Airtable AI Credits: Estimate What Your AI Fields Will Cost a Month

One Airtable AI field costs roughly 1 credit a record for simple tagging, 10 for a web lookup and about 200 for a 10-page document. Multiply by your monthly records. Free gives 500 credits per editor, Team 15,000 per paid user and Business 20,000, all pooled. Extra packs start at $20 a month for 10,000.

The per-record figures are Airtable's typical examples, not guarantees. Cost depends on the model chosen, how much text goes in and how long the answer is, so the right method is to estimate with the typical numbers, test on 20 records, then multiply by what you measured. The sections below do that with one realistic field, then work out which plan covers it.

Follow me on Instagram@sagnikteaches

What Airtable counts and what it lets you do for nothing

Airtable's AI billing article lists the actions that consume credits: deploying field agents (the AI-powered fields in a table), running AI automations, asking deep research questions, categorising feedback, analysing documents and generating images. Text pulled in from the internet when web access is on is charged too.

Connect on LinkedInSagnik Bhattacharya

Some things are free. Building a base or editing one with Omni, Airtable's built-in assistant, costs nothing, and neither does creating or updating records through it, such as importing a CSV or cleaning existing data. Asking Omni a question about the data in a base is listed at 10 credits. So the design work is free, and the running costs come from the fields, automations and questions that keep firing afterwards.

Subscribe on YouTube@sagnikteaches

Airtable publishes a short table of typical credit costs. Use it as a starting point.

For larger text, Airtable gives rough ranges: a 10-page contract is about 200 credits, a document of 50 or more pages (25,000 words or more) can use 500 to 1,500, and processing 100,000 words or more can run to 2,000 to 5,000 or beyond. Its advice is to start with smaller test runs to see real consumption before processing large volumes, and that is the advice this tutorial follows.

What each plan includes, and how pooling changes the arithmetic

All the figures come from Airtable's pricing page and its AI billing article.

Monthly billing costs more: $24 a seat on Team and $54 on Business, according to the pricing page. Credits are pooled. On Free and Team the pot is shared across the workspace, and on Business and Enterprise Scale across the organisation. So a Team workspace with four paid editors holds 4 x 15,000 = 60,000 credits for everyone, and the person who builds the AI field is not limited to their own 15,000.

Two details can trip up an estimate. First, only billable collaborators, the people with edit permission, bring credits. Read-only collaborators, form submissions and share links are not charged and add nothing to the pot. Second, Free plans give 500 credits to each editor, with up to five editors, so a Free workspace of five editors holds at most 2,500 credits.

Seat counting is covered in Airtable Permission Levels: Creator, Editor or Commenter for Staff?, which matters because demoting a rarely-used editor to a commenter removes both a $20 seat and 15,000 credits from the pool.

Estimating one field, step by step

The method has five steps and takes about half an hour.

1. Write down what the field does in one sentence, and which action in the table above it is closest to (categorise, summarise, search, analyse a document).
2. Count how many records will trigger it each month, from your current volume, not a hopeful one.
3. Find how often it re-runs per record. A field that runs once per new record is cheaper than one that runs again whenever someone edits the inputs. Test this by editing a source cell on one record and watching whether credits drop.
4. Measure the credits on 20 real records and take the average, rather than relying on Airtable's typical figure.
5. Multiply, add 25% as a buffer, and compare with the plan's pooled credits.

A summary of the arithmetic is: monthly credits = records x runs per record x average credits per run x 1.25.

The main example: a translation agency's enquiry table

Consider a small translation agency with four people who can edit its Airtable base. Each month about 900 enquiries land in a Requests table, from a web form and forwarded emails. The owner wants one AI field that tags each enquiry by language pair and urgency, and a second step that analyses the client's attached brief for the ones that become quotes. Here is the estimate, with illustrative numbers.

The Omni line assumes four people each ask six questions a working day for 21 days, which is 504 questions. It is the line most owners forget, and in this example it is the second largest.

Against the plans: on Free, four editors hold 2,000 credits, far below 15,140. On Team, four paid seats hold 60,000, so the total uses about 25% of the pot. The Team plan costs 4 x $20 = $80 a month on annual billing, so the AI part, at 15,140 credits, adds no cost beyond the seats. Credits would only become a cost question if usage passed 60,000.

Business is not needed for credits here. Its extra 5,000 credits per seat only matters if you would otherwise buy packs, and $45 against $20 is $25 more per seat for 5,000 more credits, a poor price for credits alone. The reasons to choose Business are records, admin controls and security, covered in Airtable Team vs Business Plan: Records, AI Credits and Admin Controls.

