Power Tools: using AI well when your clients use it too

Scarcely a week goes by without a designer bringing me a fresh grievance: some new use of AI, wielded against them by a client drunk on faux omniscience.

Talking clients down from the impossible, the unbuildable and the simply ugly is now part of the job. The real skill is in doing this without sounding like a bitter ex.

Here's a typical scenario: your client receives the concept presentation, uploads it to ChatGPT with a request to make it cheaper, or warmer, or more like something seen in a hotel in Lisbon. By morning the client has three variations of your scheme and, hey presto, the client has become the designer.

Sadly, this is a legitimate business threat, and I will come to what you can do about it. But it is worth noticing first that the act itself is not new to the profession. Many designers have already performed it, in the other direction. The floor plan with the client's address on it, pasted in to ask for furniture layouts. The budget spreadsheet, uploaded for a tidier version. Photographs of a child's bedroom or a dressing room, fed to a free image tool to test a colour. Designers admit to putting their accounts into chatbots. In a UK survey, two in five office workers said they or a colleague had put sensitive customer or financial data into a public AI tool.

It is the same upload. Only the direction differs. And that symmetry helps us see the most useful way to organise everything a small practice needs to decide about AI, because the standard you can credibly ask of a client needs to be the standard you can show you keep yourself.

Your side of the upload

Start with the law, because it is less ambiguous here than people assume. The Information Commissioner's position is that there is no AI exemption from data protection. If you put personal data into an AI tool, you are processing it, and you remain responsible for what happens next. Personal data is broader than names: a floor plan tied to an address, a photograph of the inside of someone's home, an email about a divorce and its effect on the brief. The ICO is explicit that you should not rely on the tool to strip any of this out. Responsibility rests with the user.

In practice, good conduct comes down to five habits.

Use business tiers, with training switched off. Free and consumer versions of most tools reserve the right to use what you type to improve their models. Paid business plans generally offer proper data-processing terms and an opt-out from training. The difference in cost is small; the difference in your position is not.

Strip before you share. Remove names, addresses and anything identifying before a document goes in. Ask for help with "a three-bedroom Victorian terrace, family of four", not with the Hendersons at number twelve.

Treat health as off-limits. A client's mobility, a child's sensory needs, a diagnosis that shapes the brief: this is special category data, which the law protects more heavily than anything else you hold. It should not go into a general-purpose chatbot at all.

Keep a list. Write down which tools you use, for what, on which plan, and where their servers are. Most are American, which makes their use an international transfer of data. This list is the backbone of everything that follows, and it takes about an hour.

Keep a person at the beginning and the end. One New York practice has a rule it calls Human Start and Human Finish: AI may help in the middle, but nothing it generates leaves the studio as a final deliverable without a designer having drawn, checked or verified it. This is a quality rule as much as an ethical one, since accountability for every recommendation stays with you, whatever produced the first draft.

Telling clients how you work

Transparency has two layers, and most practices need both.

The first is legal. Your privacy notice must tell clients who receives their personal data. If AI providers process it, they belong there, at least by category, along with the fact of any transfer abroad. A notice written five years ago that says nothing about AI while you use it daily is a gap, and the ICO is preparing transparency guidance aimed specifically at small businesses.

The second is professional, and it is where an obligation can become a mark of quality: a short, plain-English statement in your onboarding pack saying what you use AI for and what you never use it for. Something like this. We use AI tools to help draft correspondence, organise schedules and explore early ideas. We use paid, private versions. We never put your personal details, photographs of your home or anything about your health into them. Every drawing, specification and recommendation you receive has been made or checked by a designer. Any AI-generated image we show you will be labelled as such.

In a business built entirely on trust, transparency is simply a form of honesty, and ambiguity is a luxury no designer can afford. Some designers now label every AI-assisted image in a mood board, and I understand why. A few years ago, before AI imagery was everywhere, I watched a designer rework her public portfolio over several months, filling it with aspirational “projects”, presented as real, that had all the dayglo authenticity of a plastic lawn. It coloured everything I thought I knew about her business. Once I doubted the photographs, I doubted absolutely everything else.

The statement does something else, too. It makes the reciprocal request possible. Having said plainly how you treat what clients give you, you are in a far better position to say how you would like them to treat what you give them.

Their side of the upload

Here the news is less comfortable. Copyright protects the things you put on paper or screen: your drawings, renderings and presentation boards. It does not protect the room itself. Nor, in most cases, does it protect the joinery or furniture someone makes from your drawings. The law treats a drawing of an object as a set of instructions, and following instructions is not copying them. So if a client hands your cabinetry drawings to another joiner, your copyright has not been infringed. A bespoke piece may have some separate protection for its shape, but that protection is limited and lasts only a few years.

