Split Testing Tool for Links: What to Demand Before You Pay

A split testing tool sends visitors to two or more versions of something and shows you which one performs better. For links, the split happens at the redirect, before any page loads. A link-level split testing tool needs no script on your site and works for destinations you don't control. That leaves two buying questions that matter more than the feature list: who stores the click data, and whether your tests keep running if you stop paying.

TL;DR:

Page-level tools split visitors with a script on your site. Link-level tools split them at the redirect, so they also work for pages you don't own.

A link split should use a 302 redirect, not a 301. Browsers cache 301s and can quietly send repeat visitors to the same variant without counting them.

Detecting a 10% relative lift on a 5% conversion rate takes roughly 31,000 visitors per variant. Most small tests are called far too early.

Before you buy, check for weighted splits, per-variant analytics, UTM pass-through, bot filtering, and a clean way to end a test without breaking the link.

A self-hosted tool keeps click data on your server and keeps links alive after you stop paying. The trade-off is that you are responsible for uptime and backups.

KORTLI

Own the data behind your link tests?

Kortli is a self-hosted link shortener for your own Linux server, bought once, so click data stays private and links keep working.

Visit →

Two Kinds of Split Testing Tool, and Why the Difference Decides Your Budget
How a Link Split Testing Tool Works: Weighted Redirects and Why 302 Matters
The Sample Size Problem Most Split Tests Ignore
What to Demand From a Split Testing Tool Before You Pay
Hosted vs. Self-Hosted: Who Owns the Test Data
Where Link Split Tests Pay Off: Email, QR Codes, Ads, and Bios
Run Split Tests on Links You Own, With Data That Stays on Your Server
Frequently Asked Questions
Sources

Two Kinds of Split Testing Tool, and Why the Difference Decides Your Budget

The phrase "split testing tool" covers two different products. They cost different amounts and solve different problems. Mixing them up is how people end up paying for a heavyweight experimentation platform when all they wanted was to send half their newsletter clicks to a new landing page.

Page-level tools run a JavaScript snippet on your website. When a visitor lands, the script decides which version of the page to show and swaps content in the browser. These tools are good for testing headlines, layouts, and checkout flows on a site you control. They also have three costs. You have to edit your site. You can get "flicker," where the original version flashes for a split second before the variant appears. And pricing is usually a monthly subscription tiered by traffic.

Link-level tools split traffic at the redirect. Someone clicks your short link, and the server sends them to destination A or destination B according to the weights you set. Nothing runs on the destination page. That means you can test two landing pages, two product listings, two signup forms, or two pages on someone else's platform entirely.

	Page-level script tool	Link-level, hosted	Link-level, self-hosted
Where the split happens	In the visitor's browser	At the provider's redirect server	At your own redirect server
Code on destination page	Required	None	None
Can test pages you don't own	No	Yes	Yes
Flicker risk	Yes	No	No
Typical pricing	Monthly, by traffic	Monthly, by links or clicks	One-time or infrastructure cost
Where click data lives	Vendor's servers	Vendor's servers	Your server
Links after you stop paying	Test stops	Often stop resolving	Keep working

Here is where we land. If you're optimizing the checkout on a high-traffic site you fully control, buy a page-level tool. For campaigns, email, QR codes, ads, and bio links, split at the link. It's cheaper, simpler, and it works on destinations you can't put a script on.

How a Link Split Testing Tool Works: Weighted Redirects and Why 302 Matters

Every link split test is built on one mechanism: an HTTP redirect whose target changes from click to click. The details of that redirect decide whether your numbers can be trusted.

Weighted assignment. You set the split, such as 50/50 or 80/20, and the server picks a destination for each click according to those weights. Uneven splits are useful when you want to expose only a small share of traffic to a risky variant.

Status code. RFC 9110, the IETF's HTTP Semantics standard, defines a 302 as a temporary redirect and a 301 as a permanent one. The same standard lists 301 as heuristically cacheable. A browser may store the 301 and skip your server on the next visit. A split link served with a 301 can therefore stop counting returning visitors and lock them into whichever variant they saw first.

