What Survives a Tunnel: Traffic Metadata
A tunnel hides what your traffic says, not that it exists or what it looks like. An observer watching your encrypted connection still sees when you are active, how much you send and receive, in what bursts, and to which single address. That residue is called traffic metadata, and it is the reason serious threat models treat a VPN as a partial measure rather than a solution.
Most people never need to care about this. It is worth understanding anyway, because it explains where the ceiling on a VPN’s protection comes from.
What remains visible on the wire
Even with everything wrapped in a tunnel, an observer positioned between you and your VPN server can see:
That you are connected, and to which endpoint. The address is not hidden; it cannot be, because routers need it.
When you are active. Session start and end times, and the pattern of activity in between.
How much data moves, in each direction. Volume is not concealed by encryption, and padding it costs bandwidth that consumer services do not spend.
The rhythm. Bursts, gaps, and packet sizes over time. Different activities have different shapes: a video stream looks nothing like reading text, which looks nothing like a file upload or a voice call.
Protocol characteristics. Encrypted traffic still has structure, and the handshake at the start is often recognisable enough to identify which protocol is in use.
What is not visible is the substance: destinations beyond the tunnel, hostnames, page contents, credentials. The tunnel does its job. It just does not do a job it was never designed for.
What can be inferred from shape alone
More than intuition suggests, and this is a well-established area of study rather than speculation.
Activity class. Whether you are streaming video, on a call, downloading something large, or idly browsing is often apparent from volume and rhythm without any decryption.
Timing patterns as identity. When you are online, for how long, and how regularly is a behavioural signature. It does not name you; it does distinguish you.
Correlation across observation points. This is the significant one. An observer who sees traffic entering your tunnel and traffic leaving the VPN server can match the two by comparing timing and volume. No cryptographic weakness is required. It is why a commercial VPN is not enough against a state adversary — an adversary with visibility at multiple points in the network can defeat a single hop analytically.
Website fingerprinting, in constrained settings. Research has shown that the pattern of a page load can sometimes be matched against known patterns. Real-world reliability depends heavily on conditions, and it is a research finding rather than something to assume about your own traffic — but the direction of travel is worth knowing.
We are deliberately not putting numbers on any of this. Published results depend entirely on their experimental setup, and quoting a percentage as though it applied to your situation would be exactly the kind of false precision this site avoids.
What your VPN provider sees, in this frame
The provider is in the strongest position of anyone, because for it the traffic is not just shaped — it is decrypted at the tunnel boundary and forwarded. It sees destinations directly, alongside your real address and all the timing information above.
This is structural rather than a flaw, for the reasons in what data a VPN must hold, and it is the core of the trust relocation described in what your VPN provider can see. Metadata analysis is what an outside observer must resort to. The provider does not need it.
Why nobody fixes it
Concealing traffic shape is possible in principle and expensive in practice.
Padding — adding filler so all packets look the same size — wastes bandwidth continuously, for every user, whether or not they are doing anything.
Constant-rate transmission — sending at a fixed rate regardless of demand — conceals rhythm completely and is deeply impractical for general internet use on metered or mobile connections.
Cover traffic — generating decoy activity — multiplies bandwidth costs.
Bandwidth is a real cost that scales with users, so a consumer service competing on throughput has no commercial route to spending it on padding. Systems that do make these trade-offs exist, are built for anonymity rather than convenience, and accept the performance penalty deliberately. That is a different category of tool, not a better VPN.
Does this matter to you?
Probably not, if your adversary is a café network or your ISP. Neither is running correlation analysis on your browsing, and neither can see the far side of your tunnel. Metadata visibility does not undermine the ordinary use case described in what a VPN changes on public Wi-Fi.
It matters if your adversary has broad network visibility, because that is exactly the capability metadata analysis requires. If that describes your situation, the conclusion is not “buy a better VPN” — it is to get advice from organisations specialising in digital safety for people at risk.
It matters for calibrating claims. A provider promising invisibility is promising something the physics of the network does not permit for a consumer-grade service. Knowing what remains visible is what lets you discount that kind of claim without needing anyone’s help.
The takeaway
Encryption is a strong tool with a precise scope: it protects content and integrity. Everything about the envelope — existence, endpoints, timing, size — travels in the open, because the network needs it to deliver anything at all.
A VPN moves the envelope’s visible endpoint from your destination to your provider, which is genuinely useful. It does not and cannot make the envelope disappear. That boundary is where the honest description of these products ends, and it is worth carrying into every marketing page you read next — as is the general scope in what a VPN actually protects against.