Starting an AI chat
The AI chat is a peer of the guided form — a different way to interact with the same search engine. It’s also a research conversation surface for follow-up questions, “what about” pivots, and analysis of results.
In this guide:
- Open the chat
- Send a first message
- Recognize the two modes
- Run a search inline
Step 1: Open the chat
From the Product Research home, click Chat in the toggle row, or open it from the sidebar.
The chat is anchored as a chat panel — a familiar messaging UI with a composer at the bottom and a thread above it.
Step 2: Send a first message
Type whatever you want — a search request, a question about the SellerScore, a follow-up on a previous result. Claude figures out which.
Two examples:
A search request:
“Show me bamboo socks under $20 in the UK with low competition.”
A conversational question:
“How does the demand sub-score work when SFR isn’t available?”
Both are valid messages. Claude routes them differently.
Step 3: The two modes
Claude has exactly two tools available:
| Tool | When Claude picks it | Cost |
|---|---|---|
extract_search_params | The message describes a research goal. | 1 credit (for the triggered search). |
no_search_intent | The message is conversational, exploratory, or a question. | Free. |
You see the difference in the response:
- Search: A short ack (“Searching for…”) → search runs inline → results render in the chat → AI analysis of the top 5.
- Conversational: A streamed, plain-text answer.
→ Intent extraction for the full mechanics.
Step 4: Run a search inline
When Claude triggers a search, you see:
- Acknowledgement — “Searching for [extracted parameters]…”
search_triggeredevent — the parameters Claude extracted, displayed for transparency.search_completedevent — top 5 results render as cards inline in the chat.- Streamed AI analysis — Claude reads the results and offers a paragraph of synthesis.
The cards in the chat are tappable just like results in a list — open a detail, add to watchlist.
Continuing the conversation
After a search runs, you can refine without starting over:
- “Tighten that to 25%+ margin.” (re-runs with adjusted params; 1 credit)
- “Why did the top one score that way?” (conversational; free)
- “Add product 3 to my watchlist.” (executes the action; free)
The AI keeps context across the conversation.
When chat shines
- Exploratory phase — “what’s a good niche for a new seller in the UK in 2026?” — broad questions Claude can think through.
- Iterative refinement — easier to say “narrow to under $30” than to re-fill the form.
- Cross-question synthesis — “compare the top 3 in this search to the top 3 from yesterday.”
- Analysis on demand — “explain why the trend sub-score is dragging this product.”
When to fall back to guided
- Reproducibility — if you need exactly the same parameters every time.
- Power-user workflows — when filter/weight tuning matters more than chat-style flow.
- Faster — for repeated searches with known parameters, the guided form is fewer keystrokes.
The two modes use the same engine; pick whichever fits the moment.