The New Shape of Search: How Conversational AI Recomposes Information Seeking

7 hours ago

Conversational AI is often expected to initiate search journeys, but observed data shows the opposite: content typically follows search and more often precedes AI assistant use.
Search anchors the front of the journey, while assistants sit deeper within it, with a paired difference in direction of +20.6 percentage points between the two anchors.
Assistant sessions are more often self-contained: 34.1% show no external web step, compared to 19.5% for search-centered sessions, a within-user contrast of +13.0 percentage points.
Other assistant sessions are categorized as AI-first (10.5%), AI-last (18.3%), or bridge/interleaved (37.1%), illustrating recomposition of activity across dialogue, search, and browsing.
The study introduces a cross-surface topology for the search journey and distinguishes observed containment from inferred resolution, noting that self-contained does not imply task completion.

AI visibility tools promise to measure brand visibility in AI answers but often provide false precision by presenting tidy claims like mention rates and rankings.
Scraping the frontend of AI products like ChatGPT or Claude captures only one synthetic session with many uncontrolled variables, leading to biased measurements.
Even with identical prompts, AI systems can produce varying answers due to factors like model batching, personalization, and nondeterministic behavior.

CrucibleBench evaluates language models in a persistent MUD text world over 50 turns with hidden social objectives.
It uses 'lateral thinking with withered technology'—applying mature, inexpensive tech (MUDs) to measure AI behavior.
Key constraints include an enumerable action space, explicit social feedback from NPCs with trust/suspicion states, and within-run persistence.

The author's AI token usage increased from $600/month to $25K/month after joining Lovable, with an overall spend of $85K since January.
Development evolved from a solo dev with few agents to a system with a dedicated agent managing swarms of subagents, boosting merged PRs from 20-30 to 150+ per week.
Human review is now reserved for high-impact changes like RFCs and ADRs, not traditional code review, shifting focus to system design and infrastructure decisions.

ChatGPT uses different data pipelines: serp for open web, labrador for licensed publishers, bright for Bright Data scraper, oxylabs for Oxylabs scraper.
Some queries (like 'text' type) skip web searches and use training data, affecting SEO opportunities.
The 'fan-out' queries reveal ChatGPT runs multiple sub-queries, including site-specific probes, for complex tasks.

94.8% of audited websites are never named in AI assistant answers to buying-intent questions.
Only 1.9% of 5,978 assistant answers across ChatGPT, Claude, Gemini, and Perplexity mentioned the business being audited.
Gemini had the highest mention rate at 2.9%, while 8.9% of sites block at least one AI crawler, mainly training crawlers.

AI visibility tools promise to measure brand visibility in AI answers but often provide false precision by presenting tidy claims like mention rates and rankings.
Scraping the frontend of AI products like ChatGPT or Claude captures only one synthetic session with many uncontrolled variables, leading to biased measurements.
Even with identical prompts, AI systems can produce varying answers due to factors like model batching, personalization, and nondeterministic behavior.

CrucibleBench evaluates language models in a persistent MUD text world over 50 turns with hidden social objectives.
It uses 'lateral thinking with withered technology'—applying mature, inexpensive tech (MUDs) to measure AI behavior.
Key constraints include an enumerable action space, explicit social feedback from NPCs with trust/suspicion states, and within-run persistence.

The author's AI token usage increased from $600/month to $25K/month after joining Lovable, with an overall spend of $85K since January.
Development evolved from a solo dev with few agents to a system with a dedicated agent managing swarms of subagents, boosting merged PRs from 20-30 to 150+ per week.
Human review is now reserved for high-impact changes like RFCs and ADRs, not traditional code review, shifting focus to system design and infrastructure decisions.

ChatGPT uses different data pipelines: serp for open web, labrador for licensed publishers, bright for Bright Data scraper, oxylabs for Oxylabs scraper.
Some queries (like 'text' type) skip web searches and use training data, affecting SEO opportunities.
The 'fan-out' queries reveal ChatGPT runs multiple sub-queries, including site-specific probes, for complex tasks.

94.8% of audited websites are never named in AI assistant answers to buying-intent questions.
Only 1.9% of 5,978 assistant answers across ChatGPT, Claude, Gemini, and Perplexity mentioned the business being audited.
Gemini had the highest mention rate at 2.9%, while 8.9% of sites block at least one AI crawler, mainly training crawlers.

Hasty Briefsbeta