AI credits reset each month.

When a credit is used

A live AI reply on a rule uses the monthly balance. The product shows used, remaining, and the reset time.

Turn AI on only for rules that need a written answer. A fixed keyword DM does not need a credit.

Budget for model work, not follower count

Credits measure AI operations in the implemented product, so they are not simply a count of contacts or delivered conversations. A generated live reply reserves a credit. A topic match can also reserve a credit separately, and preview drafts reserve one even when provider processing later fails. This differs from a competitor promise that only successful replies cost credits. Growth includes 1,000 monthly credits and Business includes 2,500; both the balance and the reset time matter when planning a campaign.

A workflow to try

Before a launch, list which rules genuinely need generated text. Keep a fixed RESOURCE keyword for the same file and use AI only where questions vary. Budget preview tests along with live questions, especially if a topic-based matcher precedes generation. Review used, remaining, and resetAt in the product instead of estimating from follower size. A few complicated active rules can use credits differently from a much larger audience that requests one static link.

Checks before you change the setup

Read fallback behavior carefully. When the monthly reserve is exhausted, the live reply service returns the saved static template; a daily AI cap uses fallback text. Noise also returns static text, so skipping AI does not necessarily mean silence. A provider exception in the live path attempts to release its claim, but a preview failure keeps its reserved credit because provider work may have happened. Do not promise zero-cost tests, a universal one-credit conversation, rollover, or an automatic paid top-up.

Separate previews, topic checks, and generated answers

Keep a small usage diary during your first AI test: configuration tested, sample type, whether a topic matcher was involved, preview outcome, and displayed balance afterward. This helps explain why answered-message counts and credits do not always match. If a provider preview fails, do not rerun it indefinitely under the assumption that failures are free. Save the error, review the note and request, and check readiness before another test. For the live path, keep the static response acceptable at monthly exhaustion. Credit budgeting should protect useful customer answers, not encourage generating a reply for an input that a fixed message already handles well.