> ## Documentation Index
> Fetch the complete documentation index at: https://docs.uselayers.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Configure search instructions

> Use the Layers dashboard to write store-wide merchandising guidance that attaches to every AI step in the search pipeline, with examples and tips.

<Note>
  Search instructions are currently in **beta**. You'll see a **Beta** badge on the settings page.
</Note>

## Overview

The **Search Instructions** settings page lets you write a single block of merchandising guidance in plain English. Layers attaches your instructions to every AI step in the search pipeline. The affected steps are query expansion, intent detection, facet value ordering, redirect approval, and result evaluation.

Changes take effect on the next search. You don't need to re-index your catalog or run anything manually.

## Before you start

* Search instructions affect AI steps in the search pipeline. They don't change deterministic ranking weights, request-time request transforms, or autocomplete curation.
* The field accepts free-form text up to 5,000 characters.
* Leaving the field empty disables the instructions block in every pipeline step.

## Steps

1. Go to **Settings → Search & Discovery → Search Instructions** in the Layers dashboard.
2. In the **Merchandising Instructions** card, write your guidance in plain English. Treat it like a short brief: one rule per line works well.
3. Click **Save**. Cached prompts for your store are invalidated, so the next search uses the new text.
4. Run a test search to confirm the guidance is having the effect you want. Adjust and re-save as needed.

## Example instructions

### Lifecycle and engagement

```text theme={null}
- Promote items with high 30-day purchase rates.
- Demote products in their first 30 days on sale — they don't
  yet have enough engagement data to rank confidently.
- Treat out-of-stock items as last-resort results.
```

### Seasonal direction

```text theme={null}
- When the query is generic (for example, just a category
  name), prefer the current seasonal collection.
- Outside of seasonal campaigns, surface evergreen
  bestsellers first.
```

### Catalog rules

```text theme={null}
- This store is kosher-certified. Never promote products
  tagged with pork or shellfish, even if their relevance
  scores are high.
- Prefer full-price items over clearance for the same query.
```

## Tips

* **Be specific.** Tell Layers the rule, not the goal. *"Demote items in their first 30 days"* works better than *"Don't show new products too much."*
* **Keep it broad.** Search instructions apply to every AI step in the pipeline. For one-off, query-specific behavior, use a [request transform](/platform/request-transforms) or [merchandising rule](/platform/merchandising) instead.
* **One rule per line.** Short, scannable rules give the model the clearest signal.
* **Iterate.** Save, run a few representative searches, refine, save again.

## Troubleshooting

**My instructions don't seem to be having any effect.**
Search instructions guide AI judgement; they don't override deterministic ranking. If your store is heavily weighted toward keyword signals, AI guidance has less room to move results. Check your [ranking relevancy](/platform/ranking-relevancy) signal weights, then re-test.

**A specific query isn't behaving the way my instructions describe.**
Search instructions are store-wide guidance, not per-query overrides. For query-specific behavior — pinning, hiding, or rewriting a particular query — use a [request transform](/platform/request-transforms), [merchandising rule](/platform/merchandising), or [semantic redirect](/platform/semantic-redirects) instead.

**I cleared the field but old behavior is still showing up.**
Save the empty field to invalidate the cached prompts. The next search request will run without an instructions block in any pipeline step.

## Next steps

* [Search instructions](/platform/search-instructions) — How the feature works end to end
* [Request transforms](/platform/request-transforms) — Request-time query and filter modifications
* [Ranking relevancy](/platform/ranking-relevancy) — Signal weights and ranking rules