AI for Designers in 2026: Feedback Rounds, Handoff and Who Owns the Output

AI can already sort client feedback into a change list, widen a moodboard, name layers, suggest design tokens and cluster research notes. It can't hold a concept together, read what a client really objects to or hand you clean copyright: the US Copyright Office says prompts alone don't make you the author.

This page is for working designers, freelance, in a studio or in-house. It goes through a design project stage by stage, compares Figma AI, Adobe Firefly and Canva Magic Design, and covers the rights questions a client's lawyer may eventually ask.

Where a designer's hours go besides designing

Scoping, presenting and revising are written into the job. The US government's handbook entry for graphic designers lists meeting clients or art directors to define scope, presenting concepts, revising them from feedback and checking for errors before print or publishing. O*NET's task list adds keeping an archive of past work and preparing print files. About 20% of graphic designers are self-employed, so for many of them proposals, contracts and late invoices sit on top.

Picture a normal Tuesday. The client's marketing lead leaves six comments in the file, the founder forwards a screenshot with arrows on it, and someone replies at 9 p.m. asking whether the logo can feel "more premium." Before you move a single vector, you're reconciling three people's opinions. That reconciling is a good first job for AI, and it needs no generator at all.

From brief to handoff: where AI helps

AI is strongest at the text around the work and at repetitive file chores, and weakest at the decisions a client pays you for. Prices and limits change often, so check the linked pages before buying.

Figma AI, Adobe Firefly or Canva Magic Design?

Figma AI, Adobe Firefly and Canva Magic Design solve different problems, so the choice is rarely either-or. Figma AI works inside your design files, Firefly generates assets, and Magic Design gets you to a finished template quickly. Read the last column before any of them touches client work.

The practical split is exploration versus final. A generated image on an internal moodboard carries far less risk than a generated mark on a client's packaging. Inside Firefly, note which model made each image, because Adobe's own and a partner's come with different assurances in the same interface.

There's a craft risk too. Show a client a generated direction and you may have promised a look you can't reproduce, license or extend across a system. Label early AI explorations as references, not concepts.

Can you copyright AI-generated design work?

In the US, you can copyright the human parts of a design, not the parts a machine determined. The Copyright Office's registration guidance says prompts work more like instructions to a commissioned artist, because the system decides how to carry them out. Rounds of corrective prompts don't change that: you gain influence, but the tool still decides how to apply each note. The Office's January 2025 report holds the same line.

Human work around the output can still earn protection:

• Selection and arrangement. Composing AI elements creatively can make the overall work protectable, though not the elements themselves. A graphic novel with human text and Midjourney images was protected as a work, while its images were not.
• Real modification. Redrawing or altering output enough to meet the normal standard for original authorship protects your changes.
• Ordinary editing. An artist editing in Adobe Photoshop remains the author, the guidance says, though a footnote puts Photoshop's AI-powered features under the same case-by-case analysis.

Registration adds a duty. Applicants must disclose AI-generated content, exclude anything more than a trivial amount (de minimis), use the Standard Application and describe what the human contributed. Skip that and the Office may cancel the registration, or a court may disregard it if you knowingly gave inaccurate information.

The courts have settled only part of this. The Supreme Court declined to hear Thaler v. Perlmutter on March 2, 2026, leaving in place the ruling that an author must be human. How much human involvement is enough remains open, and Allen v. Perlmutter, over a Midjourney image made with more than 600 prompts, is testing it.

For client work the consequence is concrete. If a client is buying exclusive rights to a logo and the mark came straight out of a generator, you may have little copyright to transfer. Keep your sketches, layered files and version history. This is general information about US rules, not legal advice.

Licensing, contracts and settings to check before client work

The rights question usually reaches you through a contract, a tool setting or a trademark search, long before any court. Check these before a client file goes near a model:

• Training settings. Content training is on by default on Figma's Starter and Professional plans, and off on Organization and Enterprise. A team admin can switch it off under Settings, in the AI section.
• Your NDA. A client's unreleased product shot is still confidential inside a prompt box. If the agreement is silent, ask.
• Your warranty clause. If your contract promises original, non-infringing work, compare that with your tool's terms. Figma's and Canva's don't promise uniqueness.
• Trademark clearance for logos. The USPTO tells you to search design marks by design code, and warns that a clean search is no guarantee of registration.
• Your own AI clause. Consider a clause barring clients from training models on delivered work or using it as input for variations. The Dieline reported in October 2026 on designers adding them.

Telling clients is a judgment call, and AIGA's statement on AI leans toward it, asking designers to disclose the role AI plays in the creative process. A short line in your scope of work settles it before anyone asks:

AI tools may be used for research, internal exploration and production tasks such as background removal. Any generated element in a final deliverable will be identified in writing before delivery, along with the tool used. Client confidential files will not be uploaded to AI tools without written approval.

Taste, accountability and the room stay with you. AI can tell you three people disagree about the logo. Deciding that the founder's objection is really about price positioning, then defending a direction in the presentation, is the job. So is the final proof: check bleeds, color and copy yourself before anything goes to print.

Merlin and the client side of design work

Merlin, a Mac assistant, covers the client side of a designer's work, not the canvas. It generates no layouts or brand assets and doesn't do invoicing, and Figma and Adobe aren't among its featured app connections, so your design tools stay where they are. It helps around a feedback round, drawing on Gmail and Google Calendar:

• Before a presentation: Prep me on a calendar event sets up a sourced brief on who's attending, with earlier notes and email threads, decisions, commitments and open questions.
• During the feedback call: the app records the call, and on paid plans you can search a speaker-labeled record of it, so the change list starts from what the client actually said. The audio is kept locally, but paid plans upload it for transcription and your Merlin account holds the transcript and notes, so check your NDA first. Free accounts can record but get no transcript. Get everyone's consent first, since some US states require it.
• After it: tell Merlin what the client agreed and have it draft the recap email; nothing sends until you approve the draft.
• Between projects: the People panel shows clients waiting on your reply, and, under Gone quiet, old clients whose regular emails have stopped. Both lists come from email alone, so a note in a Figma comment or on Slack won't register there.

How to try AI on your next project

1. Check your training setting and your NDA before any client file goes into a tool.
2. Start with the feedback round, since it carries the least creative risk. Paste the comments, emails and any call transcript into an assistant with the prompt below.
3. Run your next research session's notes through Cluster stickies in FigJam, or a recording through a transcription tool, then check the themes against the raw notes.
4. Keep a simple log for each generated asset: tool, model, plan, date and what you changed. It answers a client's question later and helps with any copyright registration.
5. Add an AI line to your next scope of work, and have a lawyer read it once.

Below is all the client feedback on version 3 of this design, from comments, email and a call transcript. Make one change list, grouped by screen or page. For each item give who asked, their words, and the change requested. Mark any items where two people contradict each other. List anything too vague to act on as a question I should ask. Don't suggest design solutions.