Deep Search

Deep Search is an automated research orchestrator that runs multiple searches with progressive refinement, converges when results stabilize, and returns deeper, more grounded answers. Unlike standard single-shot search, Deep Search keeps iterating to fill gaps until it finds your ideal products.

In this guide:

• Starting a Deep Search session
• Understanding the setup form
• Monitoring convergence during iterations
• Viewing and working with results
• Recovering stuck sessions
• How Deep Search costs work

Starting a Deep Search session

Deep Search is only available on paid Hilal Software plans (Professional, Business+, Enterprise). Navigate to Product Research → Deep Search and you’ll see the setup form.

The setup form

1. Research goal — Describe what you’re looking for at a higher level than standard search allows:

   • “Find three uncrowded sub-niches in pet care I could enter at $30+ margin.”
   • “Identify the trend in home fitness most likely to grow over the next 6 months.”
   • “Compare my current top SKU to its closest 5 competitors and tell me what they’re doing differently.”

2. Marketplace — Select US, UK, DE, FR, IT, ES, CA, or AU.

3. Category — (Optional) Narrow to a specific Amazon category.

4. Iteration Budget — Set how many search rounds Deep Search can run (1–10 iterations). More iterations = deeper results but higher cost. Typical sessions use 3–8 iterations.

Once you set the goal and click Launch Deep Search, the system:

1. Extracts your goal into a research plan with sub-questions.
2. Runs an initial broad search to map the territory.
3. Scores how well the current results answer your goal (Result Quality Score, or RQS).
4. Proposes the next search to fill gaps.
5. Iterates until results converge or you hit your iteration limit.

Monitoring convergence

As Deep Search runs, you’ll see a live streaming view showing each iteration as it completes:

• RQS sparkline — A chart showing how the Result Quality Score evolves across iterations. Flat or declining RQS means the system has found stable results and is ready to converge.
• Iteration details — What the system searched and what it learned from this round.
• Stop Search button — Stop early if you want; you’ll only be charged for iterations completed so far.

You can watch in real-time or close the page — the session continues in the background and you can check back anytime.

Converged results

Once Deep Search converges (RQS plateaus or you hit your iteration limit), you’ll see the best iteration shortlist — a curated list of top results from the iteration that performed best during the search.

From the converged results table, you can:

• Amazon links — Click product titles to jump directly to Amazon.
• Add to Watchlist — Save a result for ongoing tracking.
• Export — Download results as PDF or CSV (1 credit).
• New Search — Start a fresh Deep Search with new parameters.

You cannot edit or view per-product details from the converged view. To see full product analysis, click Amazon links to view on Amazon, or add to your Watchlist and open the result from the main Product Research page to view details.

Stopping or recovering a Deep Search

While searching:

Click the Stop Search button (top right of the iteration view) to end early. You’re only charged for iterations completed so far.

If your session times out or gets stuck:

• Refresh the page to reconnect and continue. Iterations already completed aren’t re-charged.

If you hit a concurrent-session error:

A new session will cancel the previous one. Go back to the Deep Search setup form and click Cancel session and start over to restart with new parameters.

How Deep Search costs work

Deep Search charges:

• 1 credit for the first iteration
• 0.5 credits for each additional iteration

Cost formula: 1 + (iterations − 1) × 0.5

Examples:

You set the iteration limit before starting, so you always know the maximum cost upfront. If you want to stop early, the cancel button refunds unused iterations.

Why iterative research is harder than single-shot

Deep Search has to solve problems that single-shot search doesn’t:

• Know when to stop. Convergence detection is the hard part — it must stop too early and miss opportunities or arrive too late and waste credits, but it balances both using RQS trends.
• Avoid redundant searches. The system avoids searching the same niche twice with slightly-different parameters.
• Handle unanswerable goals. Some goals don’t have a good answer; Deep Search must identify and signal this rather than spinning through iterations.
• Stay within your credit budget. The iteration limit is your hard cap.

Related articles