HighLevel Adds Fixed-Rate AI Billing for Agencies, and Quietly Moves Token Risk to You

August 10, 2026. Also on August 7, HighLevel added the option to bill sub-accounts a fixed rate for Conversation AI and Voice AI instead of only a markup on usage. Its example is 50 cents for every message sent through Conversation AI, the same amount every time, whatever that message actually cost to process. You pick markup or fixed rate per product and per sub-account, and you can switch whenever you like.

Why this is not free money

The reason this is worth ten minutes of thought is one sentence HighLevel puts in its own notes: you are still billed based on actual usage on your end. A fixed rate does not remove token variance from the transaction. It moves it onto your balance sheet.

The two units problem

1. HighLevel's Conversation AI pricing article states the usage-based rate as 2 cents per message sent to the contact or suggestion shown to the sub-account user. Two different events, one price.
2. The changelog's fixed-rate example is priced per message sent. If the billable event you charge on is narrower than the event you get charged on, a sub-account whose staff browse many AI suggestions and send few generates cost with no matching revenue.
3. HighLevel has not published how the fixed rate treats a suggestion that is shown but never sent. That is not a gotcha, it is a gap, and it is answerable in one email to support. Ask it in writing before you set a rate.
4. Rebilling itself is plan-gated. HighLevel's pricing article lists rebilling as available on the 497 dollar plan and on the 297 dollar plan at lower markup. Bot training and bot trial are free of cost, so testing does not burn budget.
5. There is a flat option on the cost side too. The same article lists an unlimited monthly plan at 49 dollars per sub-account for unlimited Conversation AI messages.

What it means for operators

Fixed-rate pricing is genuinely easier to sell, and HighLevel is right about why. Tokens are impossible for a client to estimate, and a line item that moves every month invites a monthly argument. One predictable number per message closes deals faster.

But the risk profile flips entirely depending on which cost model you are on. If your own cost is usage-based, a fixed rate makes you the insurer: you keep the margin in a quiet month and eat the overage in a heavy one. If you move that sub-account to the unlimited plan first, your input cost is flat, and a fixed rate charged on top becomes close to pure margin with the variance already capped by the vendor. That ordering is the whole decision, and it takes one settings change.

So before you set a number, pull 30 days of real usage per sub-account and price off your heaviest account rather than the average. Averages are what make fixed-rate pricing look free. Then reprice on a schedule rather than never, because HighLevel lets you switch models at any time and a rate set against last quarter's usage is not a rate, it is a guess. Agencies packaging this into a productised offer should also decide whether the fixed rate is a profit centre or a door opener, since the same 50 cents can be either. We build this kind of margin model into GoHighLevel implementations as a matter of course, and it is the same arithmetic behind any AI service you resell: know your unit cost, know your worst month, then price.

One limit worth printing. The prices above are what HighLevel published on its own pricing and changelog pages as of today, and vendor pricing changes. Check your own agency billing screen before you commit a number to a client contract.