Will they say yes

This is the page that decides the business. If working creatives will not sell judgement about AI output at any price, there is no supply and nothing else on this site matters. The evidence says they will — but the reason is not the one most decks give, and the strongest counter-argument is not about money.

#Stated attitude 81%, revealed behaviour 10%

Ten percent is the best available estimate of the hard-refusal population among working creatives. It is a lower bound — many users never found the setting — and an upper bound, since clicking a toggle costs nothing while refusing paid work costs money. Either way it is far from a majority. For a business that needs to recruit thousands rather than millions, a one-in-ten hard-refusal rate is not a constraint.

Adobe rounds the picture out: 86% of global creators report using generative AI in their work (Adobe 2025, via Figma), matching Creative Boom's 86% — against Creative Boom's finding that only 10% think the overall effect is positive, with 58% "mixed" and 28% negative (Creative Boom). Usage is near-universal, approval is at 10%. Those are not in conflict. They describe a workforce complying under duress — and a recruiting proposition that offers agency rather than tools is aimed exactly at that gap.

#Sentiment is a sampling artefact, and that is exploitable

Put two surveys of the same profession side by side.

81% say AI dulls creativity. 80% say AI will enhance it. Same profession, opposite answers. These are not contradictory measurements of one population. They are accurate measurements of two — a general creative-professional sample, and a paid, self-selected sample drawn from a marketplace that had already onboarded them to AI work.

One academic finding sharpens the funnel design. Across three experiments of 100 participants each, learning how AI systems actually work makes people judge AI art as less morally acceptable — especially where financial gain is involved — while leaving aesthetic appreciation unchanged (Scientific American, on Cognition). The more you explain about mechanism, the more moral resistance you generate — yet transparency is what makes you credible. There is no clean resolution. The least-bad answer is to be maximally transparent about terms and minimally lecture-y about mechanism.

#The Sora revolt: every demand was a term sheet

This is the strongest evidence for the distinction the whole business rests on — that selling judgement about AI output is a different transaction from selling work into it.

In November 2024, artists in OpenAI's Sora alpha leaked access to the model in protest. Roughly 380 verified artists and creators signed the accompanying letter:

Read what is absent. Nobody said evaluating AI output was beneath them. Nobody said critiquing a generative model betrays artists. They objected to unpaid or undercompensated feedback labour, to being used for PR, and to the requirement that "every output needs to be approved by the OpenAI team before sharing."

Unpaid. Uncredited. Approval-gated. Every one of those is a term sheet, not a principle. This is the closest thing to a controlled experiment anyone has run on the question, and it points the right way: the same artists who will not license their portfolio for training will do red-teaming and evaluation — they demand to be paid and named for it.

The sequel confirms it. OpenAI's better-structured Sora Selects programme funded 10 artists and artist teams out of a $3 million initiative built over roughly ten months of relationship-building, with participants including Boris Eldagsen and shy kids (Artnet). Eldagsen still voiced ethical concerns about copyright being "outsourced to the users." The programme did not eliminate objection. It converted objection from a boycott into a negotiation — which is exactly the outcome to model.

Two supporting facts. The largest organised expression of creative opposition is Ed Newton-Rex's Statement on AI Training, now carrying more than 48,000 signatures (ed.newtonrex.com). Its entire text is one sentence: "The unlicensed use of creative works for training generative AI is a major, unjust threat to the livelihoods of the people behind those works, and must not be permitted." The whole thesis lives in one word — "unlicensed" — and the sentence is textually about training. It is silent on judgement, critique, evaluation or red-teaming. And no equivalent mass statement against paid evaluation exists [WEAK — the absence of an anti-evaluation movement is not proof that one would not form].

The litigation record pushes the same way, for an unobvious reason. Getty largely lost in the UK: it abandoned primary copyright infringement mid-trial because training happened overseas, and the court rejected secondary infringement on the ground that the model "does not store the training data itself" (judgment PDF; Ropes & Gray). Andersen v. Stability survived a motion to dismiss on direct and induced infringement and Lanham Act trade dress, with trial set for 5 April 2027 (AI Lawsuit Tracker). Meanwhile Bartz v. Anthropic settled at $1.5bn, roughly $3,000 per work, final approval July 2026 (Authors Guild). Litigation will not deliver a paycheque soon; voluntary settlement already has. The rational response to that pair of facts is to stop waiting for a court and start negotiating a price. That is the psychological opening.

#The complication money does not fix

This is the most important qualification on the entire recruiting thesis and it cuts against a purely money-based pitch. For a large minority the objection is not that they are underpaid. It is that a company profits from their work at all. A higher rate does not touch that.

What partially solves it is agency: the person chose the engagement, priced it, can decline it, and gets named. And the disclosure that would do the most work is the one nobody offers.

Not a single platform in this market names its client. Contra, Mercor, Outlier, Taste Labs and Handshake all sell "work for frontier AI labs" without saying which. For a designer deciding whether to participate, the identity of the ultimate beneficiary is the single most load-bearing fact — recall the 50.97%. Naming the client, even at the cost of a harder sales conversation, is the largest unclaimed trust differentiator in this market [WEAK — my judgement, not an observed finding; it conflicts directly with buyer-side NDAs, and the honest version may be tiered disclosure — "this brief is for a US-based image-model developer; you may decline" — rather than a name].

The precedent for how not to do it is fully documented. Adobe's Firefly contributor bonus produced, in Adobe's own forum, the line "Now I got $1 and I have 25000+ AI assets?"; by the 2025 round the thread shows a $1 minimum payment threshold and a contributor reporting "$4.88 added to existing balance" (Adobe community; 2025 thread). Shutterstock's Contributor Fund is better on exactly two dimensions — a stated 20% average corporate royalty rate and a real opt-out — and worse on one: it is pooled and periodic, so no contributor can predict or verify their own payment (Shutterstock). An opaque pooled bonus that pays $1–$5 does more damage than paying nothing, because it converts a grievance about theft into a grievance about insult. Pay a stated rate per unit; never let a payment round to a number that reads as contempt; honour the opt-out.

#Terms are the competitive ground, and one rival has left a gift

Sources: Contra terms per Contra Labs; Mercor legal docs; Background IP clause analysis; Outlier Terms of Use.

Mercor's Background IP clause is a competitive gift. A perpetual, irrevocable, sublicensable licence over pre-existing IP that a contractor "incorporates or uses" means a designer who brings their own type library, brush set, grid system or component kit to a task has arguably licensed it forever. For a population whose entire economic asset is a portfolio and a personal toolkit, that is a fatal term — and most applicants will never see it [WEAK on severity — this rests on a secondary analysis, the CIIA text being behind login]. Outlier's are worse still: irrevocable assignment plus an explicit moral-rights waiver plus zero attribution.

Contra's package — licence not assignment, author byline, no platform fee, an explicit right to decline, and seven-day payment — does not appear together anywhere else. Taste Labs matches on no-markup and no-exclusivity but publishes no IP terms and no rates at all. That combination is the recruiting proposition, and The clause that expires your corpus in year ten is where it gets drafted; Copyright is the weakest thing you own covers what can actually be owned.

One more asset from the Fast Company / AIGA pricing-transparency project, 1,426 responses fielded February–April 2026: 64% of freelance designers are not seeking salaried positions, only 12% are "very confident" pricing their work, and 49% sometimes fear asking for higher rates (Fast Company). Publish the rate card. It is cheaper than negotiating and it is a trust signal in a market where half the supply is afraid to ask.