SHOMEDIAHow a user phrases a question can materially affect the sentiment ChatGPT expresses about a brand. A study published by SE Ranking on November 4 found that negative phrasing moved sentiment roughly twice as far below a neutral baseline as positive phrasing moved it above that baseline. That matters for companies benchmarking how AI tools describe their products.
The researchers tested 180 prompts, split evenly between positive, neutral and negative framing. Each prompt was run five times, producing 900 responses. The cases covered Samsung and Sony TVs, Hers and Start Willow telehealth services, and Delta SkyMiles and United MileagePlus, with questions about recommendations, comparisons, trust and value.
SE Ranking reported average Net Sentiment scores of −36 for negative prompts, +20 for neutral questions and +46 for positive ones. The Spearman correlation between prompt and response sentiment was 0.66. These results measure response text within a defined sample; they do not establish a causal effect on consumer purchases or advertising performance.
Repeatability was limited. Only 44% of prompts produced the same sentiment classification across all five runs. For 96%, at least three of five answers agreed on the majority sentiment. A single screenshot of an AI answer is therefore a weak basis for declaring a company's reputation improved or deteriorated.
For SEO and communications teams, credible monitoring should include neutral and adversarial questions about refunds, reliability, safety and alternatives. Repeat the prompts, preserve the exact phrasing, and segment results by industry, language and competitors. This is a measurement method, not a guaranteed way to manipulate an AI model.
Search Engine Land's independent summary notes that trust-related questions behaved differently: positive framing often failed to raise sentiment much, while negative framing still reduced it. Applying the result to other markets requires new tests.
Комментарии