← Compendium

The link you don’t own

A short link like bit.ly/aB3x sits in posts, emails and forums nobody touches anymore, and it still leads exactly where it was once pointed. That depends on a third party hardly anyone still has in mind, because whoever set it up ten or fifteen years ago has often long left the company. The service still reads along with every click.

On 15 November 2026, a Burger King tweet from July 2011 suddenly led to a Wikipedia page that had nothing to do with the restaurant finder it advertised. The chain’s account had not been hacked, and nobody had touched the tweet itself. Bitly is a service that turns long web addresses into short links like bit.ly/aB3x and passes every click on them on to the real address. The Bitly link in the tweet had simply been registered again by someone else, and with that, this someone decided where a post from fifteen years ago sends its readers (Ransomnews, 2026). Bitly acknowledged that old, inactive links could be repurposed, suspended the accounts involved and disabled the affected links. Other old posts were hit as well, among them one from McDonald’s.

Whether this was an attack, a flaw in the system or both matters less to us. Something like this can hit any product, not just a global brand, because links like these also sit in your old newsletters, forum posts and on partner sites that nobody can edit anymore. The more interesting question lies behind it: How does a company end up hanging its reach on an address it does not control, and what is really inside such a tiny shortcut?

What short links were once for

There was a time when people still typed addresses by hand, hoarded bookmarks and even used services for sharing interesting ones. Short links arrived with the social networks that capped posts hard. Twitter initially allowed 140 characters, and when the address already ate half of them, there was hardly any room left for the actual message. A service like Bitly turned a long address into a few characters on its own short domain, and back then that was a real selling point.

That argument has disappeared, the technology has not. Today the big platforms shorten links themselves, with their own domains such as t.co at X, youtu.be at YouTube or lnkd.in at LinkedIn. Anyone who brings a mechanism in-house this consistently probably knows what is inside it. So let us first look at what actually happens when someone clicks a short link.

What happens on a click

A short link is not magic, it is a table. The browser sends the request to the service’s domain, the service looks the short code up in a database and answers with a redirect that contains the real address. Only then does the browser turn to the target. It takes a blink, and nobody notices.

Reader clicks the link Short-link service looks up the code sees every click Target page the long address ① Request ② Redirect ③ The browser fetches the target directly
The path of a click: request to the service, redirect with the target, then the actual fetch. The detour is what makes a short link convenient, and it is also the spot where someone reads along.

That detour is exactly the catch. Every click runs through a server that the sender of the link does not operate, and what happens there is up to its operator.

What becomes visible along the way

On every click, a short-link service gets to see more than it needs for redirecting: the time, the rough origin, browser and device, the page the click came from, and of course the target. On its own that is harmless, and for the operator it is routine anyway.

It gets interesting as soon as a service has many customers. The same person clicks a link in a news article in the morning, one from a newsletter at noon and one from a forum in the evening. To the person, these are three pieces of content. To the service that runs all three short links, it is a connection nobody else can see. Whoever recognises the person, for example through a cookie, the IP address or browser characteristics, can link their clicks across all customers into one profile.

the same person one short-link service, many customers short.link/aB3x Post by a restaurant chain short.link/Q9wz Newspaper article short.link/mK2p Shop newsletter one profile interests, times, devices three senders, one picture
Each sender only sees their own link. Only the service that runs all three can join the clicks of the same person.

Profiles like these are a sought-after raw material for the advertising industry, and it is plausible that services offering short links for free or very cheaply fund themselves at least in part this way. We cannot prove this for any single provider, and that is not the point. The point is the logic of the model: whoever operates the redirect sits exactly where this data arises, and what they do with it is written in their privacy policy and not in your hands.

The dashboards such services give their customers fit this picture. Reach and clicks by country and device are useful to customers. Because they consider these numbers credible and report it when something looks wrong, the data behind them gets better at the same time, and with it the value of the profiles that arise on the side.

The usual reflex goes: I don’t care, what could anyone do with my clicks? For a single click that is even true. But volume counts, and above all duration, because short links sit in posts that still draw reach years later. Which brings us to what happened in November.

Links age more slowly than contracts

A post from 2011 is still being shared, quoted and rediscovered in 2026. The link inside it was created back then without much thought, and nobody ever touched it again. Who owns it today?

Technically, the service does. It decides how long a code stays valid, whether it becomes free after an account is closed, and whether the service eventually changes its pricing, gets acquired or shuts down. Handing out an address that was considered unique to someone else is the most unpleasant case, because suddenly a stranger decides the target of a link you distributed years ago. According to Ransomnews, Bitly’s help pages rule out reuse, and a bug that allowed it was reportedly fixed as early as February 2025. So the provider’s promise was there, and it still was not enough.

