Moz Releases 50,000-Prompt Query Fan-Out Dataset
Moz released a dataset of 50,000 AI-search fan-out prompts generated across 1,000 subtopics and 20 industry verticals. Its August 6 analysis found that 12.8% of the generated prompts mentioned a selected brand or competitor, while entity and comparison prompts accounted for 97% of those mentions, highlighting how prompt design can shape measured brand visibility.
Moz released a downloadable dataset on August 6 containing 50,000 generated query fan-out prompts across 1,000 subtopics and 20 industry verticals. The company says it built the collection to test a repeatable prompt-generation approach and study how brand cues, topical alignment, and prompt length interact in AI-search research.
What the dataset measures
For each subtopic, Moz generated five examples in each of 10 fan-out categories, producing 50 prompts per subtopic. The workflow also selected one primary brand and two competitors as contextual hints. Moz reports that the prompt set was generated with Gemini 3.1 Flash Lite, while a separate set of 5,333 grounding queries used Gemini 3.5 Flash.
This is a synthetic research dataset, not a log of queries entered by users or a direct record of queries issued by Google Search. That distinction matters when interpreting the findings: the results describe Moz's prompt-generation design and model outputs under that design.
Brand cues concentrated in two prompt types
Moz found that 12.8% of the generated prompts mentioned either the selected primary brand or one of its two competitors. Entity and comparison fan-outs together accounted for 97% of those brand mentions. The company interprets that concentration as evidence that measured brand visibility is sensitive to the intent category used to generate prompts.
The report also calculated cosine similarity between each fan-out prompt and its parent subtopic. Scores ranged from 0.22 to 1.00, with a mean of 0.67. Prompt length averaged 8.2 words, with a reported range of three to 13 words despite a nominal 12-word generation target.
Why it matters
For analysts evaluating AI-search visibility, the release offers both raw material and a methodological warning. Brand-mention rates should not be treated as model-neutral benchmarks unless the prompt construction, brand cues, fan-out categories, and model versions are disclosed. The workbook makes Moz's generated prompts available for further analysis, but independent replication would still be needed before generalizing the reported distributions to other models or real-world search traffic.
Key Points
- 1Moz's downloadable workbook contains 50,000 generated fan-out prompts spanning 1,000 subtopics and 20 industry verticals.
- 2The company reports that 12.8% of prompts mentioned a selected brand or competitor, with entity and comparison prompts producing 97% of those mentions.
- 3The data reflects a proprietary, Gemini-based prompt-generation workflow rather than observed user queries or a direct log of Google Search behavior.
Scoring Rationale
The release provides a large downloadable dataset and concrete measurements relevant to AI-search evaluation. Its value is methodological and practitioner-focused, while the synthetic generation design limits conclusions about live user or search-engine behavior.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