A client who uploads your scheme to an AI tool and asks for variations may therefore end up with something that does not reproduce enough of your work to infringe at all. Based on my (AI) search of the UK government's report on copyright and AI in March, it made no immediate changes, and indicated that the provision protecting computer-generated works is likely to go. That cuts both ways. A designer who leans heavily on AI-generated imagery has a thinner claim to it than they suppose.

So the protection has to be built where you have control, which is the contract. The useful clauses are not exotic:

• You retain copyright in everything you produce. Once your fees have been paid in full, the client receives a licence to use it for the named project and property only.
• If the engagement ends early, the licence extends only to work that has been paid for.
• That licence expressly excludes uploading your drawings, renderings or presentations to generative AI tools, or using them to produce derivative schemes, without your written consent.
• Drawings are issued as PDFs, not editable files.
• Alternatives the client generates elsewhere and asks you to assess are treated as change requests, within your revision allowance or charged at your stated rate.

That last point is apposite, as the most recent exchange I had with an exasperated designer arose from her client’s request that she review the client's husband’s AI homework as a favour.

I am not a lawyer, and you should have a solicitor look at the wording. But it is worth understanding what such a clause does even where it would be hard to enforce. It states a norm before any work begins, at the moment the client is most disposed to agree with you, and it means that when the issue arises you are pointing to something already accepted rather than raising an objection.

What clients are actually doing - AI, and fluency without situational awareness

There are no surveys yet, but the anecdotes are consistent, and they fall into four patterns.

The second opinion that never sleeps. Before discovery most clients have already consulted AI. AI attends our client meetings, like an upstart intern, always on hand to contribute its tuppence worth. Budgets run through AI are challenged, (when we know that asking the tool to generate the budget in the first instance would produce a different outcome because the tool tends to agree with whoever is asking).

Mitigation: firmer recommendations, with the reasons given. Designers who once guided lightly are finding that a client with a tireless second opinion needs a clearer first one.

The unbuildable picture. AI renders are persuasive precisely because they ignore everything that makes interiors difficult: clearances, structure, fire ratings, lead times, the radiator under the window. A designer recently told me of money “saved” by their client getting AI to create an utterly impossible joinery solution. Time spent disproving the picture is time nobody is paying for.

Mitigation: say at the first meeting what an image is and is not, and keep visualisation subordinate to drawings.

The endless tweak. Because variations now cost nothing, some clients generate them indefinitely. Some designers have stopped sharing early concept sketches for exactly this reason.

Mitigation: defined revision rounds; concept work presented in person or as fixed files; the change-request clause above. 

The fee question. If AI does the work, the argument runs, the work should cost less.

Mitigation: stop selling time. A fee built on hours invites exactly this challenge, and every efficiency AI gives you becomes a discount you are expected to pass on. A fee built on the outcome, the judgement and the responsibility you carry does not.

I’m jumping in here, as I feel defensive. I argue hard that your fees should be linked to time on the job. However, while I advise using hours behind the scenes to calculate fees (so that they cover all of your time-on-task), I also advise that the fee proposal as presented is linked to a given scope – therefore, your fees are for outcomes, and the hours calculation is secret back-office data.

What links all four is that AI gives clients fluency without situational awareness. It can produce a convincing room; it cannot know this room, this budget, this contractor, this family. That gap is not a threat to your value. It is a precise description of it.

Showing them how to use it well

Which is why the most generous response is not prohibition but guidance. A client who uses AI well is a better client, and it costs you nothing to tell them how.

It is genuinely useful for articulating a brief before you meet; for collecting and naming what they like; for learning the vocabulary so that conversations move faster; for drafting the questions they want to ask you; and for living with an idea for a few days to see whether they still love it.

It is unhelpful as a specification, a quantity surveyor, a source of building regulations advice, an arbiter of your fees, or a way of producing a scheme to take elsewhere. Say so kindly, early, and in writing, ideally in the same document as your own AI statement, so that the two read as one standard applied in both directions.

The standard you can ask for

None of this requires you to be anxious about AI or wary of clients who enjoy it. It requires a list; a paragraph in your privacy policy and another in your Onboarding pack; five-ish contract clauses; and a clear conversation at the start. Most of it is an afternoon's work.

But the order matters. Put your own side right first: the private tools, the stripped documents, the honest labels, the notice that tells the truth. Then you can ask clients for the same courtesy without it sounding like suspicion, because it isn't. It is simply the standard you already keep.