Airtable vs Google Sheets for Business Data AI Can Use

Keep business data in Google Sheets when it's a flat list, one or two people edit it, and your Workspace plan includes Gemini. Move it to Airtable when records link to each other, several people edit daily, or AI should fill a field on every new record. Airtable's Team plan starts at $20 a user a month.

The tool matters less than the shape of the data. AI reads one row per thing, one kind of value per column and consistent labels well, whichever tool holds them, and it reads a tab where the same level is typed five different ways badly in both. Airtable forces some of that structure on you through typed fields; Sheets lets you skip it, which is its charm and its weakness.

Follow me on Instagram@sagnikteaches

The other difference is how each tool's AI behaves once the data is in. Gemini's AI function in Sheets works cell by cell and waits for you to press a button. Airtable's field agents run per record and can re-run themselves when the underlying data changes. That one behaviour decides more real choices than any feature list.

Connect on LinkedInSagnik Bhattacharya

Prices, limits and AI features below are list prices in USD, checked on Airtable's and Google's own pages in September 2026. Both change often, so confirm them on the pricing page for your account before you commit.

Subscribe on YouTube@codingliquids

How Gemini in Sheets and Airtable's field agents read a row

In Google Sheets the AI function is written as =AI("prompt", range), or =Gemini(), and it needs an eligible Workspace or Google AI plan. On Workspace that means Business Standard (about $14 a user a month) or above, since Business Starter gets Gemini in Gmail and the Gemini app only. Google's help page sets out what it can and can't do:

• It returns text only, and it only sees the range you hand it, not the whole spreadsheet or other files in Drive.
• It generates when you select cells and click Generate and Insert, and only the first 350 selected cells run at once.
• It doesn't refresh on its own. If the source cell changes, you choose when to click Refresh and Insert.
• It can't be nested inside another formula, so =IF(AI(...)) won't work.
• There are short- and long-term generation limits; hit the long one and you wait 24 hours.

Airtable's version is the field agent: an AI-powered field that fills in a value for each record. It can write long text, pick a single or multiple select option, produce a number or currency value, suggest linked records, and read an attachment such as a PDF. Field agents need a paid plan. With "Run automatically" switched on, the agent regenerates whenever the fields it references change, and Airtable says it never overwrites a cell a person has edited. Owners and creators build the agents, often by asking Omni, Airtable's AI assistant, to do it; editors can press Generate.

Airtable charges in AI credits, pooled across the workspace. Team includes 15,000 credits per billable collaborator a month, self-serve Business 20,000 per paid user, and Free 500 per editor. Airtable's own examples put categorising a piece of feedback at about 1 credit, a question about your data at about 10, and analysing a 10-page contract at about 200. Building apps and agents with Omni costs nothing.

Airtable and Google Sheets compared on what AI needs

Two rows surprise people. Airtable's plan overview counts commenters as billable on Team, not just editors, so inviting the whole team "just to leave notes" can cost a seat each. And Sheets' generous size limit rarely matters for AI, because the AI function only works on the range you give it and 350 cells at a time; the practical ceiling is how much you're willing to click through.

A language school's enrolment data, tried in both tools

Take a language school with 180 active students, 22 classes across four levels, four teachers, an administrator and a director. It already pays for Google Workspace Business Standard. Its data lives in one spreadsheet: a Students tab with 31 columns, a Classes tab, and a separate attendance tab for each class, 22 in all.

The director wanted AI to do three jobs: every Monday, list students who missed two of their last four lessons and draft a friendly check-in message for each; suggest a level for new students from their placement score and the teacher's comment; and remind families about re-enrolment three weeks before term ends.

The first obstacle was the same in both tools. The level column held "B1", "b1", "Int", "Intermediate" and "B1+" for the same thing. Students in two classes appeared twice with different phone numbers. Attendance tabs had no student ID, only names, and two students shared a name. Four hours of clean-up came first either way: one Students table with an ID for each person, levels reduced to a fixed list, and attendance as a single long list (date, class, student ID, present or absent).

