Original insights by Chris Long
30-second rundown
Key learnings:
- What changed: ChatGPT expanded an average prompt into 7.61 searches in Nectiv’s repeated test, up 250% from the earlier study.
- What works: Complex buying categories generated wider search chains, so one broad page may not answer the comparisons and verification questions AI creates.
- What to do: Map supporting pages and third-party evidence to likely follow-up searches so the brand survives more stages of retrieval.
This week: Run five important buyer questions, record every visible follow-up search using a tool like FanoutFox, group them into comparison, evidence, pricing, and verification needs, then improve the one page that misses the most repeated need.
Coming 3 months: Create an automated monthly fan-out capture and content-gap workflow across a stable buyer-question set. Give one owner responsibility for approving fixes and watch whether the same supporting pages begin appearing across more generated searches.
Chris Long and Nectiv reran roughly 4,000 prompts across software, ecommerce, travel, and other verticals to see how ChatGPT searches before it answers. The single query has become a nervous posse, kicking open seven or eight doors before breakfast and occasionally searching 29 alleys deep.
1. The Search Party Tripled
Average fan-out rose from 2.17 searches in the earlier study to 7.61, a 250% increase, while the longest observed chain grew from four searches to 29. Fifty-five percent of chains contained six to twelve searches. Fan-out means the extra queries an AI creates to gather evidence beyond the user’s original words.
2. Complex Purchases Trigger More Hunting
Software prompts averaged 10.7 searches, ahead of real estate at 8.7 and fashion at 8.5, while travel and credit-card prompts averaged fewer than six. The more variables the purchase requires, the wider the search tends to spread. A page can answer the original question and still miss the machine’s follow-up interrogations.
3. Authority Entered the Vocabulary
The study found longer queries and changing added terms, suggesting the system increasingly refines and verifies rather than merely repeating the prompt. Content must survive comparison, evidence, and source-checking queries. Publishing one broad page and praying is now the digital equivalent of mailing a compass into a sandstorm.
This is a repeated observational study, not a guarantee that every model or prompt behaves the same way. But the direction is loud: AI search is doing more searching, and every extra query is another checkpoint where weak evidence gets left beside the road.
