Your AI Email Script Stops at 100 Sends a Day. Gmail Is Not the Thing Stopping It.
The Apps Script email quota is 100 recipients a day on Gmail and 1,500 on Workspace — tighter than Gmail's own limit. What each route past it costs.
A personalisation run that dies two-thirds of the way down the sheet does not look like a quota problem. It looks like a bug: ninety rows sent, three hundred untouched, an exception in the execution log, and nothing written down about where it stopped. The Apps Script email quota is 100 recipients a day on a gmail.com account and 1,500 on Google Workspace, and those numbers belong to Apps Script rather than to Gmail. The mailbox underneath will carry considerably more than the script is allowed to put through it. Whatever an AI email pipeline costs to run, the model is almost never what caps it — and the cap is a great deal cheaper to design around than to discover at volume.
The Apps Script email quota is the number that caps the pipeline
Google publishes the figure on its Apps Script quotas page, alongside every other per-service daily allowance. Email recipients per day: 100 on a consumer account such as gmail.com, 1,500 on a Google Workspace account. There is a second, more generous row for mail sent within your own domain — 2,000 a day on Workspace — which does nothing for outreach, since prospects are by definition outside your domain.
Three details about that number matter more than the number itself.
It counts recipients, not messages. One message addressed to three people spends three of the allowance. Google’s Workspace sending-limits page is the one that spells out what counts as a recipient — every address in the To, Cc and Bcc fields of a message, each time it is sent. A team that measures its send volume in messages is systematically under-counting its own consumption.
It is held per user, not per script. Google’s own wording is that quotas are set per user and reset 24 hours after the first request. A second script does not come with a second allowance, and neither does a second copy of the same script under a different project. If the same account is also running an appointment reminder automation sending from the same account, those reminders are drawing down the same 100 or 1,500 that the outreach is.
And the reset is not midnight. It falls 24 hours after the first request of the cycle, which means a run that starts at four in the afternoon has reset the clock to four in the afternoon. Teams that schedule a second batch for “tomorrow morning” on the assumption of a calendar-day boundary find the quota has not come back yet.
When the allowance runs out, the script does not degrade gracefully. Google states it plainly: exceed a quota and your script throws an exception and execution stops.
Why the AI step lowers the ceiling instead of raising it
This is where an AI pipeline is worse off than the mail merge it replaced, and the mechanism is not obvious.
Apps Script allows 50 recipients per message. A generic announcement can therefore reach all 100 of a consumer account’s daily recipients in two messages, and a Workspace account’s 1,500 in thirty. That allowance is the one structural piece of slack in the system — and personalisation forfeits every bit of it. A message written for one named person has exactly one recipient, by definition. The per-message allowance becomes dead weight the moment the body stops being identical, and the daily recipient quota is left as the only number that governs anything.
So the pipeline described in the Gmail and Google Sheets build for personalized cold email sequences is not merely subject to the same limits as a bulk send. It is subject to the strictest reading of them, permanently, and it gets there by doing the one thing that made it worth building.
Meanwhile the model call — the part everyone budgets for — is nowhere near a ceiling. Every outbound API request from Apps Script is a UrlFetch call, and a consumer account gets 20,000 of those a day against 100 permitted sends. That is roughly two hundred model calls available for every email the account is allowed to deliver. You could draft, critique, rewrite and score each message a dozen times over and still not approach the fetch quota. Whatever is expensive about automating personalized cold email sequences with AI, the drafting is not the thing rationing it. The constraint is the stamp, not the writing.
Gmail would carry more mail than your script will
The comparison that reframes this sits one documentation page away. On Workspace, Gmail’s own daily sending limit is 2,000 messages per user, with 10,000 total recipients a day and 3,000 of those permitted to be external. The limits run over a rolling 24-hour window rather than resetting at a set time of day.
Those units are not directly interchangeable with Apps Script’s — Gmail counts messages where Apps Script counts recipients — but for personalised outreach the two converge, because one message is one recipient. On that basis the mail layer will carry 2,000 a day and the script layer will carry 1,500. Google’s limits page points at the gap without quantifying it, in a single line: quotas might be different if you’re using Google Apps Script.
