check the ugly rows on purpose. Sort by the length of the input text and read the longest few. Ambiguous, rambling, sarcastic entries are where it fails — a “thanks, this broke immediately” review gets tagged happy more often than you’d like.
Do those three things and you’ve spent a minute. It’s the difference between a tool and a liability.
What this means for you
If you run a small business: the highest-value first job is the one you’ve been avoiding. A messy customer list, a year of uncategorised expenses, 200 product descriptions you never wrote. Do one column, spot-check it, and you’ll know within ten minutes whether this earns a place in your week.
If you’re the office admin or the operations person: this quietly kills a category of task you were never thanked for. Cleaning an import, standardising a country column, tagging a feedback dump. Take the hour you get back and don’t tell anyone.
If you sell online: bulk product copy is the obvious win — export listings to a sheet, fill titles and descriptions in your own voice, re-upload. Just don’t let it write the facts. It’s good at tone and terrible at knowing your actual dimensions and materials.
If you do the books: the categorisation is genuinely useful and the arithmetic is genuinely not. Let it sort transactions into buckets; do not let it total them. That split follows the benchmark data exactly, and it’s the whole discipline in one sentence.
If you’re on a free Gmail account: you don’t have this, and no article can give it to you. Your realistic options are a paid Google AI plan, or pasting a column into a chatbot and pasting the result back — clumsier, but free.
What this can’t do
- It won’t tell you when it’s wrong. There’s no confidence score, no flag, no error. Wrong values arrive looking exactly like right ones. That’s the whole risk, and it’s why the spot-check isn’t optional.
- It won’t do your numbers. Text columns only — and given that complex aggregation runs around 33% accurate, that limit is a feature.
- It won’t fix a messy table first. Merged cells, blank rows and vague headers degrade it. Ten seconds of tidying beats any prompt.
- It won’t work on your phone, or in another language. Desktop and English, for now.
- It doesn’t make you an expert on your own data. It makes the typing faster. Deciding what the categories should be is still the part that requires you.
The bottom line
The reason to care isn’t that the AI is clever. It’s that a job which used to cost you an evening now costs you a drag and a one-minute check — and that the search results are still telling you to buy an add-on for it.
Learn the two triggers, know whether your plan actually includes it, hand it text and not arithmetic, and never skip the spot-check. That’s the whole skill.
If you want to go from “I opened a spreadsheet” to “I clean up any messy list without thinking about it,” our AI for Spreadsheets course starts free — the first two lessons are open, no signup.
Sources
- Effortlessly automate data entry in Google Sheets using Fill with Gemini — Google Workspace Updates
- Fill with Gemini in Sheets now available in 11 additional languages — Google Workspace Updates
- Use enhanced Smart Fill with Gemini in Google Sheets — Google Docs Editors Help
- From Simple Lookups to Complex Reasoning: Where LLMs Break on Spreadsheets — arXiv
- Large Language Models on Tabular Data — A Survey (arXiv)
