Rate limiting vs abuse prevention: what each one stops
Rate limiting vs abuse prevention, layer by layer: what each one stops, where the basic fixes break and how to keep free accounts off your paid capacity.
For a year you've been telling investors the same thing, and you were about to tell your board again on Tuesday. It's on slide 7 of the deck, the slide you like most: "Free tier: 12,400 developer accounts, up 31% this quarter."
A message lands from the CTO at your largest customer, who pays you $1,900 a month. "Our 10 a.m. batch sat in your queue for 15 minutes again. That's three days this week. Should we be looking at other options?"
You open the rate limiter first, because you set those limits so this couldn't happen. Free accounts get 5 requests a minute and 500 a day. The dashboard is green, with almost no 429s all week. By 6:15 p.m. you've pulled traffic by account, and there they are: 600 free accounts sending one request a minute, all through the working day. That's a fifth of the per-minute limit, and at 480 requests a day every one of them is under 500. The numbers are invented, so use your own. Say those 600 accounts belong to 30 people running 20 each, and each request costs you a tenth of a cent.
That's $8,640 over 30 days and $103,680 over a year. If you burn $90,000 a month, you've handed about 1.15 months of runway to 30 people who were never going to buy. Your limiter did exactly what you asked, which is why you're reading about rate limiting vs abuse prevention instead of finishing the deck.
The queue is where this reaches your customers. Your workers clear 1,200 jobs a minute. At the 9 a.m. peak your real users need 900 of those, 700 from paying customers and 200 from free ones. Those 600 accounts take demand to 1,500 a minute, so the line grows by 300 every minute. By 10 a.m. there are 18,000 jobs waiting (300 × 60), and 18,000 ÷ 1,200 is the 15 minutes your CTO just wrote to you about.
Slide 7 gets worse the longer you look at it. Your 12,400 accounts are 11,830 people (12,400 minus 600, plus 30). If your other 11,800 accounts average 12 requests a day, those 600 send two-thirds of the free tier's traffic: 288,000 of 429,600 requests. The 10% service credit you send the CTO costs $190. Losing them costs $22,800 a year ($1,900 × 12), and you'd planned to put their name on slide 9.
By 7 p.m. you're down to two bad options. You can buy 25% more capacity (1,500 ÷ 1,200) to serve 30 people faster. Or you can cut the free tier you've spent a year bragging about, and explain why to 11,800 real developers. Either way, someone on Tuesday's call is going to ask why the rate limits didn't stop this.
Rate limiting vs abuse prevention: two different questions
A rate limit asks whether one key, account or IP address is sending too much right now. It keeps a counter for each and returns a 429 when the count goes over. Abuse prevention asks something earlier that has nothing to do with speed: should this person get this free benefit at all, given what they've already had?
To a rate limiter, each account is its own counter starting at zero, so a new free account gets a fresh 5 a minute and a fresh 500 a day. One person with 20 accounts has 20 allowances, or 10,000 requests a day against the 500 you meant to give them.
| Rate limiting | Abuse prevention | |
|---|---|---|
| Question it asks | Is this key sending too much right now? | Has this person had this benefit already? |
| When it runs | On every request | When a benefit is claimed |
| What it stops | Bursts, runaway loops, retry storms | Extra accounts collecting a free allowance again |
| What gets past it | Many accounts, each under the limit | One account hammering your API |
The obvious fix is to tighten the limit, and it doesn't hold. Drop the per-minute limit from 5 to 1 and one request a minute still passes, while a developer testing an integration, who sends 20 requests in a minute and goes to lunch, gets nothing but 429s on their first day. Drop the daily limit from 500 to 100 and the 9,600 requests a person was getting (20 × 480) now take 96 accounts instead of 20 (9,600 ÷ 100). Your honest users lose four-fifths of their allowance, and the people you were aiming at open more accounts.
Each layer compared: what it stops and where it breaks
Every one of these catches something real.
Rate limits
Rate limits catch runaway loops, retry storms and one key scraping flat out, and every API needs them. They miss anything spread across accounts, they can hit real users in a busy hour, and they need retuning whenever a big customer's traffic changes.
Usage caps
A daily or monthly cap stops one account running away with your budget, and a hard spending cap belongs on any free plan. Each new account arrives with a fresh cap, though. The person who reaches one is often mid-project and close to paying, and you have to build whatever happens next.