The practical consequence is a lever that does not work. A Workspace user stuck at 1,500 does not get past it by moving to a larger Workspace edition, because the ceiling is not coming from the edition. It is coming from the automation layer, and the only way through it is to stop sending from there.
What each route past the wall costs
Four routes, with the condition that comes attached to each.
A Workspace seat. This is the move from 100 to 1,500, and for most small teams it is the only one worth considering, because the mail keeps coming from and returning to the address the business already uses. Google’s Help Center prices Business Starter at $8.40 USD per user per month on the Flexible plan, or $7 USD per user per month on an annual commitment — $84 USD a year against $100.80 USD for the same seat bought month to month. Business Standard is $16.80 USD and $14 USD on the same two plans. The annual plan is cheaper and less forgiving: licences can only be reduced at renewal, and cancelling early still owes the balance of the contract.
The trap here is timing, and it is in the quota documentation rather than the billing documentation. A newly converted domain does not receive the full allowance on day one. Google states that account limits increase automatically only once the domain has cumulatively paid at least USD $100 and at least 60 days have passed since that threshold was reached. Buy a seat on Monday for a campaign on Tuesday and you may still be sending against trial-account limits.
Leaving Apps Script for the Gmail API. This costs nothing in licence fees and lifts the ceiling from Apps Script’s 1,500 to Gmail’s own 2,000 messages and 10,000 recipients a day. It is not free of work: you host and schedule the code yourself, handle OAuth yourself, and lose the convenience of a script that lives beside the spreadsheet. And it lifts the ceiling rather than removing it — Gmail’s account limits still apply, including the 3,000 external recipients a day that caps outbound regardless of how the send is made.
Amazon SES. Amazon’s own pricing page puts the Essentials plan at $0.16 per 1,000 emails for the first 10 million a month, falling to $0.14 and then $0.11 at higher volumes, with no per-account monthly fee. Pro is $0.22 per 1,000 plus $105 per account per region per month; Enterprise is $0.23 plus $500. There is also an à la carte option at $0.10 per 1,000 emails plus $0.12 per GB of attachment data — and that $0.10 figure is worth isolating, because it is the number most third-party writing still quotes as the SES price. It stopped being the headline rate when AWS introduced the named plans in July 2026. At a Workspace-equivalent volume of 1,500 a day across twenty working days, SES Essentials runs to about $4.80 a month.
Resend. $0 a month for 3,000 emails, capped — with some irony — at the same 100 a day a consumer Google account allows. Pro is $20 a month for 50,000 emails with no daily cap; Scale is $90 for 100,000. Dedicated IPs are a $30 a month add-on on Scale.
The four are not interchangeable, and the division is not about price. The first two send from the mailbox a reply arrives in. The last two do not: they are built for transactional mail from an application, and a reply to one goes wherever you have configured it to go, if anywhere. For outreach whose entire purpose is to start a conversation, that difference outweighs the per-email rate.
Build the send loop so hitting the quota is survivable
The quota is not really the problem. The problem is that most implementations treat it as impossible rather than as inevitable, so the first time it fires there is no record of what happened. Four changes fix that.
Ask before you start. MailApp.getRemainingDailyQuota() returns the number of recipients still available for the rest of the day. Read it before the loop, compare it against the rows left to process, and stop deliberately with a logged reason rather than running until Google throws.
Write the marker before the send, not after. Stamp the row as in-flight, send, then stamp the timestamp. A crash between the two leaves a row that is visibly ambiguous rather than invisibly unsent, and an ambiguous row can be inspected. A row that looks untouched after a successful send is how people get mailed twice.
Size the batch against the runtime, not the quota. Apps Script allows six minutes per execution. A row that makes one model call and one send at, say, four seconds apiece will clear roughly ninety rows before the execution limit ends the run — well short of a Workspace account’s 1,500. On a consumer account the runtime and the quota land in nearly the same place, which is coincidence rather than design; on Workspace the six-minute limit binds first by a wide margin, and a loop sized to the quota will be cut off by the clock every time.