What makes this treacherous is how quiet such a failure stays. If a link points to a page of your own that is broken, your monitoring shows it and you can act. If it points to someone else’s page instead, the visitor never reaches you at all. Your own numbers can then tell you at most that traffic is dropping away, but not where it goes.

A provider’s promise is not control.

Anyone who outsources a part of their operations hands over not just work but also the final say over what happens in that part. That is no criticism of outsourcing as such. Hardly anyone builds their own mail system or their own payment processing. What remains is the question of how essential the outsourced part is and how hard it would be to run yourself. With short links, the second half of that question deserves a look.

You need redirects anyway

Anyone who runs a digital product will eventually move it onto a new foundation: another shop system, a new platform, from a monolith into separate services, into the cloud. In migrations like these, addresses change almost incidentally, so you need the ability to redirect precisely anyway. Old addresses with good search engine coverage should point to a new product, the new structure must not cost reach, and forums, social networks and emails hold links you can never edit again. Redirects are a small part of the migration, but one that decides whether your reach moves along.

What this looks like when moving to a new platform is covered in the article Moving house without losing your reach. It deals with inventory, mapping and the choice between 301, 302 and 410, and these very building blocks are also the foundation of your own short-link service.

Monitoring is part of this. If an address returns a 404 or even a 500, that is no side issue but a visitor running into nothing right now, possibly through a link that took years to build. The error rate of exactly these redirect addresses belongs on a dashboard someone actually looks at, with an alert on unusual spikes going to the phone of whoever is responsible.

Once you run this capability cleanly, short links come almost for free. All it takes is one additional short domain pointing at the same redirect logic. A long address like pos-for-cafes.com/guides/keeping-a-cash-book then becomes pfc.com/cashbook, an invented domain that serves only as an illustration. And since the operation is in your hands, you also decide what gets counted. Anonymous numbers on requests and errors give a good picture without anyone having to be recognised.

How do you build this?

That depends on how often the addresses change and how many there are. Three orders of magnitude cover most cases, and that holds just as well when you rebuild an existing service.

  • Up to roughly a hundred addresses that barely change, for example after a platform migration: the old sitemap provides the list, a mapping to the new addresses lives as data in the code, and the redirects are generated from it and set up in the web server. A test checks that every address arrives.
  • Considerably more addresses that keep being added: Redis as a fast key-value store is often a good choice here. It answers very quickly, but it has to be operated as a service, secured and sized properly.
  • Very many addresses, and you are on AWS anyway: Amazon S3 together with CloudFront, as described below.

The third way sounds unusual and is surprisingly calm in operation. For every short link you create an empty file in an S3 bucket, named after the short code, and put the redirect target into its metadata. In CloudFront you set up an origin group, a fixed order of sources that asks the bucket first and only falls back to the actual website if there is no entry. The answer arrives within milliseconds, caches extremely well and costs next to nothing to run.

Request /cashbook CloudFront keeps answers cached, in milliseconds ① S3 bucket one empty file per short link, the target sits in the metadata ② The actual website only if S3 does not know the code ask first otherwise go on
CloudFront asks the bucket for the short code first. Only if nothing is stored there does the actual website serve the request, so normal page requests keep working unchanged.

When your own service wobbles

Many teams already run their own redirect service, often grown out of a small script, a table in the main database or a Redis next to the application. For a long time it runs quietly, until a campaign, a viral post or a newsletter to every customer bundles the clicks and the redirects fail exactly when most visitors arrive. Of all things, the simplest part of the system becomes the bottleneck.

That is usually because a redirect takes the path of a full request, although it is only a lookup. Every click passes through the application, queries the database or cache and waits for the answer, and since redirects are rarely cached, this repeats for every single visitor. The service therefore grows with the clicks and not with the number of links, which hardly changes. If it also shares a database or server with the actual product, it drags the product down with it at peak load.

The remedy turns this relationship around. Because the mapping changes rarely and is read very often, it belongs as close to the edge of the network as possible, where the answer comes from the cache and the application never even sees the click. That is exactly what the route via S3 and CloudFront does, and if you rebuild a wobbly service this way, you do not have to change the mapping itself, only the place it is served from.

In a customer project we created several million redirect files this way. Creating them cost a little over 300 US dollars once, and running costs afterwards were practically zero. It replaced a Redis cluster that worked at the limit of its gigabit network connection and kept failing under load because it was overloaded. Speed, stability and cost improved at the same time, which rarely happens in engineering.

