Nobody on a jury is running your application through a detector. That is worth saying first, because the anxiety around AI writing in open calls has attached itself to the wrong fear. Detection software is unreliable enough that most institutions have quietly stopped relying on it, and it penalises non-native English writers at a rate that makes its use indefensible. What actually happens is simpler and harder to game: a panel reads four hundred applications in two weeks, develops an extremely fast sense of which ones say something, and your text either survives that filter or it does not.
The useful question, then, is not will they catch me. It is what does assisted writing tend to do to a text, and does that help or hurt in a room where someone is reading your statement for ninety seconds.
What a jury is actually doing
It helps to picture the mechanics. A residency jury with 400 applications and five readers typically works in two passes. The first pass is a sort: two or three minutes per application, images first, then a skim of the statement and project description. Most applications are eliminated here, and they are eliminated not because they are bad but because nothing in them gave the reader a reason to put them in the second pile.
The second pass is where text matters. Here the panel is reading closely, often aloud, and looking for three things: whether the project is specific enough to actually happen, whether this applicant needs this residency rather than any residency, and whether the writing and the images describe the same practice. That last one is where assisted text most often fails, and it fails without anyone naming AI as the reason.
The four things that get noticed
Fluent emptiness. This is the signature failure, and it predates AI by decades — press releases and MFA statements have produced it for years. Assisted drafting makes it much easier to generate at volume. The text is grammatical, well-paced, confident, and contains no information. A reader finishes the paragraph and cannot say what the applicant makes.
Juries respond to this with a specific irritation, because fluent emptiness is more frustrating than clumsy specificity. A panel can work with a badly written sentence that tells them you cast concrete in a disused swimming pool in Tbilisi. It cannot work with a beautiful sentence about interrogating liminal spaces.
Statement–image mismatch. When the text is generated from a short prompt, it tends to describe a generic version of the practice rather than the actual one. The images then show something more awkward, more specific, and often more interesting than the text promised. Jurors notice this as a credibility problem: either the applicant does not understand their own work, or someone else wrote the description. Both readings are damaging.
Register that does not match the rest of the file. Applications have several text fields and usually a CV, sometimes a short email. When the project description is in polished institutional English and the answer to “why this residency” is in plainer, more hesitant prose, the discontinuity is audible — especially when a panel reads aloud. The honest version is often the better one, which makes this doubly unfortunate.
Vocabulary at saturation. Certain words have become so overrepresented in generated prose that they now function as a tell, not because any one of them is wrong but because they cluster: delve, tapestry, testament, nuanced, intricate, multifaceted, resonate, underscore, landscape (used metaphorically), in today’s rapidly evolving. Paired with tricolons and em-dash parentheticals at a density no human sustains, the effect is recognisable. This list will date — the vocabulary shifts — but the underlying principle will not: any phrasing that arrives without being chosen reads as unchosen.
A worked example
What follows is an illustrative pair — not a real submission, but constructed from the patterns that recur in open-call text. Read both as a juror on application 212 of 400.
Version A: assisted draft, unedited
My practice delves into the intricate relationship between memory, material, and place. Working across sculpture and installation, I interrogate how objects carry the residue of lived experience, creating immersive environments that invite viewers to reconsider their relationship to the everyday.
Drawing on a multifaceted approach that bridges personal narrative and broader socio-political concerns, my work explores themes of displacement, belonging, and the liminal spaces between presence and absence. Through a nuanced engagement with found materials, I seek to question dominant narratives and open up new possibilities for understanding.
This residency would provide an invaluable opportunity to develop my practice in dialogue with a vibrant international community of artists. The unique resources and critical framework offered by your programme resonate deeply with my current research trajectory, and I am confident that this experience would be transformative for my work.
Read that as a juror. You cannot say what she makes, how big it is, what it is made of, or what she would do in the studio. The third paragraph could be sent to any residency on earth — which means it tells the panel nothing about why she chose theirs. The prose is competent. That is the whole problem: nothing survives the reading.
