← Back to blog

Strategy · November 5, 2026 · 8 min read

Should You Tell Clients You Use AI to Write?

Yes, almost always, and put it in the scope of work. Clients rarely object to AI drafting; they object to finding out after the fact.

By FluxWriter

Should I tell clients I use AI to write content is a question most agency owners ask far too late, usually after the work has already shipped. The answer is almost always yes, and the place to do it is the contract rather than an awkward conversation six months in. Almost nobody objects to the tool itself; they object to finding out from someone else. This walks through three levels of disclosure, the exact sentence for each, and the two cases where saying nothing is the expensive option.

Clients Object To The Surprise, Not The Software

The objection agency owners rehearse for is "you used a robot". The objection that actually arrives is "you never told me". Those are different problems, and only one of them is common.

Every agency runs on tools the client never asked about. Keyword research software, a grammar checker, stock photography, a transcription service, a subcontracted writer in another time zone. None of that gets disclosed and none of it causes trouble, because the client never formed a belief about it. AI drafting causes trouble for one reason: the client has formed a belief, usually from your own proposal, about who writes the words.

So this is not really an ethics question. It is a contract question, and you can answer it by rereading what you sold. If your proposal says "written by our senior strategists", AI drafting contradicts a promise you made, and no amount of quality in the output repairs that. If it says "eight publish-ready posts a month, briefed, edited and approved by a named editor", the drafting method was never part of the promise and you are on much steadier ground.

The difference between those two proposals is about fifteen words. That is the whole exposure for most agencies.

The Disclosure Question Has Three Settings, Not Two

Most people treat this as binary: tell them or do not. There are three settings, and the middle one is where almost all client work belongs.

Silent tool use means no client-facing statement at all. The tool is a production choice, like your CMS or your rank tracker. This holds up when the deliverable is defined by output rather than method, and when nothing you signed promises human authorship. Even here, write one line into your own process document so everyone on your team answers the question identically: "All drafts are machine-assisted and pass a named-editor review before delivery." The failure mode of the first setting is not the silence. It is two employees giving one client two different answers.

The SOW clause is the default, and the one worth adopting this quarter. One sentence in the scope of work, signed and dated: "Content is produced using a combination of AI drafting tools and human editing. Every deliverable is researched, fact-checked and approved by a named editor before it reaches you." Whether that wording is sufficient for your situation, and how it sits alongside the rest of your agreement, is a question for whoever reviews your contracts. The point here is only that it exists, in writing, before the first invoice.

The on-page note is a line on the published article: "Drafted with AI assistance and edited by [Name], [Title]." This one is not your decision to make. It belongs to the publisher, and the publisher is your client. Offer it, mention that some audiences read it as confidence rather than confession, and let them choose.

The Three Settings Map Onto Deliverables Predictably

Below is how the settings line up against the work most agencies actually sell, with the line to use when the question arrives by email rather than in a meeting. There are no numbers in it, because the right setting depends on what you signed rather than on volume.

Two of those rows do most of the damage when they go wrong. The executive byline and the regulated client are the cases where silence is the risk rather than the shortcut. The other rows usually survive a late disclosure with nothing worse than an awkward call, which is why they are worth taking one at a time.

Silence Costs More Than Disclosure In Exactly Two Situations

Regulated industries. Finance, health, insurance and legal clients usually run an internal review process that exists regardless of what you think about AI. Do not assume you know which rules apply to them, because you almost certainly do not, and they differ by market and by client. The useful move is to hand the question back: ask who on their side approves how content is produced, and whether they need the method recorded in writing. Then put their answer in the file with a date on it. You are not making a compliance judgement. You are making sure somebody qualified made one.

Ghost-bylined executive content. A named human byline carries an implied claim that the named human wrote it, or at minimum said it. When a founder's name sits on a post they have never read a draft of, the exposure is personal and reputational, and it lands on them rather than on you. That asymmetry is the entire reason to ask first.

Ask the person whose name is going on it, not their marketing manager. The question is short: do you want this drafted for you to edit, or written from an interview with you? Those are two different products at two different prices, and having the client choose one of them is worth more to the relationship than any clause you could write.

The Client Is Buying The Review Step, Not The Keystrokes

The instinct behind staying quiet is usually a pricing fear. If the client knows a machine drafted it, they will want to pay less. Look at where the hours actually go before accepting that.

Take a 1,500-word post and assume the following split, which is illustrative rather than a benchmark. Brief and research, 45 minutes. Human first draft, four hours. Editing and fact-checking, one hour. Internal links and publishing, 30 minutes. That is 6.25 hours per post. Now swap out the draft: 15 minutes to prompt and generate, with editing rising to 90 minutes because you are reworking someone else's structure rather than your own. That comes to three hours.

Across eight posts a month, the human path is 50 hours and the assisted path is 24. The saving is real, and it is a little over half. But notice what did not disappear. The brief, the research, the fact-check, the internal linking and the publishing all survive intact, and those are where both the value and the risk live. The client is buying those 24 hours of judgement.

The Rate Conversation Arrives A Month After The Disclosure Does

It is easier than people expect, provided you have already reframed the deliverable. The price was never a function of drafting hours. It was a function of an editor who is accountable for what gets published under the client's domain, and that person still exists.

You can hold a rate on that basis. What you cannot do is hold a rate while refusing to say what changed. If a client discovers a method change and a flat price in the same week, they will read the two as connected whether or not they are.

There is also a version of this that runs in your favour. Some agencies raise output instead of cutting price: the same retainer, twelve posts instead of eight, with the extra four aimed at long-tail queries nobody had time for before. On the illustrative numbers above that is 36 hours rather than 50, so the economics still improve. More usefully, it turns the conversation towards their results and away from your costs.

FAQ

Will a client leave if we tell them?

The disclosure itself rarely ends a relationship. The ones that end are the ones where the client found out some other way, months in, and then went back and reread the proposal. Telling them costs one uncomfortable email. Not telling them can cost the account and the referrals attached to it.

Does content drafted with AI get penalised in search?

Google's published guidance focuses on whether content is helpful and original rather than on how it was produced, and it treats automation aimed at manipulating rankings as the problem. Thin, unedited output tends to underperform on its own merits regardless of what produced it. The practical read is that the edit pass matters more than the drafting method.

What do we say when a client asks which tool we use?

Name it. Refusing turns an ordinary production detail into a secret, and the client will then assume the worst version they can imagine. Most clients ask once, hear the answer, and never raise it again.

The Practical Takeaway

Open the last three proposals or scopes of work you signed and search each one for the word "written". Whatever it promises about authorship is the contract you are genuinely operating under, and for most agencies at least one of those three will say something they can no longer honour. Then add the SOW sentence to your template so every new agreement carries it, and make a short list of the clients whose content runs under a personal byline, because those need a direct conversation rather than a clause. Start with the search: five minutes, three documents, and you will know whether you have a disclosure problem or only a disclosure preference.

← All posts