How to Use the =AI() Function in Google Sheets

Most Australian businesses have data they do not know what to do with. Customer survey responses sitting unread in a Google Form. Product return reasons logged in plain text. Columns of feedback that someone planned to categorise manually but never quite got to.

The =AI() function in Google Sheets changes that calculation. It brings Gemini's language capabilities directly into your spreadsheet as a native formula — the same way you would write =SUM() or =VLOOKUP(), but instead of adding numbers or looking up rows, you can classify text, extract key information, analyse sentiment, or generate structured summaries — one row at a time, automatically, across thousands of cells.

This tutorial covers exactly how to set it up, how to write prompts that produce consistent results, and four practical business use cases that deliver real time savings for Australian SMBs.

What is the =AI() Function in Google Sheets?

The =AI() function is a native Google Sheets formula powered by Gemini. It does not require installing an add-on from the Workspace Marketplace or connecting an external API. If your Workspace plan includes Gemini, the function is already available in your spreadsheets.

The basic syntax is straightforward:

=AI(prompt, [reference_cell], [temperature])

In practice, you almost always combine a text prompt with a cell reference:

=AI("Classify this customer feedback as Positive, Neutral, or Negative: " & A2)

The formula sends the contents of A2 to Gemini along with your instruction and returns the model's response into the cell where the formula lives. You can drag the formula down a column and it processes every row automatically — exactly as you would expect from any other Sheets formula.

This is fundamentally different from the Gemini side panel or "Help me write" features. Those tools are conversational and generate one-off outputs. The =AI() function is batch-capable, repeatable, and integrated into your sheet's data structure. It is the difference between asking an assistant a question and building a process that runs itself.

Prerequisites: Which Google Workspace Plans Support =AI()

The =AI() function is not available on every Workspace plan. As of early 2026, it is included in plans that bundle Gemini capabilities:

If your team is on Business Starter, you can either upgrade to Business Standard or add the Gemini Business add-on at approximately AUD $30 per user per month to unlock the function.

If your team is already on Business Standard or higher, =AI() is included at no additional cost. For a team of five already on Business Standard, that means AI-powered data processing in Sheets costs exactly nothing extra.

To confirm availability in your account, open a Google Sheet, click on any cell, and type =AI(. If autocomplete suggests the function with its syntax hint, you are ready to go. If nothing appears, check your Workspace plan in the Admin console under Billing > Subscriptions.

Basic Usage: Your First =AI() Formula

Step 1: Open a Google Sheet with data

For this tutorial, assume column A contains customer feedback responses — raw text exactly as customers typed them into a form. Column B is where you will build the AI formula.

Step 2: Write a prompt that defines the task clearly

Click on cell B2 and type:

=AI("Classify this customer feedback as Positive, Neutral, or Negative. Return only the classification word, nothing else: " & A2)

Press Enter. The function sends A2's content to Gemini and returns a single word: Positive, Neutral, or Negative.

Step 3: Drag the formula down

Click B2, grab the fill handle at the bottom-right corner of the cell, and drag it down to B100 (or however many rows you have). Each row processes independently. Google Sheets queues the API calls and fills results progressively — you will see cells populating over a few seconds.

Step 4: Freeze the results

Once you are satisfied with the output, select the formula column (B), copy it (Ctrl+C or Cmd+C), then paste as values only using Edit > Paste special > Values only. This converts the live formula results into static text, so they do not recalculate on every sheet refresh and are not affected by API availability.

Prompt Engineering for Consistent Results

The quality of =AI() output depends almost entirely on prompt quality. These four principles produce reliable, processable results:

1. Constrain the output format explicitly

Bad: =AI("What do you think about this review? " & A2)

Good: =AI("Classify this review. Respond with exactly one of these words: Positive, Neutral, Negative. No other text. Review: " & A2)

When you constrain the output, downstream formulas like =COUNTIF() and =FILTER() work reliably against the AI column.

2. Provide category definitions for ambiguous classifications

=AI("Classify this support ticket into one category. Categories: Billing (payment, invoice, refund), Technical (bug, error, not working), Account (login, password, access), General (everything else). Return only the category name. Ticket: " & A2)

Giving Gemini your category definitions eliminates the guesswork and dramatically reduces edge-case errors.