Check which trigger you are on. Google notes that scripts running on a trigger have a lower daily limit for total computer time than those executed manually, and the quota table puts a consumer account’s total trigger runtime at 90 minutes a day against Workspace’s six hours. A scheduled overnight batch and the same code run by hand are not operating under the same ceiling.
The limit that bites before the quota does
For cold outreach specifically, the sending quota is rarely what stops you first. The deliverability rules bind well below 100 messages a day, and what they put at risk is the domain the business receives mail on.
Google’s email sender guidelines require SPF or DKIM of all senders, and SPF, DKIM and DMARC of bulk senders; sending to personal Gmail accounts requires a DKIM key of at least 1024 bits, with 2048 recommended. Unauthenticated mail may be marked as spam or rejected outright with a 5.7.26 error. Above 5,000 messages a day, marketing and subscribed mail must carry one-click unsubscribe through the List-Unsubscribe-Post and List-Unsubscribe headers.
The number that actually governs, though, is the spam rate reported in Postmaster Tools. Google’s guidance is to keep it below 0.10% and never to let it reach 0.30%. At 100 sends a day, that is a budget of roughly one complaint in a thousand messages — a handful of irritated recipients a month, and a smaller allowance than most senders assume they have. And if the plan was to tune the sequence on open rates, note that Google says it does not track open rates and cannot verify the figures third parties report for them, which makes it a poor instrument for exactly this decision. Reply rate is the signal that survives contact with the documentation, which is an argument for the lighter cadence that AI-generated first-draft follow-ups make practical.
Exceeding a Gmail sending limit blocks new sends for up to 24 hours, and Google is explicit that an account may be suspended, with accounts that send spam permanently restricted from sending. That is the same account the business receives its mail on.
Where this goes wrong
The failures here are specific, and none of them is exotic.
Buying Workspace the week of the campaign and expecting 1,500 immediately, when the domain is still on trial-account limits and will be for 60 days after it has paid USD $100.
Counting messages rather than recipients, which understates consumption by however many addresses are sitting in Cc and Bcc — and understates it invisibly, because nothing reports the discrepancy until the quota fires.
Assuming a second script, a second project or a second copy of the code carries its own allowance. The quota is per user. Splitting a campaign across several Google accounts to multiply it is not a workaround either; it is the pattern that gets accounts restricted.
Treating a transactional provider’s rate card as the whole answer. Both of the ones priced above will sell you the volume; neither price quoted here says anything about what either permits you to send, and the acceptable-use terms are the thing to read before the invoice is.
And one caveat that applies to every number in this article, printed by Google on the page itself: all quotas are subject to elimination, reduction, or change at any time, without notice. That is the argument for reading the remaining quota at runtime rather than hard-coding 100 or 1,500 anywhere in the script. A figure checked each morning cannot go stale; a constant written in March can, silently, and the first sign of it will be a run that stops two-thirds of the way down the sheet.
Sources
Every figure in this article traces to one of these. Dates are when the page was read.
- Quotas for Google ServicesGoogle for Developers — Apps ScriptAccessed Nov 20, 2026Official docs
- Gmail sending limits in Google WorkspaceGoogle Workspace HelpAccessed Nov 20, 2026Official docs
- Email sender guidelinesGmail HelpAccessed Nov 20, 2026Official docs
- Compare Flexible and Annual/Fixed-Term payment plansGoogle Workspace Help — Billing & subscriptionsAccessed Nov 20, 2026Official docs
- Amazon SES pricingAmazon Web ServicesAccessed Nov 20, 2026Vendor pricing
- Amazon SES introduces pricing plansAWS What's NewAccessed Nov 20, 2026Official docs
- Resend pricingResendAccessed Nov 20, 2026Vendor pricing
Written by
Full-stack developer who builds and runs AI automation systems in production. Runs local LLMs on personal hardware, builds N8N pipelines that actually ship, and deploys on Cloudflare Pages. Every guide on Pipeline Monk is tested on real consumer hardware — a Ryzen 7 5800HS with 16GB RAM and a GTX 1650. If it works on that, it works.