Credits and metering

RankX AI meters work with credits, from one wallet per account. Every action that calls an outside provider follows the same three-step lifecycle: credits are reserved before the work starts, consumed if it succeeds, and released in full if it fails. You are never charged for work that did not produce a result.

There are no per-feature quotas. There is one balance, and what you spend it on is your decision.

The lifecycle

graph LR
  A[Action requested] --> B{Enough credits?}
  B -->|No| R[Refused, with the reason]
  B -->|Yes| C[Reserve]
  C --> D[Provider call]
  D -->|Success| E[Consume the reservation]
  D -->|Failure| F[Release in full]

Two properties follow from that shape and both matter:

A failure costs nothing. If an assistant times out, a crawl cannot reach your site or an article generation fails, the reservation is released and your balance returns to where it was. Refunded work does not appear as spend in your usage report either, because it was never spend.

A partial run is honest about what it did. If a scheduled run is interrupted or the wallet runs dry halfway, RankX AI charges for the checks it completed, marks the run partial, and reports the shortfall. It does not silently return a half-finished result as a whole one.

What is charged, and per what

The unit is where people are surprised, far more often than the price. Two examples do most of the work:

An AI visibility check is charged per prompt, per assistant, per run. Twenty five prompts across five assistants is 125 charged checks on every run, not twenty five. Multiply by your cadence to get a monthly figure.

An AI Shopping check is charged per shopping prompt, per surface, per run. The same shape over a different and smaller set. Shopping prompts are a separate allowance from AI visibility prompts, and the two do not draw on each other.

The full list of actions with their units and current prices is in the credit cost reference, which is generated from the product rather than typed by hand. Prices are deliberately not repeated anywhere else in this documentation: rates are versioned and admin-editable in RankX AI, and a number typed into prose is wrong the first time a new rate version is published.

What is free

Reading never costs credits. Every report, every dashboard, every export and every read tool on the MCP server is free to call.

So is a cache hit. Keyword discovery and trends are cached for a period, and repeating a recent request inside that window returns the stored result without a charge.

So is the onboarding package. The brand scan, the Brand Hub build and the homepage check that run during setup are platform-funded: they do not touch your balance. See the free trial.

Structural limits are not the wallet

This is the distinction that most often gets muddled, and the two things behave completely differently.

Structural limits are enforced in the database rather than in the interface, so a bulk import that exceeds one is refused rather than silently trimmed. The figures are in plans and limits.

Legacy per-feature monthly quotas no longer exist. If you have seen a claim about "20 manual checks a month" or "3 article generations a month" anywhere, it describes a model RankX AI retired.

Rates are versioned

RankX AI keeps its prices in a versioned rate catalogue. A published version is immutable: changing a price publishes a new version rather than editing the old one, so the price a past charge was made at stays knowable.

That is why prices are resolved live everywhere they are shown, including in the descriptions of the MCP tools that spend, and why this documentation points at one generated page rather than repeating figures.

When a spend is refused

There are four refusals, and they have four different remedies. RankX AI keeps them apart deliberately: telling a customer whose subscription has lapsed to buy credits sends them to do something that cannot work.

The third is agency-only and it catches people out, because the agency wallet can be full while a client is stopped. A per-client budget is a ceiling on the shared wallet, not a separate pot: no credits are moved or set aside, and topping up the agency wallet changes nothing for that client until the allocation is raised. See per-client credit budgets.

Controlling what you spend

Recurring spend in RankX AI comes almost entirely from four things, in this order:

1. How many prompts are active, multiplied by how many assistants are enabled, multiplied by how often they run. This is usually most of the bill.
2. How many keywords are tracked, multiplied by the check cadence.
3. How many shopping prompts are active, multiplied by two surfaces.
4. Everything else, which is occasional rather than recurring: audits, research runs, briefs and drafts.

So the effective levers are:

• Prune the prompt list. Pausing a prompt stops it costing credits immediately, keeps its history, and is fully reversible. This is the single most effective change available.
• Match the cadence to the plan. Every plan states a fastest cadence, and no plan's monthly grant funds the fastest cadence on a full prompt list. Running daily when weekly answers the question is the most common cause of an unexpected balance.
• Untrack keywords you do not act on. Reversible, and it frees the slot against your plan limit as well as the spend.
• Check the balance before a big run. Every surface that spends says what it will cost before it does it.

Where the credits go

The usage report groups spend by action or by Website over a period, so "why did my balance drop" has an answer rather than a theory. Work that failed and was refunded is excluded, because it was never spend.

Credits do not roll over. An unused monthly grant expires at the end of its period on every plan, which is another argument for matching cadence to plan rather than buying headroom you will not use.

Where to go next

• Credit costs, the generated per-action table.
• Plans and limits, for every structural ceiling.
• The free trial, for what the trial grant funds.
• Per-client credit budgets, for the agency ceiling.