Ahmed Shaami
Two numbers from the same report sit oddly next to each other. In Upwork’s In-Demand Skills 2026 report, AI image generation and editing grew 95% year on year. Logo design, the work that AI image tools were supposed to have made obsolete, grew 44% over the same period, and graphic design held its place as the single most in-demand skill in the Design and Creative category. Clients are buying more image generation and more logo design at the same time, which only looks like a contradiction if you think a logo is a picture.
Where those numbers come from
Upwork’s methodology is worth knowing before you lean on the figures. The data covers freelancer earnings on completed jobs between 1 January and 31 December 2025, compared with the same period in 2024, with demand originating in the United States. Only skills with at least $100,000 in aggregate freelancer earnings in a category were counted, which filters out the tiny-base growth that makes for misleading headlines.
So logo design at plus 44% is real money changing hands, not a rounding artefact. It placed third-fastest-growing in Design and Creative, behind AI video generation and editing at plus 329% and AI image generation and editing at plus 95%.
Why does a client still pay for a logo they could generate themselves?
Because generating a mark and owning a brand are different purchases. A client who types a prompt gets an image in about ten seconds. What they do not get is a mark that works at 16 pixels on a browser tab and on a shop sign, a horizontal lockup for a website header and a stacked one for a social avatar, a monochrome version for a single-colour print job, defined clear space, a colour system that survives on a dark background, and a written rationale they can use to defend the choice to a business partner.
They also do not get someone accountable. When a logo needs to be redrawn because it turns to mud at small sizes, a freelancer fixes it. A prompt does not.
What the copyright rules actually say
This is the part that changes the sales conversation, and it is the part most freelance advice gets wrong in both directions.
The US Copyright Office has been explicit that copyright protects only material that is the product of human creativity, and that the term author excludes non-humans. Its registration guidance for works containing AI-generated material sets out the test it applies: whether the work is basically one of human authorship with the computer merely an assisting instrument, or whether the traditional elements of authorship were conceived and executed by a machine rather than a person.
Two practical consequences follow. First, purely AI-generated elements do not carry copyright protection, so a client cannot rely on copyright to stop a competitor reproducing that part of the artwork. Second, anyone registering a work that contains more than a trivial amount of AI-generated material has a duty to disclose it and to explain what the human author contributed. The Office’s Part 2 report on copyrightability goes through the reasoning in detail. Work where a human made the creative decisions and used AI as a tool can still be registered, and has been.
Copyright and trademark are not the same problem
Here is where a lot of confidently-written advice goes astray. You will read that an AI-generated logo cannot be trademarked. That overstates it considerably.
Trademark protection turns on use in commerce and on whether a mark distinguishes one seller’s goods from another’s. It is a different legal regime from copyright, with different requirements, and it does not hinge on who or what authored the artwork in the way copyright does. A business using an AI-assisted logo consistently in commerce is generally not locked out of trademark protection by that fact alone.
The honest version to give a client is narrower and more useful: copyright in the artwork itself is where the exposure sits, disclosure obligations apply on registration, and the two systems protect different things. That is a more credible thing to say than a blanket warning, and clients can tell the difference. None of this is legal advice, and a client making a real brand investment should hear it from a lawyer in their own country.
How to position the work now
Stop selling a file and start selling a decision that has been made properly. In practice that means quoting a package rather than an image: discovery on the client’s market and competitors, two or three distinct directions rather than forty variations, the full family of lockups and file formats, and a short guidelines document covering colour, spacing and misuse.
It also means being able to explain your process when asked. A client who has already generated fifty logos on their phone and disliked all of them is not sceptical of designers, they are exhausted by options and looking for judgement. Walk them through why you cut the other directions. That conversation is the thing they are actually paying for, and it is the reason this work sits on the more durable side of the line described in which AI freelance work pays more and which is getting cheaper.
Price accordingly. If you quote against the bottom of the market you are implicitly agreeing that you sell images, and no freelancer wins that comparison against a tool that costs a few dollars a month.
Where AI genuinely belongs in the process
None of this argues for avoiding the tools. Using them to explore directions quickly, to mock up how a mark looks on packaging or a storefront, or to generate throwaway variations for your own elimination process is sensible, and the market clearly rewards people who work this way given how fast AI-adjacent design skills are growing.
What matters is where the human judgement enters and whether you can describe it. Keep your sketches, your rejected directions and your reasoning, because that record is both your defence of the fee and, if a client ever registers the mark, the evidence of human authorship the process depends on. It is also worth agreeing up front how AI use is handled in the engagement, which is what the AI clause your freelance contract probably does not have yet is for.
The wider pattern holds across this year’s data. Demand is moving toward people who can direct a tool and take responsibility for the result, not toward the tool itself, which is roughly what the numbers in will AI replace freelancers have been saying all year.
