TL;DR. The question “should I bill fewer hours because AI did some of it” has a written answer in at least one profession, and it is stricter than most independent workers assume. The American Bar Association says lawyers billing hourly must bill actual time, and that charging the same flat fee for work an AI made much faster may itself be unreasonable. That closes the escape hatch most pricing advice recommends. Meanwhile the EU AI Act’s transparency article has been in force since 2 August 2026, and the US Copyright Office has stated that prompts alone do not make you the author of the output. Three instruments, three different obligations, and together they decide what you may charge, what you must say, and what your client actually receives.
The question nobody wants to answer out loud
A deliverable used to take you ten hours. With AI assistance it takes four, and the quality is the same or better.
You now face a question with real money attached. Do you invoice for four hours, or ten? If you quoted a fixed price built on ten hours of work, do you keep it? And does the client have any right to know how the job got done?
Most of the advice circulating on this is confident and unsourced. It generally lands on the same recommendation: stop selling hours, sell outcomes, and the problem disappears.
It does not disappear. In the one profession that has written formal rules about exactly this situation, that recommendation is explicitly addressed and partly rejected. Independent workers are not bound by legal ethics rules, but the reasoning is instructive precisely because someone finally had to apply a duty of fairness to this fact pattern rather than a marketing preference.
What the written rules actually say
Three instruments now govern different parts of this question. None of them is a blog post.
On billing. The American Bar Association’s Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on 29 July 2024, addressing generative AI tools. Its fee section applies Model Rule 1.5, which requires fees and expenses to be reasonable.
The opinion states that generative AI “may provide lawyers with a faster and more efficient way to render legal services to their clients, but lawyers who bill clients an hourly rate for time spent on a matter must bill for their actual time.” It quotes an earlier opinion from 1993: “the lawyer who has agreed to bill on the basis of hours expended does not fulfill her ethical duty if she bills the client for more time than she has actually expended on the client’s behalf.” And it repeats the principle that where a lawyer “is particularly efficient in accomplishing a given result, it nonetheless will not be permissible to charge the client for more hours than were actually expended on the matter.”
It gives a worked example. If a lawyer uses an AI tool to draft a pleading and spends 15 minutes inputting the relevant information, the lawyer may charge for those 15 minutes, plus the time spent reviewing the resulting draft for accuracy and completeness. Not the hours the task used to take.
Then comes the part that closes the escape hatch. The opinion applies the same reasonableness factors to flat and contingent fees: “if using a GAI tool enables a lawyer to complete tasks much more quickly than without the tool, it may be unreasonable under Rule 1.5 for the lawyer to charge the same flat fee when using the GAI tool as when not using it.” It cites a case in which charging a flat fee of 1,000 dollars for work barely performed was a rule violation, with the principle stated as “a fee charged for which little or no work was performed is an unreasonable fee.”
So switching to fixed pricing does not automatically resolve the ethics of capturing the efficiency gain. It changes who has to justify the number.
On the cost of the tools themselves, the opinion draws a line between overhead and pass-through. A tool embedded in your word processor to check grammar is overhead and should not be billed to the client absent contrary disclosure. A third-party service charged per use for a specific client engagement may ordinarily be billed as the actual out-of-pocket expense, with no surcharge added and any discounts passed along. And under Model Rule 1.5(b), before charging the client for AI tools or services at all, you must explain the basis for the charge, preferably in writing.
On disclosure. Article 50 of the EU AI Act, Regulation (EU) 2024/1689, sets transparency obligations. It has applied since 2 August 2026 under Article 113, and the 2026 Digital Omnibus on AI, Regulation (EU) 2026/1744, did not postpone it. That regulation amended only paragraph 7 of Article 50, removing the Commission’s power to approve codes of practice by implementing act, and it added a transitional rule elsewhere: providers of generative AI systems placed on the market before 2 August 2026 have until 2 December 2026 to comply with the machine-readable marking duty in Article 50(2).
