Training capacity is the amount of source text your chatbot can store. It is separate from the message credits used for conversations and other credited operations.
Where to check your limits
Create your chatbot, select it, and open Training. The Limits card in the left sidebar shows Training size, Links, and Re-trainings. Read your actual used/allowed figures here; there is no single fixed allowance for every account.
Plans, add-ons and account-specific settings can change the available capacity. Open Subscription to review the options offered for your account.
What counts as a training character
The stored content of each source contributes to usage. Website extraction removes HTML markup, scripts and styles, but retained link URLs, Markdown syntax and other stored text can still count. Uploaded file bytes are a different measurement: a large PDF may extract little text, while a small text file may contribute most of its characters.
There is no reliable fixed percentage that converts website megabytes into training characters. Use the source's character count and View to inspect what was extracted. See Training basics and limits.
Per-chatbot capacity
Each chatbot has its own training character allowance. Sources within that chatbot share it, including website pages, uploaded documents, Text, Q&A and corrections. A bot reaching its limit does not use up another bot's allowance.
For example, 800,000 characters from website sources plus 300,000 from documents and 200,000 from Q&A total 1,300,000 characters. Compare that total with the selected chatbot's displayed limit. Connected catalog and AI Search features can have their own catalog limits; do not infer those from this text-source example.
Storage does not reset monthly
Training usage measures the amount currently stored, not a monthly training budget. Adding sources increases usage, and removing obsolete sources frees capacity after the removal is processed. Monthly message-credit and automatic-retraining allowances are separate.
Re-training replaces existing source content. When an existing 5,000,000-character source grows to 5,200,000 characters, its contribution increases by 200,000; the resulting stored source still occupies 5,200,000 characters. This is not a guarantee that uploading another copy creates a replacement.
If growth would exceed capacity, a retrain can retain the old content instead of updating it. Scheduled retraining can also be skipped when capacity is effectively full. Check the source's last update and View after a refresh.
If training reaches a limit
- Open Training and check which allowance was reached: characters, links or automatic retrains.
- Review large or obsolete sources with View. Delete only content you no longer need, or narrow a future scan with URL and element filters.
- Review the available options in Subscription if you need more capacity.
- Retry the affected training and wait for Training completed. Processing time depends on the source; you can leave the page after the source is accepted.
- Check the updated usage, then open Overview and test a question that needs the newly trained information.
Reduce usage without losing answers
Keep product details, policies, FAQs and support content that answer your visitors' questions. Exclude irrelevant pages or repeated layout text, but check that filters do not remove needed information. A sitemap selects candidate URLs; it does not remove headers and footers from those pages.
Follow How to reduce training characters for the filtering controls, and use the Knowledge Base Optimizer to check retrieval after changes.