Version B: same artist, same facts, written out
I make sculpture from domestic materials salvaged from houses before demolition — floorboards, kitchen tile, curtain fabric, the brass fittings from doors. My family left Shiraz in 1998 and I grew up with the objects that came in two suitcases, so I am interested in which parts of a house end up portable and which get left standing.
The works are usually floor-based and modest in scale, between 40 cm and a metre, and they look like furniture that has lost its function. I joint them badly on purpose. Over the last two years I have been casting the gaps instead of the objects — pouring plaster into the spaces between salvaged pieces, so the sculpture is the shape of an absence rather than a thing.
I am applying here because of the ceramics kiln and because the programme runs for four months rather than four weeks. The plaster casts keep failing at scale and I think the answer is to fire them, which I have never done and cannot do in my current studio. Four months is roughly how long it would take to learn badly enough to make the first ten.
Same artist, same themes — memory, displacement, material. But a juror finishes Version B knowing the scale, the material, the method, the biographical fact that grounds it without explaining it away, and exactly what the applicant would do with the residency. “I joint them badly on purpose” and “learn badly enough to make the first ten” are the two lines a panel would quote to each other, and they are the kind of sentence no assistant produces unprompted, because they come from knowing the work from the inside.
Note also what Version B does not do. It does not claim significance. It does not use the word interrogate. It is not more beautifully written than Version A — arguably it is plainer. It is simply full of information, and that is the entire difference.
How to use the tools without producing Version A
The tools are genuinely useful in this context. The question is where in the process they sit.
Use them after you have the content, not to produce it. Write your material badly first, in your own words, in whatever language comes easiest. Then use the tool to tighten, cut, restructure, or translate. The order matters more than anything else in this article: a tool given facts improves them, and a tool given a prompt invents a generic practice.
Use them to interrogate rather than to generate. Paste your own draft and ask what is unclear, what a sceptical juror would not believe, which sentences could describe any artist, what is missing. This is where assistance is most valuable and least detectable, because the output is a list of problems and the fixes are yours.
Use them on the structural work. Cutting 900 words to 500 is a genuinely tedious task that tools do well. So is reformatting a CV, checking a translation, and confirming you have answered every part of a three-part question — a surprisingly common reason for rejection.
Translate, then de-smooth. For non-native English writers this is the real benefit and it comes with a specific risk: machine translation tends to iron out exactly the idiosyncratic phrasing that makes a statement memorable. Translate, then go back through and restore the odd constructions that are actually yours. A slight accent in the prose is not a flaw in an international open call. Perfect neutral English is more suspicious than imperfect specific English, and considerably less interesting.
Never let it write the “why us” answer. This field is the single highest-signal part of most applications and the one generated text handles worst, because the honest answer depends on facts a tool does not have: which technician, which kiln, which four months, which conversation you want to have. If you cannot answer it from your own knowledge, you may be applying to the wrong programme.
Disclosure, briefly
Policies are arriving unevenly. Some funders now ask applicants to declare AI use in the preparation of materials; most do not yet. Read the guidelines and answer honestly if asked — undeclared use in breach of a stated rule is a withdrawal risk, and it is not worth it.
Where no policy exists, the ethical line most panels would recognise is between assistance and substitution. A tool that helped you express your own practice is a spellchecker’s descendant. A statement describing a practice you do not have is a misrepresentation, and the place it falls apart is not the written application at all — it is the interview, the studio visit, or the first week of the residency, when someone asks you to talk about the work and you have to be the person who wrote that paragraph.
The test
Before you submit, run one check. Delete every sentence that could appear in another artist’s application without alteration.
If more than a third of your text goes, you have a Version A problem, whatever produced it. What remains is the part a jury would have read anyway — and it is usually the part you were most tempted to cut for being too plain, too small, or too specific to matter.
It is the only part that matters.