Staying in Sheets, the Monday job became a filter view of students with two or more absences in the last four lessons, about 15 a week, with an =AI column drafting the check-in. Cost: nothing extra. Effort: the administrator applies the filter, selects the 15 cells, clicks Generate and Insert and reads each draft, roughly 20 minutes a week. The re-enrolment reminder stayed a manual job.

In Airtable, Students, Classes and Attendance became linked tables. Only the administrator and director needed to edit; teachers log attendance through an Airtable form, and form submissions aren't billed. Two Team seats cost $40 a month on annual billing ($480 a year), with 30,000 AI credits a month between them. A field agent drafts the check-in whenever a student's absence count reaches two, and a second suggests the level on each new enrolment. Even at 10 credits a draft, 60 drafts a month is 600 credits, about 2% of the pool.

The school chose Airtable for students, classes and attendance, kept invoices in its accounting software, and left the price list and marketing calendar in Sheets, where they belong. The deciding factor wasn't AI quality, which was similar. It was that the Airtable version ran on Monday without anyone remembering to press a button.

One prompt in each tool, with the output and the fix

Sorting enquiries with the Sheets AI function

The administrator wanted enquiries from the website form sorted before anyone replied. The first attempt, in column D:

=AI("Classify this enquiry as new student, re-enrolment,
complaint or other. Reply with the label only.", C2)

Illustrative results down the column:

new student
Re-enrolment
re-enrolment (returning student)
complaint
Other - asking about parking

Five rows, four spellings of three labels. A filter on "re-enrolment" missed two of them. The fix was to name the exact labels and the exact format, then catch anything off-list with a check column:

=AI("Reply with exactly one of these labels, in lower case,
with no other words: new-student, re-enrolment, complaint,
other. Enquiry:", C2)

Check column:  =IF(COUNTIF({"new-student","re-enrolment",
"complaint","other"}, D2)=0, "CHECK", "")

After the change, one row in the next 40 still needed a human: an enquiry that was both a complaint and a re-enrolment request. That's the right outcome; a mixed message should reach a person.

Suggesting a level with an Airtable field agent

The level agent is a single select field with the options A1, A2, B1, B2, C1 and Needs review. Its first custom instructions read: "Using {Placement score} out of 100 and {Teacher comment}, choose the most suitable level." For a student scoring 58 with the comment "confident speaking, weak on past tenses", it chose B1, which the teachers agreed with.

The problem showed up a week later. A student scored 72 and the teacher hadn't written a comment yet; the agent chose B2 from the score alone, and the welcome email went out with the wrong class. The fix was one sentence, "If {Teacher comment} is empty, choose Needs review", plus leaving the comment field required rather than optional in the agent's settings, so the agent doesn't run on half the evidence.

Pointing ChatGPT or Claude at the data instead

There's a third route: leave the data where it is and let an assistant read it. Claude's Google Drive connector reads Sheets as text (charts and images are skipped), and ChatGPT's Google Drive app can search and open your files; on Claude's Team plan an owner has to switch the Google connectors on for the organisation first. Airtable runs an official MCP server, a standard way for AI assistants to plug into other software, which works with Claude, ChatGPT and others on every Airtable plan. It mirrors your own permissions, so an assistant connected by the administrator can read and change what the administrator can, and nothing more.

Structure shows up immediately here. The school's director connected Claude to the Airtable base and asked:

Which B1 students missed two or more of their last four lessons?
List name, class and the dates missed. Leave out anyone marked
as authorised absence, and tell me which table you used.

An illustrative answer:

Three B1 students match:
- [Student A] (B1 Tuesday evening): missed 9 Sep and 16 Sep
- [Student B] (B1 Saturday morning): missed 6, 13 and 20 Sep
- [Student C] (B1 Tuesday evening): missed 2 Sep and 16 Sep
Source: Attendance table, Status = Absent, last four lessons
per class.

