How to Handle a Client Who Wants Unlimited Revisions

A client who wants unlimited revisions almost always wants certainty, not infinite iteration, and the fix is a scope and revision structure that gives them certainty without giving away your time for free. AI has made this worse in a specific way: since a revision now costs you a prompt instead of hours of manual rework, clients reasonably expect faster turnaround, and some read that speed as evidence the work should be free to keep adjusting indefinitely.

Why AI changed the negotiation, specifically

Before AI tools, a revision request implied real re-work time, which made a revision limit self-enforcing, nobody asked for a tenth pass because the cost of asking was obvious. AI collapsed that cost for you, but not for the underlying decision-making. A client can now generate ten variations of feedback in an afternoon, each one reasonable in isolation, and the total remains the same problem it always was: someone eventually has to decide the work is done.

Fix it in scope, before the first revision arrives

A revision limit only works if it is specific enough that both sides know when a cycle has been used, not a vague cap that invites debate about what counts.

The distinction between minor tweaks and structural changes in the second row matters most in practice, since it is the source of nearly every dispute about whether something "counts" as a revision. Define it in the contract, not in the moment a disagreement happens.

When the request arrives anyway

For an existing client asking for an unlimited fourth or fifth round with no scope language in place, the conversation goes better framed around what happens next, not what already happened.

"I want to get this right for you. We've done three rounds so far. I can do a fourth round at [rate], or if it would help, let's get on a call so I understand exactly what's not landing, since a focused conversation is usually faster than another round of written feedback back and forth." The call offer is not just politeness. Written revision requests compound because each one only addresses what the client noticed most recently, while a conversation surfaces the actual underlying concern in one pass.

What AI actually changes about how you respond

Use AI for the unglamorous part of managing this well: keeping a clear record. Ask it to summarize each round of feedback into a short changelog (what was requested, what changed, what round this was), which does two things. It gives the client a clean record that makes rounds feel concrete rather than an abstract ongoing service, and it gives you a paper trail if a dispute about round count ever comes up.

Summarize this revision feedback into a changelog entry: round
number, date, what was requested (bullet points, their words where
possible), what was changed in response. Keep it factual, no
editorializing about whether the request was reasonable.

The harder case: a client who negotiates every boundary

Some clients will push on a clearly stated limit anyway, treating each boundary as a starting offer rather than a term. If a client repeatedly negotiates past a stated scope on every project, that is information about the relationship, not just the current one, and it is worth deciding in advance what your actual walk-away point is rather than discovering it mid-negotiation under time pressure.

FAQ

How many revision rounds should I include by default?

Two to three is standard for most freelance and agency work, enough to get real alignment without becoming open-ended. Complex or highly subjective work (branding, creative direction) sometimes warrants one more round built into the base price, since disagreement there is more common and less about execution quality.

Should I mention that AI makes revisions faster for me?

Only if asked directly, and even then, frame it around your process rather than implying speed means the work should be free. Faster execution does not mean faster judgment, and judgment is what a revision round is actually paying for.

What if the client is right and the work genuinely missed the mark?

Fix it regardless of round count, but track it separately from the revision count reserved for reasonable back-and-forth on subjective calls. A genuine miss on your part is not the same category as a client changing their mind, and treating both the same either penalizes you unfairly or trains the client to treat every request as a legitimate miss.

For the disclosure question this often surfaces alongside, should I tell clients I use AI covers a related decision. For preventing scope creep before it starts, how to scope a productized AI service covers the broader version of this problem. On proving the delivered work holds up, how to prove AI-generated work is correct to a client is a useful companion once revisions are settled.