Skip to content

Billing for AI Work: How to Bill Clients for Work Your AI Agents Did

A practical guide to billing for AI agent work: when to disclose, how to set rates, and why tracking AI costs is the only honest way to know if your margin holds.

6 min read
A creative agency professional reviewing an invoice on a computer, with a section that itemises AI model costs alongside project tasks.
Photo by AMORIE SAM on Pexels.

Billing for AI work means being clear about how, and by what or whom, a piece of work was completed. You should tell your client when you use AI agents, decide whether to charge for this work by the hour or by the outcome, and always track your real AI costs beside your rates. This is the only honest way to know if the project made sense for you, and it is now what many clients expect. Billing for AI work is a fact of running an agency or studio in 2026.

Bring AI use out into the open

Disclosing AI use is not about appeasing hype, or ticking a box for the client’s legal team. It is about giving an honest answer to a practical question: who (or what) did what work, and how does that shape what your client was billed? You cannot brush this away now that AI is everywhere in the workflow. Clients—especially those who trust your process—do not want to discover after the fact that what looked like hand-crafted work was actually steered by a model.

The risk in hiding AI work is reputation. The reward for disclosing is credibility. There is no universal client reaction. Some will shrug and accept it. Some will challenge your price, argue for a discount, or want the option to approve AI use up-front. Others will leave for shops that promise 'all human, all local' work. None of these reactions are under your control, but disclosure is.

This matters most when the AI output is a central part of the deliverable: if the main copy, code, or design draft was done by an agent, it belongs on the bill. If you used AI for glue work—proofing, summarising, or scraping—list it, but do not overstate its share of the value. Agencies that survive the next five years will do this as a matter of habit, not compliance. This is simple to do in FlatHours: label AI outputs, assign them to tasks, and give clients exactly the record they expect. If they ask what 'claude-3-sonnet' or 'gpt-4-32k' actually did, you can show them, not just assure them.

Billing for AI work: hourly, outcome, or a hybrid?

Much of the agency world is built on the hourly rate. Billable hours are a habit: clients know the math, software does the tally, and roles within the agency align to a time-and-materials way of thinking. AI agents do not track their time, and their 'labour' cost typically comes per unit of data processed or tokens consumed, not by the minute. This causes problems, but it also opens the door to clearer outcome pricing.

Billing by outcome means committing to a result: a draft delivered, a batch of ad copy, a designed asset. This aligns AI work with value, and backs up what some clients already expect—results, not time spent. The downside is calibration. Simple AI writing can take seconds, or hours to refine. Fat margins are possible if a bespoke job is done in an instant, but you risk souring the client if they sense their budget is being captured by automation.

Hourly billing, if forced onto AI work, can generate strange results. A model can write a press release in about a minute. This is not equal to an hour of a staff copywriter. Many agencies now bill 'AI assisted' lines on the invoice—showing the task, the AI agent, and the real charge. FlatHours makes this practical: track model jobs as tasks, bill them as you wish, and let your client see what, when, and why a given item appears.

How AI work shows up in agency billing
Job Type How AI Is Used Common Billing Approach Client Expectation
Initial design draft Model generates concept visuals Outcome per asset Wants disclosure, may expect reduction
Bulk copywriting AI writes long-form text Listed as distinct line, per result Might ask for sample or explanation
Technical documentation AI structures and drafts guides Outcome or hybrid (human QA) Will expect QA and human oversight
Market research summary Model compiles and analyses Often billed hourly Expects clear savings v. old method

For work that is part-AI, part-human, use a hybrid: line item the AI’s role, then add human refinement as a distinct tracked task. Transparency is the best defence when billing for AI work. If your own workflows are mixed, FlatHours lets you show the mix as granularly as you need.

Why tracking real AI cost is the only clear margin

Billing for AI work without knowing its true cost is guesswork. AI models are not all priced alike. Rates can change overnight, and the delta between what you pay (model cost, platform fees) and what you booked as revenue will define your agency’s true profitability on these jobs. Unlike salary for a junior designer—fixed, predictable, arms-length—AI model costs depend on use, prompts, model choice, and even the provider’s daily price moves.

This is where most agency stacks let you down. If you meter AI work as admin, you have no idea what margin you made per project. If you guess at the model price, you risk underbilling or overpromising on cost. FlatHours’ AI model integration solves this the direct way: every job that goes out through FlatHours is tagged with the model, tokens, and real cost from a daily price feed. You see per-project, per-client, and cross-team reporting on where you spent, and whether your rate was justified. See FlatHours’ AI agent billing guide for demos and setup details.

Below is how tracking cost and rate can show the difference between a job that paid, and one that failed:

Billing for AI work: Billed amount versus actual AI cost
Project AI Model Billed to Client Model Cost (auto-tracked) Margin (%)
Campaign content set gpt-4-32k $1,200 $45 96.25
Longform technical docs claude-3-sonnet $700 $22 96.86
Bulk product reviews mixtral-8x7b $330 $11 96.67

Your actual model cost will change based on provider and prompt length. FlatHours always reflects the price at use. No guessing. This makes it possible to understand whether high-volume AI jobs are worth the slot in your schedule.

Honest limitations of billing for AI work: what to tell your client

Bringing billing for AI work into the open will cost you with some clients. There is a real risk that by showing how efficient (and affordable) your AI output was, a client will expect new rates—sometimes demand them. You will face negotiations about where the value is. Be ready to defend your margin in the context of your contribution: curation, editing, steering, QA, and project delivery.

There is also the risk of being undercut by agencies who still hide their AI use, or 'blend' it into admin and overhead. For those shops, ethics are downstream of close rate. If you refuse that path, you must work twice as hard to show that a clear bill is a sign of a trustworthy vendor.

No system yet solves the client trust issue in one shot. Your own experience will shape how you set expectations, bill for AI, and defend your price. FlatHours makes it easier, but does not automate trust. Only clear and honest work can do that. For more on the business and ethics of agency AI, see our dedicated billing for AI work guide, or start with our take on why traditional billing platforms are losing ground.

If you build a margin on AI, show it. If you risk a client’s trust by hiding automation, rethink. If you cannot defend your markup when the model cost is $12 and the line item is $400, reconsider what you ship. Billing for AI work is not a simple update to a timesheet. It changes the whole case for value in agency services. That is now unavoidable—and clients know it.

Time tracking and invoicing with a bill that does not move

Nothing is metered on any plan, including Free. Import your Harvest history, keep unlimited projects, clients and invoices, and take your data out again whenever you like.

Questions people ask about this

How do I bill clients when AI agents contribute to their project?

You must disclose AI involvement clearly, then decide whether to bill by hour for the time spent or by outcome for the deliverables received. Always record the real model cost to see if your rate is profitable and defendable.

Will my clients care if I use AI to do some of their work?

Some will not care, some will care deeply. The only safe position is maximum transparency: state what was done, why you used AI, and how this changed the project’s value. It may cost you a job, but it preserves long-term trust.

Is it fair to charge clients more than the AI model cost?

Yes, if you add value through planning, prompt engineering, client management, and review. The model cost is only part of the full service. However, you must be able to show what makes your service worth the margin.