When one person runs the whole thing

Now take the same workload at a one-editor business, where that one person asks every Omni question. A single Team seat gives 15,000 credits, and the estimate of 15,140 (without the buffer) slightly exceeds it. Options and their monthly costs:

The pack prices come from Airtable's billing article: 10,000 credits are $20 a month or $200 a year, 20,000 are $40 or $400, 50,000 are $100 or $1,000, 100,000 are $200 or $2,000, and larger packs follow the same $0.002-a-credit pattern up to 400,000 for $800 a month. That makes the per-credit price of extras $0.002 whatever the pack size, so there is no volume discount to chase, and buying the smallest pack that covers your gap is enough.

Compare a seat with a pack: an extra Team seat is $20 for 15,000 credits, about $0.0013 a credit, cheaper than a pack, but only if the person truly needs edit rights. Buying seats to harvest credits is a poor reason to pay for them, and on Team and Business Airtable charges for everyone with edit permission on at least one base in the workspace, so a seat added for credits stays on the bill.

The re-run trap, shown with the contract field

The biggest estimate error comes from forgetting that a field can run more than once for the same record. Suppose the brief-analysis field in the agency's table reads three input fields: the attachment, a notes cell and a language-pair cell. If staff edit the notes cell twice while preparing a quote and the field recomputes each time, the 200-credit run happens three times.

Brief analysis, as estimated:   40 records x 1 run  x 200 = 8,000 credits
Brief analysis, with re-runs:   40 records x 3 runs x 200 = 24,000 credits
Difference:                     16,000 credits, or $32 if bought as packs

That extra 16,000 would push the one-editor business well past its 15,000 allowance. I cannot tell you from outside whether your field recomputes on every edit, because it depends on how it is configured and how Airtable treats the dependency, so test it. Edit one source cell on a record and read the credit counter on the workspace settings page before and after. Airtable also warns builders when an AI field or automation action will use a significant amount of credits or keep running over time, so read those warnings rather than clicking through them.

The cure is a manual trigger. Use a checkbox or button field that must be ticked before the expensive analysis runs, so it runs once per quote and only when someone decides.

Two small examples with sample output

Here is a cheap field, with the kind of instruction you would give it and an illustrative result.

Instruction for the "Request tag" AI field:
Read the Enquiry text. Reply with two labels separated by a comma:
the language pair (for example "French to English") and an urgency of
"urgent" if a deadline within 48 hours is mentioned, otherwise "normal".
If either is unclear, write "check" for that label.

Enquiry: "Hi, we need a 4-page supplier agreement from German into
English by Thursday, please can you quote?"

Illustrative output:
German to English, normal

The output shows the sort of mistake to catch: the enquiry says "by Thursday", which might be inside 48 hours, but the model had no date to compare against, so it said "normal". Fix it by including the received date in the instruction, or by letting the field return the stated deadline as text and having a person set urgency. A one-credit field is worth a few minutes of testing.

A more expensive example: a distributor adds a field that searches the web for each new customer's company and writes one line.

Records: 60 new customers a month
Credits: about 10 per web search
Monthly cost: 60 x 10 = 600 credits

That is 4% of one Team seat's allowance and costs nothing extra. If the same field also runs on the old customer list of 1,500 records during setup, that is 15,000 credits, a month's allowance in one afternoon. Run lookups on the new records first, and process old records in batches of 100 while watching the counter.

Three quick sums for other kinds of business

A courier firm tagging delivery complaints. It receives 3,000 complaint messages a month and tags each with a category at about 1 credit. That is 3,000 credits, a fifth of a single Team seat, for a field that sorts the inbox by cause. Add a weekly Omni question, "which depot had the most late-delivery complaints this week?", at 10 credits, and it is 40 credits a month, which is rounding error.

A small manufacturer checking supplier certificates. It uploads 25 supplier certificates a month, each 5 pages, and asks the field to extract the expiry date. If a 10-page document costs about 200 credits, a 5-page one might be roughly half, so around 100, but treat that as a guess and test it. 25 x 100 = 2,500 credits, about a sixth of one Team seat's pool.

A wholesale distributor summarising 400-word supplier emails. Summaries of short text sit near the cheap end. If a test shows 3 credits per email and 2,000 emails come in a month, the field uses 6,000 credits. Two editors on Team hold 30,000, so it fits. If a manager asks to summarise the whole backlog of 40,000 old emails, the same field would use 120,000 credits, four months of the whole team's pool, so the backlog has to be sampled, not processed.

How to check your estimate against reality

Run the field on 20 records, then read the credit counter. For a Team plan, workspace owners can view used and available credits on the workspace settings page under AI credits, and Business and Enterprise Scale admins can view them in the admin panel under Settings, AI settings and View AI credit usage. A before-and-after reading around the test gives you the real average.

Credits before test:      14,310 remaining
Credits after 20 records: 14,090 remaining
Used:                     220
Average per record:       11
Estimate had assumed:     10
Verdict:                  close enough; keep the 25% buffer

If the average is more than twice your estimate, look for a larger model setting, a long prompt, web access switched on or text pulled in from many linked records. Each of those inflates input size. Shorten the instruction, link fewer fields into the input and turn web access off for jobs that do not need it.