3. Use examples for nuanced tasks

=AI("Extract the product name from this customer message. Examples: 'My blue XL hoodie is faded' → 'hoodie'. 'The coffee machine arrived broken' → 'coffee machine'. Now extract from: " & A2)

Few-shot examples (showing the model what correct output looks like) significantly improve extraction accuracy when the task is not straightforward.

4. Keep prompts consistent across a column

If you are classifying 500 rows, every cell in column B should use the exact same prompt logic. Do not customise prompts per row manually. This ensures results are comparable across the dataset.

Practical Example 1: Categorising Customer Support Tickets

The scenario: An Australian e-commerce business receives 200+ support tickets per week via a Google Form. The team wants to categorise each ticket (Billing, Technical, Account, General) and then route them using a FILTER formula.

Column A: Raw ticket text (e.g., "I was charged twice for my order")
Formula in column B:

=AI("Categorise this support ticket. Choose one: Billing, Technical, Account, General. Billing = payments, invoices, charges, refunds. Technical = errors, bugs, app issues. Account = login, password, profile. General = everything else. Return only the category name. Ticket: " & A2)

Column C — Priority flag:

=IF(B2="Billing", "High", IF(B2="Technical", "Medium", "Standard"))

Results: The team went from spending 45 minutes manually sorting tickets each morning to running the sheet once a week. The AI classification takes about 90 seconds for 200 rows. Staff now spend that time resolving tickets rather than routing them.

Practical Example 2: Sentiment Analysis on Customer Reviews

The scenario: A Melbourne hospitality business collects post-visit reviews through a Google Form. Management wants a monthly sentiment breakdown across different service dimensions — food, service, atmosphere, value.

Column A: Full review text
Columns B through E: Sentiment per dimension

Formula in column B (Food sentiment):

=AI("From this restaurant review, assess the sentiment toward the FOOD specifically. If the review mentions food, return Positive, Neutral, or Negative based on what is said about it. If food is not mentioned, return N/A. Return only the rating word. Review: " & A2)

Repeat the pattern for columns C (Service), D (Atmosphere), E (Value), adjusting the dimension name in each formula.

Monthly summary using COUNTIFS:

=COUNTIF(B2:B500, "Positive")

Management now has a live dashboard that tracks sentiment trends across dimensions by month — without anyone reading 500 reviews manually.

Practical Example 3: Extracting Structured Data from Unstructured Text

The scenario: A Brisbane trade services company receives job requests via email, which staff paste into a Google Sheet. The requests are free-text and inconsistently formatted. The team needs to extract: suburb, job type, and urgency level.

Column A: Raw job request text
Columns B, C, D: Extracted fields

Formula in column B (Suburb):

=AI("Extract only the suburb name from this job request. If no suburb is mentioned, return Unknown. Return only the suburb name, no other words. Request: " & A2)

Formula in column C (Job type):

=AI("Identify the type of trade work described. Choose one: Plumbing, Electrical, Carpentry, HVAC, Painting, General. Return only the job type. Request: " & A2)

Formula in column D (Urgency):

=AI("Based on the language used in this job request, classify the urgency. Urgent = mentions emergency, flooding, no power, today, ASAP. Standard = everything else. Return only Urgent or Standard. Request: " & A2)

This setup lets the coordinator sort by suburb and urgency in seconds, dramatically reducing the time from job request to dispatch.

Practical Example 4: Bulk Text Summarisation

The scenario: A Sydney consulting firm maintains a Google Sheet of client meeting notes pasted from documents. Principals want a one-sentence summary of each meeting's key outcome for quick review.

Column A: Full meeting notes (multi-sentence, sometimes several paragraphs)
Formula in column B:

=AI("Summarise the key outcome of this meeting in one concise sentence. Focus on decisions made and next steps. Notes: " & A2)

The result is a scannable column of one-liners that let principals track engagement history at a glance without opening individual documents.

Data Classification at Scale: Tips for Large Datasets

When running =AI() across hundreds or thousands of rows, keep these practical points in mind:

Batch in chunks: Sheets processes =AI() calls sequentially. For large datasets (500+ rows), consider breaking your data into groups of 200-300 rows per sheet and processing separately. This reduces the risk of timeouts and makes it easier to spot and re-run failed rows.

