What does an AI subscription stack actually cost?

Why this category accumulates faster than others

Three reasons, and they compound.

Everything launches with a free tier and converts. Trials in this space are short and the conversion is quiet, which is the exact combination that produces forgotten subscriptions.

Tools are adopted individually for individual tasks. A writing assistant here, an image tool there, a transcription service, a coding assistant. Each was justified at the time.

The overlap grows without anyone noticing. General assistants keep absorbing what specialised tools used to do. Three subscriptions bought a year ago for three distinct jobs may now be one subscription's worth of capability.

That third point is the one that makes an annual review insufficient.

Auditing the stack

Fifteen minutes, quarterly.

1. List everything, including free tiers you have consented to charges on. Trials that converted are the largest single source of surprise here.

2. Note which are billed as subscriptions and which are usage-based. This is the category's specific trap: API access is billed separately from a consumer plan, and it is entirely possible to be paying both for the same provider without realising. Check the billing page, not the plan page.

3. For each, name the task it does that nothing else in the list does. If you cannot, it is a consolidation candidate.

4. Check what your employer provides. Coverage is increasingly common and often includes the tier you bought personally.

5. Convert everything to an annual figure. This category's monthly prices look small individually and add up faster than people expect.

What consolidates and what does not

Usually consolidates: general writing assistance, summarisation, everyday questions, basic image generation. A single general subscription now covers what several tools did.

Usually does not: specialised transcription with speaker separation, domain-specific tools with proprietary data, anything integrated deeply into a workflow you did not build.

Depends: coding assistance. Genuinely contested and it shifts between releases, which is an argument for testing rather than reading.

The test is the same as everywhere: cancel the candidate, work for a month, and see whether you missed it. Resubscribing costs nothing but the month.

The two things to set up once

A price baseline per tool. Record what you agreed to pay. This category restructures tiers constantly — the plan you are on stops being offered and you are moved to a nearer-priced one — and without a stored original that is invisible against a charge that looks normal.

A quarterly reminder to run the audit. Annual is too slow here. Everything else in a subscription list can wait six months; this cannot.

Die Entscheidung einmal treffen und dann festhalten

Der teurere Fehler ist nicht der falsche Dienst, sondern beide zu bezahlen — oder einen zu behalten, den seit März niemand mehr geöffnet hat.

Zwei Dinge verhindern das. Den Betrag notieren, den du zugestimmt hast, damit eine spätere Erhöhung überhaupt auffällt; ohne gespeicherten Ausgangswert gibt es nichts zu vergleichen, und Dienste wandern über Jahre unbemerkt nach oben. Und eine Erinnerung vor der Verlängerung mit der richtigen Vorlaufzeit: sieben Tage bei monatlichen Plänen, dreissig oder mehr bei jährlichen, weil Kündigungsfristen von dreissig Tagen üblich sind und eine Warnung innerhalb dieser Frist zu spät kommt.

SubTracker macht beides manuell — der kostenlose Plan verfolgt unbegrenzt viele Abos mit Kalender-Feed und Export; Erinnerungen und Preis-Alarme sind Plus für $5.90/Monat. Ein Kalendereintrag pro Verlängerung tut es auch, kostet nichts und ist zwanzig Einträge Pflegeaufwand.