A hashed ID is still an ID

What an "anonymous" analytics identifier still links together

Checked against the linked sources on October 26, 2026

The short answer

Most privacy-focused analytics tools avoid names, emails, and advertising IDs, and give each installation or visitor an identifier with nothing personal in it. That is a real improvement on ad-tech tracking. But an identifier with no name attached is still an identifier. Every event that carries it belongs to one installation, so the store holds a history per installation, and any outside detail that connects that installation to a person connects the whole history.

Hashing hides the original value. It does not break the link, because the same input always produces the same hash. What holds up against a breach, a court order, or a dataset someone joins later is not storing an identifier at all. Trackless is built that way. Its design has limits too, and they are listed below, next to the cases where an identifier-based tool is the right choice.

What a stable identifier lets someone join

An identifier is a join key. It needs no name to be useful: it only has to stay the same from one event to the next.

  • It turns events into a history. Every event with the same ID is the same installation: when it was used, what it did, on which device and from which city, for as long as the vendor keeps the data.
  • A few outside details narrow it to a person. A sign-in time in your own server logs, a support ticket that mentions the time of an error, or knowing who in a small town uses a particular tablet can match one ID. After that one match, the ID hands over everything recorded under it.
  • Some identifiers are shared by design. On iOS, the identifierForVendor is the same for every app from the same developer on the same device (Apple: identifierForVendor), so a store keyed on it can link a person's use of several apps. Android's Advertising ID is a device-level ID for advertising, provided by Google Play services, that lasts until the user resets or deletes it (Google Play: Advertising ID). Android's own guidance notes that the wider an identifier's scope, the greater the risk of it being used for tracking (Android: best practices for unique identifiers).
  • Linking can run backwards. Tools that merge identities attach an account's user ID to the events a device sent before anyone signed in. Mixpanel, for example, retroactively sets the user ID on prior events that carry the same device ID (Mixpanel: identifying users). Once one event is tied to a person, the anonymous ones before it are too.

Why hashing does not change that

A hash function turns a value into a fixed-looking string, and the same value always gives the same string. That is what makes a hashed ID useful for analytics, and it is also why a hashed ID is still a persistent identifier. The US Federal Trade Commission put it plainly in 2024: "hashes aren't 'anonymous' and can still be used to identify users" (FTC: No, hashing still doesn't make your data anonymous).

  • Guessable inputs can be matched. If the hashed value is an email address, a phone number, or a device ID, anyone holding the salt can hash a list of candidates and compare. The FTC's post describes a case in which, the FTC alleged, hashed email addresses sent to an ad platform were matched back to the people seeking counseling.
  • Salting slows matching down; it does not remove the link. Two events with the same salted hash still came from the same installation.
  • Rotating salts help, at a cost. Some tools derive a visitor ID each day from the IP address and browser details, with a salt they throw away every 24 hours. That limits linking to one day, which is a meaningful improvement. It also means the server reads the IP address of every request to compute the ID.

European law draws the same line. Under the GDPR, data that has been pseudonymised but could be attributed to a person using additional information is information about an identifiable person, judged by the means reasonably likely to be used, singling out among them (GDPR, Recital 26). A breach, a court order, or a second dataset is exactly how someone acquires that additional information.

What a breach or a court order can produce

A vendor can only lose, or be made to hand over, what it holds. The design decides what that is.

A store keyed on a per-install IDTrackless
The database is copied Every event per installation, with times and device details, for as long as the vendor keeps it Daily counts by event name and coarse context; no events, no identifiers
A court order or other legal demandThe same histories, per installation or per accountThe same daily counts; there is no per-person record to produce
Joined with another datasetOne matching detail can attach a person to a whole history Nothing to join on. Someone who already knows about a person can sometimes narrow a small count to them (see below)
The live service is taken overEverything already stored, plus whatever arrives nextOnly what arrives from that point on; the stored history still holds no one
How long the exposure lastsAs long as the vendor keeps events, often years13 months of aggregate counts, then automatic deletion

What Trackless stores instead

The SDKs send no identifier of any kind. A session lives in memory on the device and ends when the app goes to the background; the next launch is a new, unrelated session. Ingest adds each event to a daily count for its name and coarse context (platform, OS version, device class, app version, region from the device's locale, language) and writes nothing per event. Application code never reads the IP address, and the ingest endpoint keeps no access log of network addresses.

A stored row looks like this:

DateFeaturePlatformOSDevice classRegionCount
2026-10-25exportiOS18phoneUS1,284

Two sessions on the same phone are indistinguishable from two sessions on two phones. On the App Store label you declare Product Interaction, Other Diagnostic Data, and Performance Data (and Crash Data, if you turn on uncaught-exception counts), none linked to the user, and nothing under Identifiers; the privacy guide walks through it.

Where Trackless's protection runs out

Storing no identifier removes the join key. It does not make every count meaningless, and these are the cases where a count can still say something about a person:

  1. Someone who already knows about a person. If you know someone used a feature on Tuesday, from Iceland, on an iPad, the day's count for that combination can confirm it. This confirms what someone already suspects; it does not reveal anyone new.
  2. Features or values only a few people trigger. An event that only one person could cause, such as an admin screen in an app with one admin, describes that person.
  3. Very small apps. A rare combination of platform, OS version, device class, region, and language can narrow to a few people. The dashboard shows small counts as they are, down to a count of 1. Performance durations are stored as digests rather than individual records, but a digest of very few samples approximates those samples: from a single sample, the percentiles and the duration histogram give that duration to within about 5%.
  4. Values the developer sends. The SDKs strip email addresses, phone numbers, and SSN patterns, and ingest rejects values that look like encoded IDs, such as UUIDs and long hex strings. A plain name or username passed as a value would still get through. The SDK guides tell developers never to send anything about the person.
  5. A takeover of the live service. Someone who gained control of the ingest service, or compelled a change to it, could record what arrives from that point on, including the network address of each request. That applies to every analytics service; what Trackless changes is that there is no history to take.

If your app serves people who could be harmed if found, keep event names to actions many people take, never instrument something only they would do, and limit who can see the dashboard. The privacy whitepaper describes the architecture and these limits in more detail.

When an identifier-based tool is the right choice

Identifiers are what make some questions answerable at all. If you need any of these, Trackless cannot provide them, and a tool that keeps an identifier is the right choice:

  • How many people used the app this week: unique users, DAU, or MAU
  • Whether people come back: retention curves, new versus returning
  • Following one installation or account through a sequence of steps
  • Marketing attribution, revenue per user, or session replay

If you need to know which features are used, what is failing, and how slow it is, without holding anything per person, Trackless is built for that. Side-by-side comparisons, each checked against the vendor's own documentation: TelemetryDeck, Firebase, PostHog, Aptabase, Mixpanel, and Plausible.

Sources

Every outside fact here comes from the pages linked above, read on October 26, 2026. If something on this page is out of date or wrong, tell us at feedback@tracklesstelemetry.com and we will correct it. This page is informational and not legal advice.