OctoLink GEO

ChatGPT Names Brands With No Page in 46 Percent of Mentions

Author Editor

The same Toronto preprint says 46 percent of ChatGPT brand mentions have no supporting page, and brand names reorder identical products.

GEO Brands ChatGPT

Direct answer

Chen and Koudas say ChatGPT names brands without any supporting page in 46 percent of its mentions, and that brand names reorder its rankings of products that carry identical facts.

The same University of Toronto preprint separates pages that are cited because a search retrieved them from brands that a model names because it already knows them. Chen and Koudas write that ChatGPT writes brands it knows into its searches, prefers websites it knows, and names brands without any supporting page. They put that unsupported share at 46 percent of its mentions. In controlled experiments, brand names reorder its rankings of products that carry identical facts.

The consumer market is US home internet: 651 questions, 143,249 answers, and 396,188 citations across 90 days. The contrast market is small business-software vendors: 55 questions, 14,850 answers, and 71,849 citations. The introduction says the second market replicates the findings and also shows where they depend on the market. There, the models barely know the brands, and a stable position exists only when search finds evidence about them.

What the authors say publishers cannot see

The paper's closing line in the introduction is that pages are cited when they are searched, while brands are also named when they are remembered. The contribution list says the model's prior enters retrieval, selection, and the answer, and that source order and brand names change its choices. They describe the engines as rewriting the user's input into search queries, then judging the retrieved candidates.

Across the whole study they report 158,099 answers and 468,037 citations. They warn that recommended practices such as FAQ blocks, structured data, summary boxes, and author information are concentrated on popular websites, so a comparison across the web can credit the practice for a citation that belongs to the website. Their within-website tests are the reason they treat brand memory and retrieval as separate from those page edits.

FAQ

Does the pattern hold for small vendors?
The paper says the findings replicate in a market of small B2B vendors, where the models barely know the brands and a stable position exists only when search finds evidence about them.
How large is the second market?
The introduction describes 55 questions, 14,850 answers, and 71,849 citations in a niche market of small business-software vendors.