What to say when a client asks for an AI discount
The question is fair and it is coming. Answer it with a record of what the work involved, not with a reflexive discount or a defensive denial.
Sooner or later a client says it out loud. You are using AI now, so this should cost less.
Two answers come naturally and both are bad. The reflexive discount concedes that your work was worth less than you charged, and you will never win that back. The defensive denial, where you insist nothing has changed, is not credible to anyone who has used the same tools themselves.
Before the conversation happens, it is worth knowing what number you are actually defending. The AI discount calculator estimates your leverage ratio and what the compressed capacity is worth, which is the figure the client is reaching for without being able to name it.
The useful answer starts by working out what the client actually means, because the sentence hides three different complaints.
The question is fair
It helps to concede this before anything else. If a task that took eight hours now takes three, and the contract bills hours, the client is right that their invoice should reflect it. Pretending otherwise damages trust more than the money is worth.
What is not fair is the assumption underneath: that faster means less valuable. The output is the same deliverable, carrying the same responsibility, produced by people whose judgment is what made the tool useful. Speed changed. Worth did not.
Pricing comparison
Six ways to price the same AI-assisted work
| Model | Price anchor | Tradeoff |
|---|---|---|
| Hourly | Time spent | Efficiency can shrink the invoice |
| Fixed fee | Agreed scope | Delivery risk sits with the firm |
| Value-based | Business value | Requires a credible value case |
| AI multiplier | Visible savings | Can turn AI use into a discount |
| Compute pass-through | Usage cost | Cost is not the same as value |
| Hybrid | Agreed mix | More flexible, but harder to explain |
Three things the client might mean
| What they say | What they mean | The honest response |
|---|---|---|
| You used AI, so charge less | I am paying for hours nobody worked | Show the hours actually spent, and bill them |
| You are more efficient now | I want a share of the gain | Negotiate it explicitly, once, across the relationship |
| AI work is not worth as much | The output is commodity now | Separate the deliverable from the method, and hold the line |
The first is a billing question and you should simply answer it with the record. The second is a commercial negotiation and deserves to be treated as one. The third is a positioning problem, and conceding it quietly is how firms talk themselves into a lower rate.
What to concede
Concede the hours. If you bill for time and the time went down, the invoice goes down. That is what an hourly contract says, and a client who catches you billing hours nobody worked will not be a client for long.
Concede the visibility too. A client asking this question is usually asking because they cannot see how the work was done. A record showing human hours, where AI was materially involved, who reviewed the output, and what it cost to run will answer the question more convincingly than any argument about value.
What to hold
Hold the review. Someone senior read the output, corrected it, and is accountable for it. That work is not optional and it is not cheaper because a machine produced the first draft. In many cases it is harder, because reviewing plausible-looking work takes more attention than reviewing obviously rough work.
Hold the responsibility. When the deliverable is wrong, the client calls you, not the model provider. That exposure is priced into your rate and nothing about AI reduces it.
Hold the judgment. Choosing the approach, knowing which output is wrong, and knowing what the client actually needed rather than what they asked for is the part that took years to learn. It is also the part that made the tool produce something useful instead of something generic.
Answer with a record, not an argument
Every position above is only credible with evidence behind it. A firm that can show a delivery record where human hours, material AI involvement, review status, and run cost sit together is having a different conversation from one that is asserting its value in the abstract.
The record also protects you in the other direction. When a project took more human hours because the AI-generated first pass was wrong twice, that is in the record too, and it is a far better answer to a scope conversation than a recollection.
A worked example
A content workstream used to take twenty hours: twelve drafting, five editing, three in review. With an AI-assisted first draft it now takes eleven: three drafting, five editing, three in review. Model spend is about thirty dollars. At a rate of 150 an hour, the client's invoice goes from 3,000 to 1,650.
Look at where the hours went. Drafting fell by nine. Editing did not move, because a plausible first draft needs the same attention as a rough one, sometimes more. Review did not move either, because responsibility did not move.
That is the whole argument in one table. The client already has the saving, and it landed in the part of the work where speed was possible. The remaining eleven hours are the part that was never about speed.
| Stage | Before | After | What changed |
|---|---|---|---|
| Drafting | 12 hours | 3 hours | The tool produces a first pass |
| Editing | 5 hours | 5 hours | Plausible drafts still need judgment |
| Review | 3 hours | 3 hours | Accountability did not move |
| Model spend | 0 | 30 | A new, visible cost |
| Invoice | 3,000 | 1,650 | The client already has the gain |
A client who sees this rarely asks for a further discount, because the number they were worried about is already reflected. A client who cannot see it will keep asking, and no amount of explanation will settle it.
Repricing without cutting the rate
If the relationship genuinely needs to reflect a shared gain, there are better instruments than a discount on the rate.
- Deliver more within the same retainer. The client gets visible additional value, and your rate stays intact for every future engagement.
- Shorten the timeline and price the speed. A deliverable in two weeks rather than six is worth something specific to the client's own plans.
- Move that workstream to a fixed fee. Where scope is predictable, you take the delivery risk and keep the efficiency, and the client gets a number they can plan around.
- Pass through run cost transparently. If model spend is material, showing it as a line item makes the rest of the invoice read as effort rather than markup.
Each of these keeps the rate as the anchor. A cut rate resets the anchor for every renewal and every referral that client makes on your behalf.
Update the contract before the next one asks
The conversation goes better when the agreement already anticipates it. Three clauses are worth adding at renewal: what the client is buying (hours, deliverables, or outcomes), whether AI-assisted work is permitted and under what review, and what evidence you will provide about how the work was produced.
That last clause turns an awkward question into a scheduled report. The client stops asking whether they are being overcharged, because they can already see what they are paying for.