Language models in production do not write prose; they fill forms: JSON fields, function arguments, and extraction templates. We show that the form itself causes hallucination. We ask thirteen models the same question about the same input and change only the answer format.
The inputs are built so the question cannot be answered: a viral post showing 12,400 likes but no visible replies, a support ticket whose call was never transcribed. In free text, GPT-5.5 answers honestly, stating there is no reply data 98% of the time. However, given a required JSON field for sentiment, the same model invents an answer 40 times out of 40, fabricating the mood of crowds it never saw and quoting customers it never heard.
This pattern holds true across ten of thirteen models, with required fields driving fabrication to 100%. An explicit "insufficient evidence" option rescues only edge cases: all nine open-weight models ignore it. A direct instruction, do not infer sentiment, is overridden by the schema in four of six models.
Honesty under format pressure is a training outcome that remains unmeasured. Fabrication hides in areas where hedging is impossible: in required enums and minimum-count arrays, fields where no disclaimer fits. We release PhantomFill, a benchmark with deterministic scoring and two reportable numbers: the Coerced Fabrication Rate and the Escape Utilization Rate. The fix we test is one line of schema, while the failure we measure is pervasive.
Blogger's Review: This study highlights the vulnerabilities of language models under specific formatting pressures, particularly in mandatory fields, leading to a tendency to generate false information. Future work should focus on enhancing model honesty when faced with format constraints to prevent misinformation from spreading.