# Throttle, Do Not Block: Rate Limiting Is the Bot Control That Forgives Mistakes

Source: https://bottrafficindex.com/blog/2026-11-16-rate-limiting-bots-throttle-instead-of-block.html

November 16, 2026

*Blocking is binary, and binary controls fail binary: the day your blocklist catches a corporate network, a carrier NAT, or a search crawler having a bad week, the control that was saving you money starts costing you customers. There is a duller control that almost never makes that headline. Slow the traffic down.*

## Why automation hates a speed limit more than a wall

Most abusive automation is an economics problem wearing a technical costume. A price scraper polling your catalogue every minute earns its keep on freshness; throttled to a crawl, its data is stale and its operator's spreadsheet stops working. A credential stuffing run testing thousands of login pairs an hour depends on throughput; at a few attempts per minute per source, the same list takes weeks and the stolen credentials age out. A rate limit attacks the return on the operation rather than the identity of the visitor, which is why it keeps working against residential proxies and rotating addresses that made IP verdicts unreliable, the shift documented in [proxies that look local](2026-11-14-residential-proxies-bot-traffic-looks-local.md).

The standard shape is a budget per source per window, per endpoint: so many requests per minute from an IP, a session, or an account before responses slow, queue, or return 429. Endpoint choice is the craft. Login, search, add to cart, gift card balance, and discount code validation are the endpoints automation abuses, and each tolerates a different human ceiling. A human checks a gift card balance twice. A bot checks ten thousand. The gap between those numbers is so wide that a limit set anywhere inside it catches the abuse and never touches a shopper.

## Where throttling beats blocking, and where it does not

Throttling wins when identity is uncertain, which is most of the time: shared addresses, new visitors, traffic that smells automated but might be a comparison tool your customers like. It loses when the abuse is slow by design. A patient scraper polling each page once an hour sits under any sane limit while still copying the catalogue over a month; that problem wants the behaviour analysis and fingerprinting covered in [the fingerprint signals file](2026-11-15-headless-browser-fingerprint-signals.md), not a smaller number in the rate limiter. It also loses at the login door during an active credential stuffing spike, where the correct answer for known bad sources is still a block, applied fast and lifted when the spike passes.

The mature setup layers them in order of certainty. Rate limits everywhere abusive automation travels, because they are cheap and forgiving. Challenges for traffic that trips the limits repeatedly. Blocks reserved for sources you can name and prove: data centre ranges with no shopper story, agents that ignore robots.txt at volume, infrastructure from a documented attack. Each layer handles the confidence level it deserves, and no layer is asked to be certain when it is only suspicious.

## Set the numbers from your own logs, not a template

Pull the busiest legitimate hour your store has had, a sale day, and read the real per source request rates for your top endpoints. Set limits at a multiple of that human ceiling, three to five times is a sane start, because a limit set from a vendor template will either strangle your biggest fan or wave the bots through, and only your logs know which. Then alert on the limit itself: a rising count of throttled sources is an early warning of a scraping wave or an attack warming up, visible days before it shows in the analytics symptoms.

Review the throttled list weekly for false friends: the uptime monitor, the feed partner, the affiliate's price checker. Whitelist the ones with a business name and a person attached, in writing, with a review date. A whitelist that only grows becomes the hole the next scraper rents. The control is dull, the maintenance is duller, and between them they quietly do more work than any blocklist you have ever bought.

## Frequently asked questions

### What is rate limiting as a bot control?

A budget on how many requests a source can make to an endpoint in a time window. Past the budget, responses slow down, queue, or return a 429 status. Humans stay far below sane budgets; automation that depends on volume and freshness breaks against them.

### Why not just block bot IP addresses?

Because identity is often uncertain. Shared corporate addresses, carrier NATs, and residential proxies mean one address can hold both a bot and your best customer. Blocking is a verdict that misfires; throttling is a speed bump that costs a misidentified human a few seconds instead of the sale.

### Which endpoints should be rate limited first?

The ones automation abuses: login, search, add to cart, gift card balance checks, and discount code validation. Set each limit from your own logs, at several times the busiest legitimate rate you have observed for that endpoint.

### Does rate limiting stop scrapers completely?

No. It defeats volume dependent abuse like credential stuffing and fast scraping. A patient scraper working slowly under the limit needs behaviour analysis and fingerprinting instead. Rate limiting is the forgiving first layer, not the whole defence.
