> ## 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 autocomplete curation

> Write a natural-language brand prompt that tells Layers which autocomplete suggestions to surface, how to label them, and which terms to suppress.

<Note>
  Autocomplete curation is currently in **beta**. You'll see a **Beta** badge on the settings page.
</Note>

## Overview

The **Autocomplete** settings page lets you control typeahead suggestions with a natural-language brand prompt. Layers uses the prompt — together with your store description — to decide which suggestions appear to customers and how they are labelled.

Curation updates asynchronously. You can save a prompt and keep working; suggestions usually update within a few minutes, depending on catalog size.

## Before you start

* Curation only affects the [Autocomplete API](/api-reference/predict). Search results are not filtered by your brand prompt.
* If both your store description and brand prompt are empty, no curation runs.
* Changing the prompt re-curates every eligible suggestion, not just new ones.
* New stores arrive with a starter prompt and brand description that Layers drafts automatically during [install](/help/onboarding/install#what-layers-seeds-automatically). Edit them to match your brand voice — Layers never overwrites text you have saved.

## Steps

1. Go to **Settings → Autocomplete** in the Layers dashboard.
2. In the **Brand Prompt** card, write the editorial rules you want Layers to apply. The prompt is limited to 2,000 characters.
3. Click **Save**. Layers confirms with *"Autocomplete prompt saved. Curation will run shortly."*
4. Wait for the **Curated** timestamp next to the prompt to update. The timestamp shows how long ago Layers finished the last curation run.

Suggestions update automatically once curation completes. You don't need to sync your catalog or restart anything.

## Preview a query

Use the **Preview** card to see curated suggestions for any query without waiting for curation to finish applying. The preview uses whatever prompt is currently in the **Brand Prompt** editor, so you can iterate on a draft without saving first.

1. Edit the **Brand Prompt** text (or leave it as-is to preview the saved prompt).
2. Type a partial query into the preview input (for example, `nike`).
3. Click the search icon or press **Enter**.
4. Review the suggestions Layers would return:
   * **Normal text** — the suggestion is allowed.
   * **Strikethrough** with a red *suppressed* label — the suggestion is hidden from customers.
   * **Label (was "original")** — the suggestion is shown with a curated display label.

Hover a suggestion to see the model's short reason for its decision. Reasons are only shown in preview — customers never see them.

Because the preview uses your unsaved draft, you can compare two versions of a prompt back-to-back: run the preview, edit the prompt, run it again. Nothing in the live experience changes until you click **Save**.

Previews are limited to 10 requests per minute per user per store.

## Re-curate after a large change

Saving the prompt automatically starts a new curation run. If a recent catalog change introduced new suggestions you want reviewed immediately, re-save the prompt even if you do not change the text. Layers also performs a full re-curation daily at 09:00 UTC for every store that has a prompt or description configured.

## Example prompts

### Brand casing and voice

```text theme={null}
Capitalize brand names exactly as they appear on our site:
Nike, Adidas, Lululemon, On. Use sentence case for every
other multi-word suggestion. Suppress suggestions that
reference brands we don't carry.
```

### Catalog alignment

```text theme={null}
This store only sells women's apparel. Suppress suggestions
for men's or kids' products, even if customers have searched
for them. Prefer singular nouns for the display label
(dress, not dresses).
```

### Compliance and claims

```text theme={null}
We're a supplement store. Do not allow suggestions that
make medical claims (cures, treats, heals). Keep category
suggestions like "vitamin C" or "magnesium" but suppress
anything that sounds like a diagnosis.
```

## Tips

* **Be specific.** Tell Layers the rule, not the goal. *"Capitalize Nike"* works better than *"Make brands look professional."*
* **Focus on editorial decisions.** The store description already tells Layers what you sell. Use the prompt for things only you know — tone, allowed claims, brand casing, seasonal guidance.
* **Iterate with the preview.** Save, preview a handful of real customer queries, refine, save again.
* **Don't duplicate typo handling.** Layers already filters misspellings and near-duplicates before curation runs. Your prompt doesn't need to cover them.

## Troubleshooting

**The Curated timestamp isn't updating.**
For large catalogs, the first run after a prompt change can take several minutes. If the timestamp still hasn't updated after 30 minutes, contact support.

**My prompt changes aren't reflected in preview.**
The preview uses whatever prompt text is in the **Brand Prompt** editor when you run it, including unsaved edits. If the preview still looks stale after changing the prompt, make sure you clicked the search icon or pressed **Enter** to re-run it. The live customer experience only updates after you click **Save** and curation finishes.

**Every suggestion shows as *suppressed* in the preview.**
If you leave the **Brand Prompt** blank and the model still chooses to suppress every suggestion, Layers falls back to showing all of them rather than hiding the entire autocomplete experience. To intentionally suppress suggestions, write a prompt that describes the rule you want applied — a blank prompt is treated as "no editorial guidance."

**Suggestions I want suppressed still appear.**
Check the preview first — if the suggestion is marked *suppressed* there, the live API is likely serving a cached response. Cached suggestions refresh within a few minutes. If the preview also shows the suggestion as allowed, rewrite your prompt to be more specific about the rule you want enforced.

**Preview returns "Too many preview requests."**
The preview is rate-limited to 10 requests per minute. Wait a minute and try again.

## Next steps

* [Autocomplete curation](/platform/autocomplete-curation) — How curation works end to end
* [Autocomplete API](/api-reference/predict) — The endpoint that serves curated suggestions
* [Semantic redirects](/platform/semantic-redirects) — Route specific search queries to landing pages
