AI credits are the units a product uses to meter AI work.

A credit is a billing unit chosen by the product.

The word “credit” sounds standardized, but it is not.

A product can use it to package model calls, browser actions, research tools, image creation, or several inputs at once. Another product may use the same word for a different mix. Read the product’s pricing and usage rules before turning a credit count into a cost estimate, and use a representative task to test the real unit of value.

Tokens, credits, and outcomes answer different questions.

Tokens describe the chunks of text a language model processes.

Credits describe how a particular product chooses to charge for the work around that model. Outcomes answer the question that matters to a team: did the run produce a sourced competitor brief, a checked data cleanup proposal, or a reliable first draft? Could someone actually use it?

A low-credit run that needs two hours of repair is not necessarily efficient. A larger run that replaces a manual research pass can be the better operating choice if its sources, scope, and stopping point are clear.

The credit cost of a task depends on the work inside it.

A short rewrite and a multi-source research assignment do not ask the same system to do the same amount of work. Scope changes matter. Number of sources, length of the input, required verification, web activity, generated media, and the amount of back-and-forth can all alter the resources behind a run. The most controllable way to reduce avoidable usage is to define the audience, evidence standard, output format, and stopping rule before starting.

Repeated work should get better before it gets cheaper.

When a useful task repeats, save its proven brief, source rules, checks, and final format as a skill rather than reconstructing them from scratch. If it has a dependable cadence, a routine can prepare the same class of work on schedule. That makes the next run more consistent and easier to review, which is a more durable gain than simply trying to spend fewer credits on an unclear request.

Frequently asked questions