> ## 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.

# Tune signal weights

> Adjust how much each ranking signal group — text relevance, popularity, recency, and personalization — contributes to your store's search result ordering.

## Before you start

* Signal weights apply globally to all text searches for your store.
* Changes take effect immediately after saving.
* Use the search preview on the same page to test adjustments before saving.

## Steps

1. Go to **Ranking Relevancy** in the main navigation.
2. On the **Signal Weights** tab, review the **Weight Distribution** donut chart and legend showing the current percentage assigned to each of the five signal groups.
3. Adjust a signal group's contribution in either of two ways:
   * Drag the signal group's **slider** to move the percentage up or down.
   * Click the percentage field next to the slider and type an exact value between 1 and 80, then press **Enter** or click outside the field to commit the change.
     The other groups rebalance automatically so the total stays at 100%.
4. To fine-tune sub-signals, click the chevron next to a signal group to expand it:
   * **Semantic** — Adjust the balance between text similarity and image similarity.
   * **Engagement** — Adjust the relative importance of product views, sales, cart sessions, and total sales. These are powered by your store's [metrics](/platform/metrics).
   * **Keyword** — View the per-field weights derived from your attribute configuration (read-only).
5. Hover the help icon next to any signal group name to see a short description of what that signal measures.
6. Below the signal weights, adjust the **AI Reranking** dial if you want to change how broad, top-of-funnel queries (such as "shoes" or "gifts") are reordered:

   * **AI Reranking** — How forcefully Layers promotes products judged as strong matches for the query. Past 50%, it also demotes products judged as wrong for the query.

   The dial applies only to broad queries — specific and long-tail searches are unaffected. To spread a broad results page across categories instead of letting one category dominate, configure a [diversity expression](/platform/merchandising/diversity) in a sort order or merchandising rule; diversity on live results comes only from merchant-configured rules. See [AI reranking for broad queries](/platform/ranking-relevancy#ai-reranking-for-broad-queries) for details.
7. Use the **Search Preview** panel on the right to type a query and see how products rank with your current (unsaved) settings.
8. Click **Save** when you are satisfied with the results.

## Reset to defaults

If your adjustments produce unexpected results, click **Reset** above the donut chart to restore the original weight distribution. The button is only enabled when your weights differ from the defaults.

## Image search configuration

Click the **Image Config** button to open image search settings where you can configure which product images Layers uses for visual search and set minimum image quality thresholds (width and height).

By default, new stores use **both** the product featured image and variant featured images for visual search embeddings. You can change this to use only product images, only variant images, or all images. See [configure search behavior](/help/configuration/configure-search-behavior#image-search-configuration) for all available options.

## Tips

* Start with small adjustments — moving a slider by 5–10%, or entering a close value in the percentage field, and previewing the effect is more reliable than large changes.
* If search results feel too keyword-literal, try increasing the semantic weight.
* If popular products aren't surfacing for relevant queries, try increasing the engagement weight.
* If recently added products need more visibility, increase the freshness weight.
* Avoid pushing any single signal group above 70%. Over-concentrating weight on one group can reduce result diversity and make rankings less responsive to other quality signals.

## Troubleshooting

* **The percentage field snaps to a different number after I type.** Each signal group is capped between 1% and 80%. Values outside that range are clamped to the nearest allowed value. The other groups also rebalance to keep the total at 100%, which can nudge values by a percentage point or two.
* **I can't click Reset.** The **Reset** button is disabled when the current weights already match the defaults. It becomes active as soon as you change any slider or sub-signal.
* **The donut chart and legend don't match what I see in live search.** Changes are local until you click **Save**. Switch back to the signal weights view and save to apply the new weights to live searches.

## Next steps

* [Ranking relevancy](/platform/ranking-relevancy) — How signal groups and ranking rules work
* [Create a ranking rule](/help/ranking-relevancy/create-ranking-rule)
* [Test text search](/help/evaluate/test-text-search)