Seer Interactive on How to Create Context Examples in ChatGPT to Ge... [Video Analysis] | Spotlight | Spotlight - Ekamoira

Seer Interactive on How to Create Context Examples in ChatGPT to Ge... [Video Analysis] | Spotlight | Spotlight - Ekamoira

Summary

AI-generated overview

Seer Interactive demonstrates how generating training data within ChatGPT significantly improves categorization accuracy for SEO competitive analysis. The test case: categorizing 150 competitors as aggregator vs. non-aggregator sites. Without training data, ChatGPT identified 16 aggregators (10.7%). After first asking ChatGPT to generate a list of 100 example aggregator sites with definitions ("includes listicle content, like top 10 payroll companies or best internet providers"), then providing the same 150-competitor list, it identified 32 aggregators (21.3%)—doubling the detection rate. The training data approach creates a "context window" priming the model with relevant examples before the actual task. Validation remains critical: manual spot-checking confirmed Backpacker.com (flagged in the trained version) publishes gear guides and listicles characteristic of aggregators, which the baseline version missed. The method applies to other categorization tasks like thematic keyword analysis, organic ranking term grouping, and qualitative research where providing examples upfront yields "more organized and efficient" outputs.

Key Insights

8 insights

1
Without training data, ChatGPT identified 16 aggregator sites from 150 competitors; after generating 100 example aggregators as training data, it identified 32—more than doubling detection accuracy from 10.7% to 21.3%.
2
Training data is created by asking ChatGPT itself: "Create a list of 100 aggregator websites" with a definition ("listicle content, like top 10 payroll companies") in a table format with domains.
3
The training data approach "creates that context window of aggregator and affiliate websites, so it's already starting to think about some of this information or generate some of this information" before the actual categorization task.
4
Manual validation revealed the trained model correctly identified Backpacker.com as an aggregator due to its "Winter Gear Guide" and other listicle content, which the baseline model missed entirely.