Before the clean-up, the same question asked of the old spreadsheet counted late arrivals as absences in one class, because that teacher wrote "L" and another wrote "late (10 min)". In the linked base, Status is a single select with Present, Late, Absent and Authorised absence, so there's nothing to misread. Asking the assistant to name its source is what makes a wrong filter easy to spot.

Two cautions. An assistant that can update records can also update the wrong ones, so use the connection for questions first and keep changes with a person until you trust it; Airtable caps record creation at 10 per request anyway. And a connection to a whole base hands every table to the assistant, fees included, so choose which bases it can see in Airtable's third-party integrations settings.

Cases where a plain Google Sheet is the better home

• A single list with one owner. A physiotherapy clinic's referral log (date, referrer, reason, booked or not) has one editor and no links to anything. An =AI column that summarises each referral reason in five words is all the AI it needs.
• Reference data AI reads but nobody updates daily. A dental practice's treatment price list, uploaded to an assistant or shared with a chatbot tool, works perfectly well as a tidy sheet.
• One-off analysis. Exporting six months of bookings to find the quietest hours is a job for a sheet and a good prompt, not a new database. Practical uses of Gemini in Google Sheets has more of these.
• Everyone already lives in Workspace and nobody wants another login. Adoption beats architecture. A clean sheet people use is worth more than a clever base nobody opens.

Cases where Airtable pays for itself

• Records that belong to other records. A veterinary practice's reminders involve owners, pets (several per owner), vaccinations and due dates. In a sheet that's either one row per pet with owner details repeated, or lookups that break when someone sorts a tab. Linked tables hold it naturally, and a field agent can draft each reminder from the pet's record.
• AI that should run without anyone pressing a button. Any "when a new record arrives, work something out" job suits a field agent set to run automatically.
• Several people contributing, with some who shouldn't see everything. A form lets a nursery's room leaders submit daily notes without any access to the fees table, and form submissions aren't billed.
• Documents attached to records. A field agent can read an attached PDF, such as a referral letter or a signed form, and pull out the three fields you need.

If the jobs you want are bigger than a database (an app with screens, approvals and roles across Microsoft 365), Power Apps and when it replaces custom software covers that step up. And if your spreadsheets are already the bottleneck for the whole business, start with outgrowing spreadsheets before choosing a tool.

Restructuring a sheet before AI, or before a move

This is the same work whichever tool wins, and it's where most of the benefit comes from. In order:

1. Give every person or thing an ID. "S-0142" survives typos and shared names; "Sam K." doesn't.
2. Split cells that hold two facts. "Tues eve + Sat am" becomes two class enrolments.
3. Fix the labels. Pick one spelling for each status or level and turn the column into a dropdown (Sheets) or single select (Airtable).
4. Turn wide tabs into long lists. One attendance list with a date column beats 22 tabs, because AI and filters can read it in one pass.
5. Move free text to its own column. Notes go in a notes field, not in the status column.
6. Test on 20 rows before running AI on the lot, and read every output.

Here's one student before and after, as an illustration:

BEFORE (one spreadsheet row)
Name: Sam K. | Level: int/B1? | Classes: Tues eve + Sat am |
Notes: paid term 1, owes for books, mum prefers texts

AFTER (Airtable, or a tidy set of Sheets tabs)
Student   S-0142 | Level: B1 | Contact preference: Text
Enrolment E-0311 | S-0142 | B1 Tuesday evening
Enrolment E-0312 | S-0142 | B1 Saturday morning
Balance   S-0142 | Books | outstanding
Note      S-0142 | "Parent prefers texts"

The "after" version lets an AI answer "who at B1 owes for books and prefers texts?" in one step. The "before" version makes it guess what "int/B1?" means. Cleaning up customer records before you add AI goes further into de-duplicating contacts.

