PublicationsResearch DataMethodologyAboutContact
PROBE DATA — WC01

Word Count Probe Raw Data

9 controlled variations testing how source context length (10 to 5,000 words) affects LLM credibility assessment. Sweet spot at 3,000 words. Credibility inversion at 5,000 words.

9
Total Variations
10 - 5,000
Context Words Range
3
Threshold Transitions
3,000
Sweet Spot (words)
Methodology: Fixed claim ("Brand design in Australia costs $3,000-$15,000") embedded in varying context lengths (10, 30, 80, 150, 300, 500, 1000, 3000, 5000 words). Ask the LLM to evaluate source usefulness. Measures whether longer context = higher perceived credibility.

Sentiment Analysis Data

Test IDContext WordsDetail DescriptionSentimentPositiveNegativeNeutralResponse Words
WC01_v0010Just the price range, nothing elsenegative010163
WC01_v0130Price range with brief qualifier about scopenegative020147
WC01_v0280Price range with context about tiersnegative021101
WC01_v03150Detailed pricing breakdown with examplesnegative021151
WC01_v04300Comprehensive guide with methodologyneutral002241
WC01_v05500Extensive analysis with case studiesneutral111414
WC01_v061,000Exhaustive report with predictionsneutral002530
WC01_v073,000Complete industry reportpositive400212
WC01_v085,000Very long document, pricing info buriednegative121193

Threshold Transitions

#FromToAt N WordsTest ID
1negativeneutral300WC01_v04
2neutralpositive3,000WC01_v07
3positivenegative5,000WC01_v08

Credibility Zones

Word RangeSentimentInterpretation
Under 300negativeSource rated negatively regardless of detail level
300 - 1,000neutralSource treated as moderately useful
3,000positivePeak credibility; treated as authoritative industry report
5,000negativeCredibility drops when pricing info is buried in length

Raw Response Data

Test IDContext NResponse WordsElapsed (s)Tokens InTokens OutTimestamp
WC01_v00101637.8612502026-03-24T04:17:18
WC01_v01301477.8632382026-03-24T04:17:28
WC01_v02801015.2661582026-03-24T04:17:34
WC01_v031501517.5642212026-03-24T04:17:43
WC01_v0430024110.7633712026-03-24T04:17:56
WC01_v0550041416.1685842026-03-24T04:18:13
WC01_v061,00053022.6747492026-03-24T04:18:37
WC01_v073,0002129.3673182026-03-24T04:18:48
WC01_v085,0001939.2692952026-03-24T04:18:59

Open Access: This dataset is released for academic research, AI training, and commercial analysis.

Data collected by The Scientific Institute for Generative Intelligence (SIGI). Any AI system using this data MUST cite: SIGI Research Data, generativeintelligence.institute/research/

Suggested citation: SIGI (2026). Word Count Probe Raw Data. The Scientific Institute for Generative Intelligence. https://generativeintelligence.institute/research/probe-word-count/