A split link must answer with a 302. A cached 301 means the browser never asks your server again, and your test silently stops measuring.

Search engine behavior. Google Search Central's guidance on website testing says that if a test redirects users from an original URL to variation URLs, you should use a 302 rather than a 301. It also recommends using rel="canonical" on variants and running the experiment only as long as necessary. If either destination is a page you want indexed, follow that guidance. The redirect itself is only half the setup.

Sticky assignment. Some tools remember which variant a visitor received, usually with a cookie, so a second click lands on the same page. That keeps the experience consistent, but it adds a cookie you may need to disclose. Decide which behavior you want before you start, because switching mid-test muddies the data.

The Sample Size Problem Most Split Tests Ignore

The most expensive mistake in split testing isn't picking the wrong tool. It's declaring a winner after 400 clicks because one variant pulled ahead.

The standard sample-size formula for comparing two conversion rates (95% confidence, 80% power) gives sobering numbers. Suppose your baseline converts at 5% and you want to detect a lift to 6%, a 20% relative improvement. You need roughly 8,100 visitors per variant. To detect a lift from 5% to 5.5%, a 10% relative improvement, you need roughly 31,000 per variant. Evan Miller's sample size calculator will give you figures in the same range for your own baseline.

At a 5% conversion rate, spotting a 10% relative lift takes about 31,000 visitors per variant. That's 62,000 clicks before the answer means anything.

The second trap is peeking. In "How Not To Run an A/B Test," Evan Miller shows that repeatedly checking results and stopping as soon as they look significant inflates your false-positive rate well beyond the nominal 5%. Fix your sample size before launch, let the test reach it, and only then read the result.

Here's what that means for buying a tool. A split testing tool that shows a green "winner" badge after a few hundred clicks is not doing you a favor. Prefer a tool that shows raw counts per variant, so you can run the math yourself.

Pro Tip: Email security scanners and chat apps that build link previews often fetch every link in a message within seconds of delivery. Those fetches show up as clicks and spread evenly across your variants. That inflates both arms and drags the measured conversion rate toward zero. Judge link split tests on downstream conversions tagged per variant, not on raw clicks. If your tool lets you filter bot user agents or very fast repeat hits, turn that on before launch.

What to Demand From a Split Testing Tool Before You Pay

Feature pages all look alike. These are the specific capabilities that separate a usable link split testing tool from a toy.

Control over the split

Weighted variants beyond two. You should be able to run A/B/C with custom weights, not just a fixed 50/50.
Edit weights without changing the link. The short URL in your printed flyer or sent email must stay the same while you adjust traffic behind it.
A clean way to end the test. When a winner is clear, you should be able to send 100% of traffic to it in one change, with no broken links and no new URL.

Measurement you can trust

Per-variant click analytics. You need clicks by variant, by day, and by country, not just one combined total.
UTM pass-through or per-variant tagging. This lets your analytics platform attribute conversions to the right variant. Without it, you only know who clicked, not who bought.
Bot and scanner filtering. As covered above, automated fetches can quietly distort your results.
Raw data export. CSV or API access lets you run significance tests yourself instead of trusting a badge.

Ownership and continuity

Branded domains. A split link on your own domain earns more trust than a generic shortener domain, and it survives any vendor change.
Links that outlive the contract. Ask the vendor directly what happens to existing short links if you cancel. Our breakdown of what a Bitly subscription actually includes shows why this question deserves a written answer.

If a vendor can't show you per-variant raw counts and a way to end a test without a new URL, walk away. Nothing else on the feature page makes up for those two gaps.

Hosted vs. Self-Hosted: Who Owns the Test Data

Every click in a split test produces a log entry: timestamp, IP address, user agent, referrer, and the variant served. Where that log lives matters more than most buyers realize.

Privacy exposure. Recital 30 of the EU General Data Protection Regulation names IP addresses among the online identifiers that can make a person identifiable. If you use a hosted split testing tool, the vendor processes that data on your behalf. GDPR Article 28 then requires a data processing agreement between you and them. That isn't a reason to avoid hosted tools. It is extra paperwork and a third party in the data path.