Then set a monthly habit. On the first working day, read the previous month's usage, compare it with the estimate and adjust. Airtable emails when you pass 75% of your limit, but a workspace that hits 75% on the 10th of the month has no time to react, so set your own mid-month check.

Guardrails that keep a credit bill from surprising you

Estimates drift, so build four small controls into the base before the field goes live.

1. A trigger checkbox on expensive fields. Name it "Run analysis" and make the automation or field wait for it. A 200-credit job should never start because someone pasted a file into the wrong row.
2. A view called "AI queue". Filter to records where the AI output is empty and the trigger is ticked, so you can see in one glance how many expensive runs are pending. If 60 records are waiting and each costs 200 credits, that is 12,000 credits you can still stop.
3. A named owner for the credit counter. Only workspace owners can read usage on the settings page, so decide who checks it, on which day, and what they do at 50% used by the 15th.
4. A written cap per field. Write "tagging: 1,000 credits, brief analysis: 9,000, Omni questions: 5,000" in a notes table. When a field passes its cap, someone asks why before it passes the pot.

A short before and after shows the effect. Before: the brief-analysis field ran whenever an attachment appeared, including duplicates and test files, and it used 24,000 credits in a month. After: it ran only when "Run analysis" was ticked, a person checked the attachment first, and its use fell to 8,400 credits for 42 genuine briefs. The change took ten minutes to build and removed the need for a $40 pack.

Free-plan reality check

A Free workspace gives 500 credits per editor and above. With two editors, that is 1,000 credits a month, enough for 1,000 one-credit tags and nothing more. The Free plan also caps a base at 1,000 records, 1 GB of attachments and 100 automation runs, so a base that gets heavy enough to need AI fields usually hits one of those limits first. If it does, the next step is Team, as Airtable's billing article confirms that Free customers needing more credits must upgrade. Airtable Pricing Plans: Is Free Enough for Bookings and Quotes? covers where the Free plan stops.

Choosing the plan that covers it

Use this short rule once you have a buffered monthly total.

• Under 500 credits per editor (2,500 at most, since Free allows five editors): Free can cover it if you stay within 1,000 records and 100 automation runs.
• Up to 15,000 times your number of paid editors: Team. Credits are rarely the reason to move up.
• Above that, with more than one editor: Add a pack at $0.002 a credit before you upgrade to Business, unless you need Business for other reasons.
• Records above 50,000 in a base, or a need for AI admin controls and an admin panel: Business, with credits as a bonus rather than the cause.

If you are still deciding whether to build the AI field at all, Airtable AI Fields: Auto-Summarise and Tag Incoming Quote Requests shows a working build, and Airtable Automations for Small Business: Six Examples to Copy shows which automations can call AI actions. If you would rather map which jobs in your own base deserve AI credits with someone, that is what my AI implementation consultation covers, and you can reach me through the contact page.

What owners ask when the credit warning email arrives

Do unused Airtable AI credits roll over to next month?

No. Airtable's billing article says the free credits expire at the end of the month, and a self-serve plan's allowance resets at the start of its billing cycle. Plan to use what you get each month rather than saving it, and set your first alert at the 75% email Airtable sends.

Are credits per person or shared across the team?

Shared. Airtable says credits are pooled across all users in a workspace on Free and Team, and across the organisation on Business and Enterprise Scale. One heavy user can therefore draw on the allowance that quieter colleagues bring, which is why a five-editor team gets a single large pot.

Do viewers or read-only people use credits?

Collaborators with Viewer permission get no allocated credits, though Airtable says they can use credits through Omni, its AI assistant. Seats are charged for people who can edit, so check who has editing rights before you assume a credit count.

Does asking Omni to build a base cost credits?

Airtable lists building apps and agents with Omni, and creating or updating records through it, as free. Asking Omni a question about your data is listed at about 10 credits. So designing the base costs nothing, and the running cost comes from fields, automations and questions.

How do I see what I have used?

Workspace owners can open the workspace settings page for each workspace and read used and available credits. On Business and Enterprise Scale, admins can view usage in the admin panel under AI settings. Airtable emails an alert when a workspace or organisation passes 75% of its limit.

Further reads

• Airtable vs Notion as a Business Database: Which Handles AI Better? — How Notion's AI allowance compares for a small database
• Airtable for Startups: Build an AI-Assisted Operations Base — A whole operations base with AI built in, for a young firm
• Airtable CRM Setup for a Service Business: Tables, Views and Alerts — A CRM base where an AI summary field earns its credits
• Airtable vs AppSheet for a Job Tracker: Cost, Offline Use and AI — A six-person trade firm's job tracker priced on Airtable and AppSheet, with offline behaviour, record limits and AI features compared.
• 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 Airtable support article on AI billing and credits, read in November 2026; credit consumption figures are Airtable's typical examples, not guarantees.