SISTRIX published data showing AI-generated B2B texts score an average of 44.1 points on a 100-point readability and quality scale, with none of the three tested language models—Claude, Gemini, and ChatGPT—reaching the acceptable "green" threshold of 60 points. The findings appeared in SISTRIX's July 2026 monthly newsletter, sent on July 31, 2026, by Johannes Beus, founder and managing director of the Bonn-based search analytics company. The joint study with content-marketing firm Wortliga examined 2,112 AI-generated B2B texts. Claude scored highest at 47.7 points with the most consistent results, Gemini followed at 46.8 points but tended toward filler words, and ChatGPT trailed at 37.7 points. The study's more consequential finding concerned prompt design: scores for the same model swung between 1 and 96 points depending on how the prompt was written. A bureaucratically phrased instruction produced difficult-to-read text, while a simple instruction such as "write clearly" raised the readability component to 79.4 points but sacrificed depth of subject-matter detail. This reframes AI text quality as a solvable instruction-design problem rather than a fixed limitation of any single model. Further detail on prompt rules appears in a companion blog post on sistrix.de.