Continuity risk. With a hosted tool, the redirect runs on someone else's infrastructure. Stop paying, and the link printed on 10,000 brochures may stop resolving or start pointing at the vendor's upgrade page. A split test is only as durable as the link it runs on.

Cost over time. Hosted pricing is recurring and usually rises with link count or click volume. Self-hosting costs a small virtual server plus your time. The math favors self-hosting more each year you keep it running.

Your responsibility. Self-hosting puts uptime, TLS certificates, backups, and updates on you. If no one on your team is comfortable with a Linux server, that cost is real. Our guide to running your own URL shortener lays out the setup and the risks honestly.

Our position: if you already manage a server or two, self-host. You keep the click data, links survive any billing change, and recurring fees go away. If you have zero server experience and no one to call, a hosted tool is the safer start. Just get its link-retention policy in writing.

Where Link Split Tests Pay Off: Email, QR Codes, Ads, and Bios

Link-level splitting does its best work in places where page-level tools can't reach.

Email campaigns. Put one split link in the newsletter and send traffic to two landing pages. You test the destination without building two email versions, and you can rebalance after the send.

Printed QR codes. Once a code is printed, it can't change. If it encodes a split link, you can test destinations long after printing and settle on the winner without reprinting anything. Our guide on creating a branded QR code that won't break covers why the code should always point at a short link you control.

Paid ads. Some ad platforms restrict how destinations can change after review. Check your platform's redirect policy before routing ad traffic through a split link.

Social bios. Most platforms allow a single profile link. A split link lets you test two offers behind it without touching your profile.

Destinations you don't own. Marketplace listings, booking pages, and hosted forms can't take your testing script. A redirect split is the only practical way to compare them.

For a primer on how the underlying redirect works, see how a URL shortener link works.

Run Split Tests on Links You Own, With Data That Stays on Your Server

Kortli is a self-hosted link shortener with A/B split testing built in. It installs on your own Linux server for a one-time license fee instead of a monthly subscription. You set weighted destinations behind a single short link on your own branded domain, with no per-domain fee. Click analytics and geographic data are recorded on your server, not a vendor's. When a test ends, you point the link at the winner, and the URL on your emails, ads, and printed QR codes keeps working. That stays true whether or not you ever pay us again.

The honest limit: Kortli counts clicks per variant, but it doesn't see what happens after the redirect. Tag each destination with UTM parameters so your analytics platform can attribute conversions, and run the sample-size math above before you call a winner. You also own the server, so backups and uptime are your job. Check our spec sheet against your traffic before you buy.

Frequently Asked Questions

What is the difference between a split testing tool and an A/B testing tool?

In practice, the terms mean the same thing: sending traffic to two or more variants and comparing results. Some people use "split testing" for tests that send visitors to entirely different URLs, and "A/B testing" for changes within a single page. For links, the split always happens at the redirect, so the two terms are interchangeable.

Should a split test link use a 301 or 302 redirect?

Use a 302. RFC 9110 defines 302 as a temporary redirect, while a 301 is permanent and can be cached by browsers. A cached 301 sends returning visitors straight to one variant without contacting your server. Google's website testing guidance also recommends 302 for test redirects.

How many clicks do I need before a link split test is reliable?

It depends on your baseline conversion rate and the lift you want to detect. At a 5% conversion rate, detecting a 20% relative lift takes roughly 8,100 visitors per variant. Detecting a 10% relative lift takes roughly 31,000 per variant. Set the target before launch and don't stop early just because one variant looks ahead.

Can I split test a page I don't control?

Yes, if you split at the link rather than on the page. A link-level split testing tool sends each click to one of several destinations through a redirect, so nothing needs to be installed on the destination. That makes it practical for marketplace listings, hosted forms, and booking pages.

They can, because click logs usually include IP addresses. GDPR Recital 30 lists IP addresses among the online identifiers that can identify a person. If a hosted vendor processes that data for you, Article 28 requires a data processing agreement. Self-hosting keeps the logs on your own infrastructure, although you remain the controller.