← Study findings
Methodology

Methodology: Artist Statement Patterns Study

This descriptive study used a purposive quota protocol frozen before collection to code structural and language features in 100 English-language artist statements available online. It reports aggregate patterns only.

Research question and design

The question was frozen before collection: what structural and language patterns appear in how publicly available artist statements introduce a practice, name media and process, address an audience, state a purpose, and use a preregistered specialist vocabulary?

The study is descriptive. It cannot identify which writing choices cause success, quality, persuasion, sales, career outcomes, or audience response.

Sample

The final sample contains 100 current, general-practice English-language statements: 20 statements in each of five practice groups—painting and drawing; sculpture and object-based practice; photography and moving image; digital and multimedia practice; and installation and performance.

Selection was purposive rather than random. Artist-owned pages were preferred; the final source mix was 99 artist-owned sites and one gallery site. No artist contributed more than one statement.

Eligibility and collection

Eligible texts were public without login or payment, written in English, attributed to one identifiable artist, presented as a statement about the artist's overall practice, and between 75 to 750 words after navigation and boilerplate were removed.

Biographies, CVs, interviews, reviews, press releases, curatorial texts, anonymous or collective statements, exhibition-only texts, duplicates, and unresolved authorship or practice categories were excluded.

Coding fields

Each included statement was coded for word count, source type, primary practice, grammatical person, opening type, whether a medium was named, whether a process was described, whether an audience was addressed, whether a purpose was explicit, biographical leakage, and occurrences of a pre-specified 14-term specialist vocabulary.

Binary fields were marked present only when explicit. The opening was coded from the main clause of the first complete sentence. Biographical leakage was limited when it occupied one sentence or at most 20% of the words, and substantial above that threshold.

Quality control

A reproducible seeded selection fixed 20 of 100 rows for an AI consistency audit. The second pass used a separate cleaned-text packet without consulting the initial coded CSV during entry. It was nevertheless performed by the same AI system with retained project context, so it was not independent or blind validation.

The audit compared eight fields across 20 rows, or 160 code comparisons. Two disagreements were adjudicated against the unchanged codebook and applied to the final coded file. The post-adjudication comparison found zero unresolved differences.

Analysis and reproducibility

A deterministic script validated the 100 final rows, calculated category counts and denominators overall and for each 20-statement practice group, and calculated minimum, median, maximum, and mean word counts. The same input produces the same aggregate CSV and findings.

Percentages are descriptive frequencies, not population estimates. No significance tests, model-based estimates, or causal comparisons were performed.

Rights and privacy

Source URLs, artist names, full statement texts, access records, row-level codes, selected audit IDs, and adjudication notes remain private. The public CSV contains aggregate values only. No quotation or screenshot from the sampled statements is reproduced.

Limitations

Return to the findings