Most of Article 50 binds providers and deployers of AI systems rather than freelancers who use them. But one subparagraph is directly relevant to anyone publishing AI-assisted text, and it contains a carve-out worth reading closely. Deployers of a system that generates or manipulates text published to inform the public on matters of public interest must disclose that the text has been artificially generated or manipulated. The obligation does not apply, in the regulation’s own words, “where the AI-generated content has undergone a process of human review or editorial control and where a natural or legal person holds editorial responsibility for the publication of the content.”
Note what the law treats as sufficient. Not a label saying AI was involved. Human review, plus a named person or entity holding responsibility for the result. The standard is accountability, not confession.
On ownership. In January 2025 the US Copyright Office published Part 2 of its report on copyright and artificial intelligence, addressing copyrightability, after receiving more than 10,000 comments from all 50 states and 67 countries, roughly half of them on this question.
Its conclusion is direct: “The Office concludes that, given current generally available technology, prompts alone do not provide sufficient human control to make users of an AI system the authors of the output.” The reasoning is that “prompts essentially function as instructions that convey unprotectible ideas,” and that providing detailed directions without influence over how they are executed is insufficient for authorship.
The report is not a blanket denial. It states that “in many circumstances these outputs will be copyrightable in whole or in part, where AI is used as a tool, and where a human has been able to determine the expressive elements they contain,” with modification, selection and arrangement assessed case by case. It also confirms that including AI-generated elements in a larger human-authored work does not affect the copyrightability of the larger work.
The commercial consequence is easy to miss. If your deliverable is a pure prompt-to-output artefact, you may be selling your client something nobody owns. If your contract promises assignment of copyright, you may be promising to transfer a right that does not exist.
What each instrument actually requires
CEOtudent editorial synthesis. Each row summarises a published instrument; the obligations are those stated in the source documents.
| Instrument | In force | Who it binds | What it requires |
|---|---|---|---|
| ABA Formal Opinion 512, Model Rule 1.5 | 29 July 2024 | US lawyers; persuasive elsewhere | Hourly billing must reflect actual time worked. The same flat fee may be unreasonable if AI made the work much faster. Tool costs are overhead unless charged per use for that client, then billed at actual cost without surcharge. |
| ABA Formal Opinion 512, Model Rule 1.4 | 29 July 2024 | US lawyers; persuasive elsewhere | Must disclose AI use if the client asks, if the engagement requires it, if client information is entered into the tool, if AI use bears on the fee, or if output drives a significant decision. |
| EU AI Act Article 50(4), Regulation 2024/1689 | 2 August 2026 | Deployers publishing AI text on matters of public interest | Disclose that text is artificially generated, unless it has had human review or editorial control and a named person or entity holds editorial responsibility. |
| EU AI Act Article 50(2) with the 2026 Omnibus transitional rule | 2 August 2026, legacy systems by 2 December 2026 | Providers of generative AI systems | Outputs must be marked machine-readable and detectable as artificially generated. |
| US Copyright Office, Copyright and AI Part 2 | January 2025 | US copyright practice | Prompts alone do not confer authorship. Protection attaches where a human determines the expressive elements. |
The arithmetic the advice skips
Now put numbers on it, because the pricing debate is usually conducted without any.
Take a deliverable that required 10 hours of work at 120 dollars an hour, invoiced at 1,200 dollars. AI assistance brings it to 4 hours, a 60 percent reduction in time. Here is what each billing model does.
Revenue and effective rate under a 60 percent time reduction
CEOtudent editorial model. Illustrative figures calculated by CEOtudent from the stated assumptions: a 10-hour task at 120 dollars per hour falling to 4 hours. These are worked examples, not market rates.
| Billing model | Invoice before | Invoice after | Change | Effective rate before | Effective rate after |
|---|---|---|---|---|---|
| Hourly, billing actual time | 1,200 dollars | 480 dollars | minus 720 dollars, minus 60% | 120 dollars/hour | 120 dollars/hour |
| Fixed fee, price unchanged | 1,200 dollars | 1,200 dollars | no change | 120 dollars/hour | 300 dollars/hour |
| Fixed fee, repriced down 25% | 1,200 dollars | 900 dollars | minus 300 dollars, minus 25% | 120 dollars/hour | 225 dollars/hour |
| Outcome price at 3,000 dollars | 3,000 dollars | 3,000 dollars | no change | 300 dollars/hour | 750 dollars/hour |
| Monthly retainer, 4,000 dollars, 30 hours falling to 12 | 4,000 dollars | 4,000 dollars | no change | 133 dollars/hour | 333 dollars/hour |
The hourly row is where the panic comes from, and it deserves a second look, because the loss is conditional rather than automatic.
