Example conversations

What marketers ask, and what the assistant does with the Eureka tools.

"How did agents and humans do last week?"

The assistant calls agents_vs_humans_summary with date_range: last_7_days and compare: previous_period, then answers with sessions by navigation mode, conversions and revenue for agents, humans and mixed sessions, and the top agents.

"Where do agents get stuck on my site?"

The assistant calls agent_friction with no steps. It answers with the blocking step first, for example "Delivery slot (/checkout/delivery-slot): 24% of agents get through vs 80% of humans; claude-in-chrome is most affected; fixing it is worth about +267 orders this month", then the suggested fixes (native controls instead of a custom picker, a default slot the agent can confirm). See Find where agents get stuck.

"Where do agents drop off in checkout?"

The assistant calls list_saved_funnels to find your checkout funnel, then run_funnel with breakdown: actor_family. It reports the step where agents fail far more than humans and the element involved, for example "24% of agents get past the delivery slot picker (delivery-slot-picker), against 90% of humans". Follow up with "show me the paths agents take after that step" (top_paths).

"Which pages does ChatGPT read, and does it send us traffic?"

pages_read_by_ai with bot_name: chatgpt-user, then run_report with dimensions: ["channel_group"] filtered on the AI Assistants channel. The answer lists the most-read pages and the visits and conversions coming back from ChatGPT.

"Is someone pretending to be Googlebot?"

pages_read_by_ai with bot_name: googlebot and verified: spoofed. The assistant reports how many hits came from outside Google's published IP ranges, and from which network.

"Show me mixed sessions where the agent completed the order"

mixed_sessions with converted_only: true. Each session comes back with its segments (human, then agent, sometimes human again) and what each one did.

"Anything unusual right now?"

realtime_snapshot: active agents and humans, top pages, agents currently on the site.

Anything else

For open questions, the assistant can call ask_analytics, which runs Eureka's own analyst on your data, or combine describe_catalog and run_report to build the table itself.