How to price AI: limits, overage, cost visibility
Make likely monthly cost, included limits, overage rules, and exit path visible before purchase, so AI pricing earns trust instead of hiding cost.
| By | Prompt & Product — Newsroom |
|---|---|
| Filed | 8 September 2026 |
| Read | 3 MIN |
Before a user buys an AI feature, they should be able to say the likely monthly cost out loud. Large language model inference prices have declined at a median rate of 50x per year, with the fastest declines after January 2024 accelerating to 200x per year. The pricing page still has to make the bill predictable, because the user is deciding on a workflow, not a token count.
Put the whole cost on the purchase screen
Design the purchase screen as a cost forecast. It should answer what is included, what happens after, and what a normal month feels like, in words the user can repeat.
- Name the model: subscription, usage, credits, seats, or hybrid.
- Show included limits as usable amounts, not raw tokens.
- Define overage: trigger, unit, rate, and cap.
- Estimate likely monthly cost from a realistic workload.
- Expose the exit path: cancellation, data, and credit rollover.
The failure is a usage model, or the way price changes with use, that appears only at billing. When overage arrives with the bill, the design has hidden the cost.
Choose subscription when the workload is steady, usage when it is variable, and hybrid when a base plan plus overage fits the workflow. The plan should be a label the user can repeat: unlimited reviews, included credits, or per seat.
Write the plan in the user's language on the purchase screen: for a subscription, what is included and what is not; for usage, what counts as use and what does not; for a hybrid, what the base covers and where overage starts.
Translate the allowance into the user's task. The limit passes when it covers a normal week, not a raw token count. For usage-based plans, state the trigger, unit, rate, and cap in one short sentence, and say what happens when the cap is reached, resets, or pauses the feature.
Put a range near the purchase button, not a single optimistic number. For a chat assistant, show a sample month with a normal number of conversations. For a code agent, show a sample month with a normal number of tasks.
Expose the exit path as part of the offer. It passes when it covers cancellation, data, and credit rollover in plain language.
Receipts show the bar for visible usage
Some AI pricing models lead with a seat price. CodeRabbit, an AI code review tool, sets seat prices between $12 and $24 per month for unlimited pull request reviews.
Other tools make usage terms explicit. Emergent, an AI app builder, sells a Standard plan for $20 per month that includes 100 credits and a Pro plan for $200 per month that includes 750 credits.
Enterprise investment in generative AI was projected to grow by 50% in the coming year, while only 6% of surveyed companies reported achieving 75% of expected return on investment.
The receipts matter because the user is comparing options while reading a price. The cost pressure makes fixed allowances easier to justify, but it also makes stale pricing pages more obvious. A predictable monthly cost is easier to compare, budget, and cancel.
Redesign the price page as a forecast
Audit the price page for hidden usage terms. Search for words like credits, tokens, compute, usage, and activity. Where a reader cannot map a word to a task, replace it with a task-based label or add a plain-language explanation.
Make the estimate live. Update it when usage patterns change. Display the actual cost from the most recent month next to the forecast. Let the user adjust the workload numbers and watch the price move.
Replace the current price page with included tasks, overage rate, and likely monthly cost. Where credits are used, list what a credit buys and what a normal month uses. The redesign should make the cost visible where the buy button is.