Your 10 hours of capacity now fits 2.5 of these deliverables. Sold at 480 dollars each, that is 1,200 dollars for the same 10 hours, identical to the baseline. The client pays 60 percent less per deliverable and you earn exactly what you did before.
That is the whole game in one line. Hourly billing does not destroy your income when AI makes you faster. It destroys your income only when you cannot refill the freed capacity. The real problem is a demand problem wearing a pricing problem’s clothing, and repricing will not fix it if the pipeline is empty.
This also explains why the fixed-fee and outcome rows feel so attractive and why regulators look at them carefully. In those rows the client’s payment is unchanged while the work behind it shrank by 60 percent. Somebody captured a 60 percent efficiency gain, and it was not the buyer. That may be perfectly legitimate, because the buyer purchased a result rather than your calendar. It is also exactly the situation Formal Opinion 512 flags as potentially unreasonable when the gap gets large enough.
A disclosure test you can actually apply
The most useful thing in Formal Opinion 512 is not a rule. It is a set of triggers, because it converts an anxious judgement call into a checklist.
Under Model Rule 1.4 the opinion concludes that in some circumstances client disclosure may be unnecessary. But disclosure is required if the client asks how the work was done or whether AI was used, if the engagement agreement or the client’s guidelines require it, if you propose to input client information into the tool, if the AI use is relevant to the basis or reasonableness of your fee, or if the output will influence a significant decision. It adds a sixth situation: where a client hired you specifically for your own skill and judgement, undisclosed AI use may violate the engagement terms or the client’s reasonable expectations.
The opinion also lists the factors that decide the closer calls: the tool’s importance to the task, that task’s significance to the overall work, how the tool processes client information, and the extent to which knowing would affect the client’s evaluation of or confidence in your work.
The disclosure decision test
CEOtudent editorial framework, adapting the trigger list in ABA Formal Opinion 512 for independent professionals who are not bound by legal ethics rules.
| Question | If yes |
|---|---|
| Did the client ask how the work was produced? | Answer honestly and specifically. This is the one absolute. |
| Does the contract, brief or client policy require disclosure? | Disclose. Check before you start, not after you deliver. |
| Will client data, documents or confidential material go into the tool? | Get explicit permission first. Confidentiality outranks convenience. |
| Does AI use change the basis of your price? | Disclose. If efficiency is the reason for the number, the client is entitled to the reason. |
| Will the output drive a decision the client cannot easily reverse? | Disclose, and state what you verified yourself. |
| Was your personal judgement the thing the client was buying? | Disclose. Substituting a tool for the skill they hired is the trust failure, whatever the output quality. |
| None of the above? | Disclosure is optional. Doing it anyway is usually a competitive advantage. |
The last line is not a platitude. Under the EU AI Act’s own carve-out, the thing that discharges the obligation is human review plus named editorial responsibility. Stating who checked the work is a stronger signal than either silence or a generic AI label, and it is the standard the law itself reaches for.
How to price this without lying to anyone
Stop selling time for work AI genuinely compresses, but fix demand first. The arithmetic above shows hourly billing is survivable at full capacity. If your pipeline cannot absorb 2.5 times the throughput, changing your pricing model is treating a symptom.
Price the residual, not the artefact. What AI compressed is production. What it did not compress is scoping the problem, judging whether the output is right, and carrying responsibility if it is wrong. The Copyright Office draws the same line from a different direction: authorship attaches where a human determines the expressive elements, not where a prompt was typed. If your price is attached to the part that survives, the compression stops threatening you.
Do not promise rights you cannot deliver. If your contract assigns copyright in the deliverable, check whether there is copyright to assign. For largely prompt-generated material under current US guidance, there may not be. Either raise your own contribution to the level where protection attaches, or say plainly what the client is getting.