At this volume nobody maintains things by hand anymore, and that did not work in Redis either. That is why there is also a text version of the mapping that lives offline: a file you can search, filter and change in bulk with ordinary tools. The files in the bucket are generated from it. It is the actual source of truth, and the bucket is only its image.

# short code    target
cashbook        https://pos-for-cafes.com/guides/keeping-a-cash-book
demo            https://pos-for-cafes.com/demo
news            https://pos-for-cafes.com/news
The mapping as a plain text file, versioned and searchable with ordinary tools. The files in the bucket are generated from it, not the other way round.

And one more thing changes: the short domain becomes a treasure itself. Whoever takes over its access at the domain provider (the registrar) or its DNS takes over every link beneath it. Why this layer is so often underestimated is covered in the article DNS is massively underestimated.

What this means for you

Whether you buy a part of your operations or build it yourself is not a matter of faith. A useful question is: What happens to everything I have distributed over the last years if this provider changes its rules tomorrow, gets sold or disappears? If the answer is “then my links are dead or belong to someone else”, that part is essential. And a redirect service is small enough to run yourself.

You do not have to overturn anything at once. New links will be created under your own domain from now on, existing ones stay where they are, and you go through them step by step: Which ones still bring visitors today, which sit in places that matter to you? Only for those is the move worth it.

For us, redirects are a small but important building block of platform migrations, where a product moves onto a new technical foundation and neither reach nor data nor availability should be left behind. If such a step is ahead of you, we are happy to look together at what has to come along.

Frequently asked questions

What role do redirects play in a platform migration?

They are a small but important part. When a product moves to a new platform, addresses often change on the side, and old links in search engines, newsletters, forums and partner sites would otherwise lead nowhere. An inventory of the old addresses, a mapping kept as data and monitoring of the error rate keep your reach intact. Alongside them, data, operations and deployments have to move along just as much.

When does it pay to run a service yourself instead of buying it?

When the part is essential and can be run yourself with manageable effort. A useful question is: What happens to everything I have distributed so far if the provider changes its rules, gets sold or disappears? If the answer is dead or foreign links, the part is essential. A redirect service is small enough to run yourself.

What is the difference between 301 and 302 for short links?

A 301 means “moved permanently”, browsers often remember it without an expiry date, and later clicks may no longer reach the server. A 302 means “temporary”, every click runs through the server and can be counted and redirected. If the target is final and search engines should carry over the signals, the 301 fits, otherwise usually the 302.

Do short-link services make money from my data?

That cannot be said in general for any single provider. Structurally, though, a service sits exactly where origin, device and target of every click arise, and it can join clicks across many customers. What it does with them is stated in its privacy policy, which is worth reading before you entrust it with important links.

Can a short link really be handed out again?

In principle that should not happen. In November 2026, however, it was reported that old, inactive Bitly links were registered again and pointed at foreign targets, including links in tweets from 2011. Bitly confirmed the problem, suspended accounts and disabled links. The case shows that a provider’s promise does not replace your own control.

What does running your own redirect service cost?

For a handful of addresses practically nothing, because they sit in the web server configuration. With S3 and CloudFront, creating the files costs a small one-off amount per million files, and running it usually stays within the free tier. In a customer project with several million redirects, creating them cost a little over 300 US dollars once, and operation afterwards was practically zero.

Our own redirect service falls over under load. What can we do?

Usually every redirect runs as a full request through application and database, although it is only a lookup, and it is not cached. The load then grows with the clicks. The fix is to move the mapping to the edge of the network, for example with S3 and CloudFront, so answers come from the cache and the application is relieved. The mapping itself stays the same.

Do I need my own short domain for this?

No, the redirects also work under your main domain. A short extra domain only makes sense if the address is actually supposed to get shorter, for example for social media, print or QR codes. It then points at the same redirect logic.

What do I need to watch out for security-wise with my own redirect service?

The service must never take a target from the request, for example via a parameter like ?url=, otherwise the domain becomes a springboard for phishing. It knows only codes from its own table and redirects only to verified targets. You should also protect the registrar access and DNS of the short domain especially well, because whoever takes them over takes over every link beneath them.

How do I monitor my redirects?

Watch the error rate of the redirect addresses, meaning responses with 404 and 5xx, and let an alert reach the responsible person on spikes. A regular test adds that every address in the mapping arrives and that the target actually exists.

What do I do with my existing short links at a provider?

Start with an inventory: Which links exist, which still bring visitors today, which sit in places that matter to you? Only for those is the move worth it. From now on, create new links under your own domain, and if you stay with the provider, do so with your own domain so that the address belongs to you.

Related topics