Email checks
Email verification proves someone can open an inbox, which clears typos and dead addresses. It says nothing about a second real inbox, it loses people whose code lands in spam, and deliverability work never ends.
A unique-email rule catches the same address twice and rarely blocks anyone wrongly, but one mailbox can have many addresses. Gmail ignores dots, so sam.lee@gmail.com and samlee@gmail.com reach the same inbox, and mail to samlee+notes@gmail.com lands there too.
Disposable-domain blocks catch known throwaway services and miss new domains and ordinary mailboxes. They turn away people using privacy forwarding, and someone has to update the list forever.
CAPTCHA, phone and card
A CAPTCHA filters simple scripts, does nothing about someone signing up by hand and makes life harder for people using screen readers or privacy tools. A phone requirement stops casual second accounts, misses anyone with another number, shuts out people without a reliable mobile and charges you for every text. A card requirement stops people who don't have a second card. It also turns away the developer who wanted to try your API before talking to finance, and that developer is the reason slide 7 exists.
IP limits and VPN blocks
One signup per IP address catches repeats on one connection and misses every other network. It also lands on offices, campuses and mobile networks that share addresses, and each of those becomes an exception and a ticket. VPN blocks catch known VPN ranges and nothing off your list. They reject staff on company VPNs and give you another list to maintain.
Shorter trials and manual review
A smaller allowance or a shorter trial lowers your cost per account. More accounts make up the difference, every honest user pays for it and fewer people reach paying. Manual review catches what a person can spot, but not after hours and not when it's spread out slowly. At 2 minutes an account, 600 accounts is 20 hours, and whoever looks odd to a tired reviewer gets turned away.
Plot each layer and you can see the corner your counters never reach.
How to use both layers
Give paying customers their own capacity. A separate queue per plan, or a strict priority, removes the CTO's 15 minutes on its own.
Put one ceiling on all free traffic. With 1,200 jobs a minute and 700 paying at peak, free traffic gets at most 500. Paying customers keep their speed, but 800 free requests a minute now want those 500 slots, and 600 of the 800 belong to those 30 people. Three in four free slots still go to them, and your real free users do the waiting.
Keep per-key limits for bursts, and set them from real usage so the integration tester gets through.
Decide who gets the free benefit before you issue it. Treat a free API key, plan or credit grant as a benefit with a written rule, such as one account per person, and apply it when the key is issued.
The first three are code you can write this week. The fourth is harder. A query can tell you what an account has had, but your rule is about a person, and no layer above can tell you that account 20 and account 1 are the same one. A wrong guess costs you either way: you keep paying for the repeat, or you turn away a real developer who never says why.
Deciding whether a claim should get the free benefit
Whether a free claim should go through is the question AbuseZero answers, and no counter in this post can. Before you issue a free key, plan or batch of credits, you make one API call and get a clear allow or deny, under rules you control. There's also a proprietary algorithm, with AI that checks the claim again. You get multi-layer AI risk analysis that examines 140,000+ signals, detects linked claims and flags abuse that single checks miss, and every layer you just read about is a single check. Where the email layer ran out, AbuseZero checks mailbox aliases, disposable domains and mail routing. Where the IP layer ran out, it sees beyond an IP address with VPN, hosting, network reputation, Tor and proxy intelligence. It recognizes repeat devices across new accounts, even when email addresses keep changing. Creating a fresh account isn't enough to erase previous activity, and a fresh account is exactly what resets your rate limit. You don't have to build your own abuse prevention system.
Keep your rate limiter. It looks after the next minute, and AbuseZero's evaluation happens when a benefit is claimed. Observe mode shows you what would be denied before anyone is blocked, and the quickstart protects your first benefit.
Frequently asked questions
Is rate limiting enough to stop free tier abuse?
It stops bursts on one key. It doesn't stop one person using many accounts, because each account gets its own counter. See how to stop free tier abuse on an API product.
What's the difference between a rate limit and a usage cap?
A rate limit controls how fast an account can send requests over seconds or minutes. A usage cap controls how much it can use in a day or a month. Both count inside one account.
Can abuse prevention replace rate limiting?
No, you need both. Rate limits run on every request, and abuse prevention runs when a free benefit is claimed.
Protect your free tier before you have to cut it
Start with Builder
Pay as you go. Evaluate your first claim today.
See pricing
Builder, Pro, Scale and Enterprise.
Talk to us about Enterprise
High volumes and custom terms.