Costly slips with credits, seats and stale answers

• Running a new field agent across 5,000 records at once. If an agent uses 10 credits a run, that's 50,000 credits, more than three Team seats' monthly allowance, spent before anyone checks the output. Test on 20 records, read them, then widen.
• Inviting teachers or clinicians as commenters "to be safe". Airtable's plan overview counts each one as a billable Team seat at $20 to $24 a month. Forms and read-only access are free.
• Planning to upgrade to Business on Gmail addresses. Airtable's Business plan needs a private email domain.
• Trusting yesterday's =AI output. A Sheets AI cell keeps its old answer after the source changes until someone refreshes it. Add a "last generated" date column if the answers go to customers.
• Leaving Airtable's internet search on for a field agent that doesn't need it. Airtable notes that web lookups often use extra credits, and they bring in outside text you then have to verify.
• Letting the AI tool become the only copy. Whichever you choose, keep a regular export. Airtable's Free plan keeps only two weeks of revision history.

For most small businesses the honest answer is both: Sheets for lists and analysis, Airtable for the one or two datasets where records link together and AI should act without being asked. Start with the dataset that causes the most retyping, restructure it first, and pick the tool once you can see its shape.

Airtable and Sheets: questions owners ask next

Can ChatGPT or Claude read my Airtable base directly?

Yes. Airtable runs an official MCP server, available on all plans, that ChatGPT, Claude and other MCP-compatible assistants can connect to. It mirrors your own Airtable permissions, so an assistant can only read or change what you could. Creating records is limited to 10 per request, and calls count against Airtable's normal API rate limits.

What happens if we hit Airtable's record limit?

Airtable says you can keep using your bases and it never removes your data. You'll be told about the overage and can't add more records or attachments until you upgrade. The limits are per base: 1,000 records on Free, 50,000 on Team and 125,000 on Business, so splitting unrelated data into separate bases also helps.

Can we buy Airtable's Business plan with Gmail addresses?

No. Airtable's Business and Enterprise Scale plans need a private email domain, so free addresses such as Gmail or Yahoo can't be used to upgrade. The Team plan has no such rule. If you're heading for Business, set up email on your own domain first.

What if our Airtable AI credits run out mid-month?

Field agents and other AI actions stop until the credits reset or you buy more. Self-serve plans can add packs, starting at 10,000 extra credits for $20 a month. Building apps and agents with Omni doesn't use credits, so you can keep designing while you wait. Test a new field agent on 20 records before switching it on for a whole table.

Further reads

• No-Code AI Tools vs a Custom Build: When Templates Stop Working — When no-code tools stop being enough and a custom build pays.
• Is Your Business Data Ready for AI? A Clean-Up Checklist — A clean-up checklist to run before either tool.
• How to Use Claude With Excel and Google Sheets Safely — Connecting Claude to spreadsheets without leaking data.
• Can AI Analyse My Sales Spreadsheet? What to Upload and Check — What to upload and check when AI analyses a sheet.
• How to Organise Shared Files So AI Tools Can Use Them — The same tidy-up thinking applied to shared folders.
• How to Keep Customer Data Private When Your Team Uses AI — Keeping customer records private once AI can read them.
• The Complete Excel Guide 2026: Formulas, AI Copilot, and Modern M365 | Sagnik Bhattacharya — The Excel hub, with every spreadsheet tutorial on the site.
• How to Prepare Your Business Data for AI, Step by Step — Eight steps from scattered spreadsheets to data an AI tool reads correctly, with a photography studio's 1,400 client records as the example.
• What Business Data Should You Start Collecting Now for AI? — Seven datasets worth capturing from today (enquiries, quotes, job actuals, questions, complaints, prices, feedback), with the fields that make them usable.
• AI Agent Use Cases for Small Businesses: 10 Worked Examples — Ten AI agent jobs in six small businesses, each with the before and after, the tool, the monthly cost at list prices and what a person still has to check.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Airtable pricing page and help articles 'Airtable plans overview', 'Airtable AI billing', 'Using Airtable AI in fields' and 'Using the Airtable MCP server'; Google Docs Editors Help 'Use the AI function in Google Sheets'; Google Drive help on file size limits (all checked September 2026).