Your subscription covers two separate things, and they are metered separately. Knowing which one you are using answers most questions about limits.
Research credits are spent on research runs.
AI usage allowance covers what the AI models cost to run while you write.
Both reset at the start of each billing cycle. You can see both on Home, and in more detail under Usage.
Research credits
One credit pays for one research. A finished research is reusable: create as much content from it as you like without spending another credit. Refreshing a research spends a new one.
Plan | Price | Research credits per cycle |
|---|---|---|
Pro | $29 / month | 5 |
Startup | $99 / month | 25 |
Business | $399 / month | 125 |
Credits do not roll over. Each cycle starts with the full allowance, whatever you used in the last one.
AI usage allowance
Writing with the agent calls AI models, and those models cost money to run. Your plan includes an allowance for that cost, shown in dollars because that is what it is.
Plan | AI usage allowance per cycle | You see a warning at |
|---|---|---|
Pro | $6.00 | $4.80 |
Startup | $20.00 | $16.00 |
Business | $80.00 | $64.00 |
The allowance is 20% of your plan price, rounded to the nearest 50 cents. It is not a separate charge. It is the share of what you already pay that is set aside for model costs.
As you approach the limit
At 80% of your allowance, you will see a warning in the app. Nothing changes yet. It is there so a long piece of writing does not stop halfway without notice.
At 100%, AI features pause until your next cycle begins. Your content and research are all still there and still editable. Only the calls to AI models stop.
You have two ways to keep going straight away:
Add your own API key on the LLM API Keys page. Calls on your own key are billed by the provider directly and never count against your allowance.
Upgrade your plan. The allowance follows your plan price, so a bigger plan raises it right away.
What never counts against your allowance
Voice transcription. Dictating notes or instructions is free.
Web searches the agent runs for you in the Agent tab. Web search in the Targeted tab runs inside the AI model itself, so it counts as normal AI usage.
Research. A research is paid for by its credit, and page audits and content briefs cost nothing extra.
Import assistance when you bring existing content in.
Anything run on your own API key.
So if you watch the meter during a research run and it barely moves, that is expected. Research spends a credit, not your AI usage allowance.
When the cycle resets
Both meters follow your billing cycle, not the calendar month. If you subscribed on the 14th, your cycle runs from the 14th to the 14th. The Usage page shows the exact dates of the current cycle.
If your billing dates cannot be read for any reason, the allowance falls back to the calendar month, so a missing date never blocks you.
On the LLM API keys page, the This month total is a calendar-month figure. The AI usage allowance card beside it follows your billing cycle, so the two can show different amounts.
Questions people ask
Do unused research credits roll over?
No. Each cycle starts fresh.
Does a cheaper AI model make my allowance last longer?
Yes. The allowance is measured in dollars of model cost, so a less expensive model uses less of it for the same work.
If I add my own API key, do I still need a plan?
Yes. The plan covers the product, research credits and support. Your own key only changes who pays the model provider.
Is the allowance a bill I might receive?
No. It is an allowance inside what you already pay. WordCrafter never charges you more than your plan price for AI usage. When you reach the allowance, AI features pause rather than costs piling up.
Do I get anything back if something fails?
For research, it depends on how far the run got: a run that fails or stalls before your outline or page plan is saved gets its credit back, while one that already saved your outline or page plan keeps the credit, because you keep it. Research FAQs lists exactly when. AI usage works differently: model calls that already ran have been paid for, so that usage is not refunded, even if you stop the agent partway.