GoWWW All articles
Internet Culture

Your Next Visitor Has No Eyes: Building Short Links That Machines Can Actually Follow

GoWWW
Your Next Visitor Has No Eyes: Building Short Links That Machines Can Actually Follow

Photo: abstract server room network cables blue light data center, via images.fineartamerica.com

There was a stretch, not long ago, when you could reasonably assume a click came from a person. A thumb, a trackpad, a momentary lapse of judgment on a Tuesday afternoon. That assumption is now a rounding error.

In 2026, a healthy chunk of requests to any public short domain arrives from something summarizing a page, filling out a comparison table, monitoring a price, or assembling a research brief on behalf of a human who will read the output and never touch the link. Your redirect layer is no longer just a front door for people. It is an API that occasionally gets read by a browser.

Most short-link infrastructure was never designed for that.

The redirect chain is now a contract

HTTP status codes were always a form of documentation, but agents read them literally in a way humans never did.

A 301 tells a machine the move is permanent. Per spec, that answer is cacheable indefinitely unless you say otherwise, and plenty of agent frameworks will hold onto it for the duration of a session or longer. If you habitually use 301s because they look tidier in DevTools, and then repoint a slug during a product launch or a post-incident scramble, you have created a class of visitor that will keep landing on the old destination and confidently reporting on it.

A 302 or 307 says this is temporary. For anything you might ever retarget, that is the honest answer. Pair it with explicit Cache-Control headers so the interpretation is not left to a library's default.

A 308 is the one that bites quietly. It preserves the request method. Most crawlers handle it fine. A surprising number of homegrown fetchers do not, and will fall back to a GET or bail entirely.

The other half of the contract is the method. Agents routinely send HEAD before GET to check whether a link resolves, how large the payload is, and what content type to expect. If your shortener returns a 405 for HEAD, or worse, a 403 because your bot rules only allowlisted GET, the agent concludes the link is dead. It is not dead. It is just rude.

Where short links break in the wild

Here is the recurring hit list, roughly in order of how often it wrecks automated workflows:

Interstitials that need JavaScript. A landing page that flashes a logo, runs a five-second countdown, then calls window.location is invisible to a fetcher that does not execute scripts. You get a redirect that resolves to a pretty page containing no destination and no content.

Meta refresh-only redirects. Slightly better than JS, still inconsistently supported. Some parsers read the refresh tag. Many do not.

Bot walls in front of the redirect. Challenge pages from CDNs and WAF vendors are doing exactly what customers pay them for, but they cannot distinguish an agent that respects robots.txt from a scraper that does not. The result: a 403 challenge for a link a user legitimately asked an assistant to open.

Multi-hop chains. Three hops is not a routing problem, it is a parsing problem. Each hop can change headers, drop query parameters, or introduce its own rate limit. Your carefully preserved UTM string dies at hop two.

Silent expiry. A link that used to resolve now returns a 404 with a friendly HTML page. Agents do not read friendly. They read 404 and move on, sometimes logging your domain as unreliable for future use.

Opaque slugs with no surrounding signal. gowww.link/8f3k2 is fine as a string. It is useless as context. When an agent encounters that as bare anchor text, it has nothing to reason about: no title, no description, no clue whether the destination is a press release or a malware staging page.

That last one is the part people underestimate. Agents do not just follow links. They decide whether to follow them, and they make that decision from the text around the link, the domain reputation, and whatever metadata the page exposes.

Link text is machine-readable metadata

If your anchor text says "click here" or "read more," you have handed an agent a coin flip. If your anchor text is a raw short URL, you have handed it a string of characters with no semantic content whatsoever.

Descriptive anchor text is free, it helps screen readers, and it is the single highest-leverage change you can make to a document destined for an agent-heavy audience. "2026 remote work report (PDF)" beats "here" every time, and it costs you nothing but a rewrite.

Short URL slugs can carry meaning too, without giving up brevity. gowww.link/q3-earnings reads to a human and parses for a machine. Your naming conventions are doing double duty now: branding on one side, machine legibility on the other.

The hygiene rules worth adopting this quarter

These are not exotic. They are the boring, unglamorous things that separate a redirect layer that survives 2026 from one that quietly degrades.

  1. Answer HEAD requests. Always. With the same status code and headers as GET, minus the body.
  2. Use 302 for anything mutable. Reserve 301 for slugs you are genuinely done touching, and set cache headers to match.
  3. Collapse to one hop. If a link needs to move, move it, do not stack it.
  4. Never require script execution to reach a destination. The redirect must be an HTTP response, not an animation.
  5. Allowlist known agent user agents if you run a WAF, or at minimum return a plain, honest 403 with a reason instead of a challenge page designed for humans.
  6. Write link text that means something. Both in your documents and in anything you syndicate.
  7. Return 410 Gone for retired slugs, not 404. The distinction is meaningful to automated tooling and costs nothing to implement.
  8. Log agent traffic separately. User-Agent plus Accept headers plus HEAD-versus-GET ratio will tell you more about your machine audience than any analytics dashboard.
  9. Publish a machine-readable sidecar. A small JSON endpoint at a predictable path, with the slug, target, creation date, and expiry, gives agents something to cite instead of guessing.
  10. Test with curl. curl -I https://your.link/slug should tell you everything the agent sees. If you have never run it, run it today.

Humans still click, but they are no longer the majority of the packets

The uncomfortable part of this transition is that human traffic did not drop. It just stopped being the whole picture. You get both audiences on the same slug, and the design has to work for a person who wants a readable page and a program that wants an unambiguous header.

That is achievable. It just requires admitting that the redirect layer is infrastructure now, not decoration. Shorten it, brand it, track it — and then make sure it answers the machine politely.

All Articles

Related Articles

Landing Pages and Short Links: The Power Combo Marketers Keep Sleeping On

Landing Pages and Short Links: The Power Combo Marketers Keep Sleeping On

Roll Your Own: The Developer's Case for Ditching Corporate Shorteners and Going Solo

Roll Your Own: The Developer's Case for Ditching Corporate Shorteners and Going Solo

The No-Redirect Movement: Why Elite Dev Teams Are Cutting Short Links from Their Stack

The No-Redirect Movement: Why Elite Dev Teams Are Cutting Short Links from Their Stack