Watch for blank outputs: Occasionally a row will return an empty cell rather than an error. Build a check column:

=IF(B2="", "REPROCESS", "OK")

Filter on "REPROCESS" and re-run those rows after the initial batch completes.

Use validation to confirm output quality: After processing, use COUNTIF to tally how many results match your expected categories. If 15% of results are not matching any expected value, your prompt needs tightening.

=COUNTIF(B2:B500, "Positive") + COUNTIF(B2:B500, "Neutral") + COUNTIF(B2:B500, "Negative")

If this number is significantly less than your total row count, review the outliers and refine the prompt.

Convert to values before sharing: Always paste-as-values before sharing the sheet with stakeholders or importing results into another system. Live =AI() formulas recalculate on open, which creates inconsistency in results and can hit usage limits unexpectedly.

Limitations You Need to Know

The =AI() function is genuinely useful, but it has real constraints that affect how you should plan your workflows.

No guaranteed determinism: Gemini can return slightly different outputs for the same input on different runs. For classification tasks, this is usually minor, but it means =AI() is not suitable for tasks where exact, auditable repeatability is required (such as financial calculations or legal categorisation).

No model selection: Unlike third-party options (GPT for Sheets, Claude for Sheets), you cannot choose which Gemini model runs your formula. Google manages this. Output quality may vary compared to leading third-party models for complex nuanced tasks.

Usage limits apply: Google does not publish granular rate limits for =AI() calls on standard Workspace plans. Very high-volume processing (tens of thousands of rows) may hit daily limits. For sustained high-volume use, a third-party solution with pay-per-use API billing may be more cost-effective.

Context window constraints: The =AI() function handles typical row-level data well (a few hundred words per cell). If you are feeding in very long documents — several pages of text — the function may truncate content or produce degraded results. Keep individual cell content under approximately 2,000 words for reliable output.

Not suitable for PII-heavy data: Review your organisation's data handling obligations before passing personally identifiable information through =AI(). Data is processed by Google's Gemini infrastructure. For highly sensitive datasets, assess whether this aligns with your privacy policy and applicable Australian Privacy Act obligations.

Resources and Getting Started with Google Workspace

If your team is not yet on a Workspace plan that includes the =AI() function, or if you are evaluating whether to upgrade from Business Starter to Business Standard, the following resource covers plan options and pricing:

Google Workspace plans starting from approximately AUD $10.80/user/month (Business Starter) through to Business Standard at ~AUD $16.80/user/month where =AI() becomes available. For teams of five or more, the productivity gains from AI-assisted data processing in Sheets typically offset the cost difference within the first month of active use.

You can explore current Australian plan pricing and sign up through the official Google Workspace referral link below. Using this link supports the production of independent, practical content like this guide at no additional cost to you:

Get started with Google Workspace

For teams already on Business Standard or higher, no action is required — open any Google Sheet, type =AI(, and start building.

Conclusion: Turn Your Backlog of Unprocessed Data into Actionable Insight

The =AI() function in Google Sheets is not a gimmick. It is a practical tool for the specific problem that most Australian SMBs face: data that exists but is not usable because processing it manually is too slow and too expensive.

Customer feedback, support tickets, job requests, survey responses, product reviews, meeting notes — all of these are sitting in Google Sheets right now, unclassified and unsummarised, because the hours required to process them at scale do not exist in most business budgets.

The =AI() function changes that. Once you understand prompt engineering basics and the formula syntax, you can build a classification system in 30 minutes that processes hundreds of rows in 90 seconds.

Start with one use case. Pick the data backlog your team most wants to act on. Build a three-column test: raw data, AI output, and a manual spot-check column. Run it on 20 rows, review the results, and refine the prompt until accuracy is where you need it. Then scale.

The function works best when prompts are specific, outputs are constrained, and results are converted to static values before downstream use. Those three habits, applied consistently, make =AI() a reliable part of your team's data workflow rather than a novelty you experiment with once and abandon.

If your Workspace plan already includes it, there is no reason not to use it today.

Need help setting up AI-powered data workflows in Google Sheets for your business? Contact our team for a free consultation.