Pass tool costs through at cost, or absorb them. This is the cleanest rule in the whole opinion and it generalises well. Subscriptions you use across all clients are overhead. Per-use costs incurred for one client can be billed at actual cost, without a markup. Marking up an API bill is the kind of small dishonesty that costs a relationship.
Write the disclosure position into the engagement, not the invoice. The opinion notes the engagement agreement is the logical place for it. Setting the terms before the work removes the incentive to stay quiet afterwards, which is where trust actually breaks.
Never let efficiency become an excuse for unreviewed output. Every instrument above converges on the same point. The billing rule assumes you spent time reviewing the draft. The EU carve-out only applies if human review happened. The ownership rule only protects what a human shaped. Review is not the overhead standing between you and the margin. Review is the product.
For the channel side of this question, see our analysis of what the Upwork and Fiverr filings reveal about freelance channel economics. For structuring fixed prices without absorbing scope risk, see our productized service pricing framework, and for converting project work into recurring revenue, see the retainer playbook.
Frequently asked questions
I am not a lawyer. Why should ABA Formal Opinion 512 apply to me?
It does not bind you. It is worth reading because it is the most carefully reasoned public answer to the exact question, produced by people who had to resolve it rather than market a position. Clients increasingly bring these expectations from regulated contexts into unregulated ones.
If I bill hourly and AI halves my time, am I obliged to charge less?
Under the legal ethics rule, yes: hourly means actual time. As an independent worker you are not bound by that rule, but the expectation is the same one your client already holds. The durable answer is to stop selling time for compressible work, not to invoice hours you did not work.
Can I just switch to fixed or value-based pricing and keep the full gain?
Commercially you can, and often you should. But Formal Opinion 512 explicitly says it may be unreasonable to charge the same flat fee when AI makes the task much faster. Fixed pricing is a legitimate answer when the client is genuinely buying an outcome. It is not a loophole that makes the fairness question disappear.
Do I have to tell clients I used AI?
It depends, and the triggers are listed above. You must if they ask, if the contract says so, if their data goes into the tool, if it affects the price, if it drives a significant decision, or if they hired your personal judgement. Otherwise it is a choice, and disclosing is usually the stronger position.
Is the EU AI Act transparency requirement actually in force?
Yes. Article 50 has applied since 2 August 2026 under Article 113, and the 2026 Digital Omnibus did not delay it. The Omnibus did give providers whose generative systems were on the market before that date until 2 December 2026 to meet the machine-readable marking duty in Article 50(2).
Does my client own AI-generated work I deliver?
Not automatically. The US Copyright Office concluded in January 2025 that prompts alone do not make the user the author of the output. Protection attaches where a human determines the expressive elements. This is a US position on US law, and jurisdictions differ, but it is the clearest official statement available and it should shape what your contracts promise.
Should I charge clients for my AI subscriptions?
Treat general subscriptions as overhead, the same way you treat your laptop. Costs incurred per use for one specific client can reasonably be billed at what you actually paid, without a markup, and with any discount passed along.
Sources and further reading
American Bar Association, Standing Committee on Ethics and Professional Responsibility, Formal Opinion 512, Generative Artificial Intelligence Tools, 29 July 2024.
American Bar Association, Standing Committee on Ethics and Professional Responsibility, Formal Opinion 93-379, Billing for Professional Fees, Disbursements and Other Expenses, 1993.
Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence, Articles 50 and 113.
Regulation (EU) 2026/1744 of the European Parliament and of the Council of 8 July 2026 amending Regulations (EU) 2024/1689, (EU) 2018/1139 and (EU) 2023/1230 as regards the simplification of the implementation of harmonised rules on artificial intelligence, Digital Omnibus on AI, Official Journal L series, 24 July 2026.
United States Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability, A Report of the Register of Copyrights, January 2025.
American Bar Association, Model Rules of Professional Conduct, Rules 1.4 and 1.5.
Upwork, Inc., Annual Report on Form 10-K for the fiscal year ended 31 December 2025, filed with the United States Securities and Exchange Commission.
Fiverr International Ltd., Annual Report on Form 20-F for the fiscal year ended 31 December 2025, filed with the United States Securities and Exchange Commission.
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
This post is also available in:
