Search relevance and ranking
Relevance means showing products that best answer the shopper’s current intent. Ranking is the order of those eligible products. VIBE combines several signals, then applies the merchant’s controls, filters, availability choice, and rules.
Use this guide when a product is present but appears in the wrong position, or when you need to decide whether to update Shopify data, the search profile, a semantic control, or a merchandising rule.
Eligibility comes before ranking
Section titled “Eligibility comes before ranking”VIBE can rank only records that are eligible and indexed. Before investigating scores, confirm that the product:
- Is Active in Shopify.
- Is published to the Online Store.
- Is not hidden from search engines.
- Does not match a VIBE product, collection, or tag exclusion.
- Is handled as expected by the out-of-stock setting.
- Finished its latest index update.
If the product never appears for its exact title, start with Sync and Index, not a boost or pin rule.
The main relevance signals
Section titled “The main relevance signals”Keyword matching
Section titled “Keyword matching”Keyword matching rewards exact or close text relationships between the query and indexed product data. It is especially important for:
- Product names.
- Brands and vendors.
- Product types.
- Model numbers.
- Material names.
- Variant options.
- Specific collection or category terms.
Use clear Shopify titles and descriptions for literal searches. A synonym helps when shoppers consistently use a different term for the same thing.
Semantic matching
Section titled “Semantic matching”Semantic matching compares meaning, not only identical words. It helps with intent queries such as:
light jacket for raingift for a new apartmentminimal dining chairwarm neutral artwork
Semantic matching is only as useful as the product information VIBE indexes. If many intent searches are weak, inspect the active search profile and source data before creating many individual rules.
Structured product data
Section titled “Structured product data”Product type, vendor, collections, tags, options, availability, price, and other structured values provide factual context.
Structured data supports relevance, but exact storefront filtering is different. A semantic query can prefer blue products; a color filter should restrict the result to records explicitly carrying the selected filter value.
Variants
Section titled “Variants”When variant search is enabled, VIBE can index variants separately and match exact sizes, colors, styles, or other options.
Variant search is useful when shoppers search for a precise combination and the product-level record is too broad. It can also increase index work and requires the plan shown in the app.
Images
Section titled “Images”When image matching is enabled, product and optional context photos can contribute to visual similarity.
- Product photos should clearly represent the item.
- Context photos can add environment, styling, model, room, or detail information.
- Variants use their own image when Shopify supplies one.
Image matching helps visual and style intent. It should not replace accurate text and structured data for factual product attributes.
Request-time modifiers
Section titled “Request-time modifiers”Semantic controls
Section titled “Semantic controls”Controls let the shopper express a preference at request time.
- A slider balances two concepts.
- A toggle chooses between two concepts.
- A single control strengthens one concept.
Controls modify the current ranking. They do not rewrite the product index or permanently change relevance for every shopper.
Starting collection
Section titled “Starting collection”A starting collection scopes or guides Explore around a selected catalog group. It is useful when the shopper starts with a department or theme before adjusting preferences.
Shopify filters
Section titled “Shopify filters”Filters apply exact constraints. If a shopper selects size M, products without that indexed filter value are removed from the eligible set.
Use filters for facts. Do not create a semantic control to imitate inventory, price, or exact option filtering.
Out-of-stock handling
Section titled “Out-of-stock handling”The Configuration tab provides three approaches:
- Hide removes sold-out products from shopper results.
- Show last keeps them discoverable but moves them behind in-stock items.
- Mix in allows normal relevance to position them.
This setting can explain why a semantically strong sold-out item is absent or below weaker in-stock results.
How merchandising changes natural relevance
Section titled “How merchandising changes natural relevance”Use Boost when a product or group is relevant but should receive additional business priority.
Good examples:
- Favor a new collection for a broad category query.
- Give higher visibility to an owned brand.
- Support a seasonal campaign while keeping other relevant results.
Do not use Boost to make an unrelated product appear relevant.
Use Bury when a product is technically relevant but should appear later.
Good examples:
- De-emphasize low-margin alternatives.
- Move clearance items down without hiding them.
- Keep an older model searchable behind a newer model.
Use Pin when a specific product must occupy a fixed position for a matching query.
Pin is the strongest placement decision. Use it sparingly because it can displace products that would otherwise rank naturally.
Behavioral ranking
Section titled “Behavioral ranking”Behavioral ranking uses query-specific shopper response to reorder close matches. It is most useful when several products are similarly relevant and real engagement provides an additional signal.
It should refine close candidates, not compensate for missing catalog data.
Synonym
Section titled “Synonym”Use a synonym when different words should be treated as equivalent.
Examples:
- trainers, sneakers
- sofa, couch
- handbag, purse
Do not create synonyms for merely related concepts. Treating dress and jacket as equivalent would reduce precision.
Redirect
Section titled “Redirect”Use a redirect when the best answer is a destination rather than a product result list.
Examples:
returnsto a returns-policy page.summer saleto a campaign collection.gift cardsto the gift-card page.
Redirects bypass normal result ranking for the matching search.
Rule priority and conflicts
Section titled “Rule priority and conflicts”More than one rule can match the same query. Priority determines which decisions apply first or which rule wins where actions conflict.
When debugging:
- Filter the Merchandising table by the relevant rule type.
- Search for the query or trigger.
- Check whether each rule is active and within its schedule.
- Review target products, groups, and position.
- Review priority and reorder if necessary.
- Test the exact query again.
Avoid creating a second rule until you understand the first matching rule.
How to read Search Preview
Section titled “How to read Search Preview”Search Preview separates the overall result score from semantic similarity.
- A high overall score can reflect a strong combination of text, meaning, and other ranking signals.
- A high semantic score means the product is close to the query in meaning.
- A product can have a good semantic score but rank lower because another product has stronger combined evidence.
- A product can match literal words but still be a poor shopper answer if its source data is misleading.
The selected-product panel exposes indexed data and sync state. Use it to verify the input VIBE actually received instead of assuming the current Shopify page has already propagated.
Choose the smallest correct fix
Section titled “Choose the smallest correct fix”| Symptom | Best first action |
|---|---|
| Product never appears for its exact name | Check eligibility and sync |
| Many intent queries are weak | Review source data and search profile |
| Shoppers use another word for the same product | Add a synonym |
| One campaign product needs more visibility | Add a scoped boost or pin |
| A relevant product should remain searchable but lower | Add a bury rule |
| A query should open a collection or information page | Add a redirect |
| Explore preference has little effect | Improve the control concepts |
| Exact size or color searches are weak | Review variant indexing and Shopify options |
| Visual similarity is weak | Review image sources and re-sync |
| Storefront differs from Preview | Check controls, filters, rules, language, and theme mode |
A disciplined relevance test
Section titled “A disciplined relevance test”For every important query, test:
- An exact known product name.
- A broad category.
- A natural-language intent.
- A common synonym.
- A variant option.
- A collection name.
- A misspelling or short prefix.
- A query that should return no products.
Record expected top products before changing anything. Change one layer at a time and repeat the same set so the improvement is measurable.