Product Discovery Settings
Product Discovery settings let you influence which products Talkom AI recommends and how closely recommendations should match a shopper's request. They are useful for aligning product suggestions with your sales strategy, stock position, and catalog.
These settings guide the recommendation process; they do not replace relevance. For example, boosting best sellers helps popular products appear earlier among suitable matches, but it should not make an unrelated product a good match for the shopper.
Where to find the settings
In Shopify Admin, open Apps > Talkom AI > Agent, then select the settings icon on the Product Discovery section.
Changes apply to new product searches and recommendations after the settings are saved. It is worth trying several representative customer questions after making a change.
Understanding boost strength
Best-selling products, inventory quantity, and new arrivals use the same four strength levels:
Strength | Expected effect |
None | The factor does not receive an additional preference. |
Low | Gives the factor a small advantage when products are otherwise similar. |
Balanced | Gives the factor a noticeable preference while keeping relevance central. |
Aggressive | Gives the factor the strongest preference. Relevant products are still prioritized. |
A boost is not a filter. Products without the boosted quality can still be recommended when they are a better match for the customer.
Best-selling products
This setting favors products with stronger sales performance. It is useful when you want proven, popular products to receive more visibility.
Expected behavior:
Relevant best sellers are more likely to appear earlier.
Less popular products remain eligible when they better match the request.
A stronger setting makes sales performance more influential, but does not turn every search into a list of the store's global best sellers.
Consider Balanced for a general-purpose preference. Aggressive can suit stores that strongly prioritize proven products, while None is appropriate when sales history should not influence recommendations.
Inventory quantity
This setting favors relevant products or variants with more available inventory. It can help guide demand toward products that are well stocked.
Expected behavior:
Well-stocked matches can move ahead of similarly relevant, low-stock matches.
A product is not excluded simply because its inventory is lower.
The setting has no effect when None is selected.
Inventory boosting is different from out-of-stock handling. Inventory quantity influences the order of available products, while out-of-stock handling decides whether unavailable products may appear at all.
New arrivals
This setting favors recently added products. It is useful for launches, seasonal collections, and stores where showing fresh inventory is important.
Expected behavior:
Newer relevant products can move ahead of older products.
Older products remain eligible when they are a stronger match.
A stronger setting gives recent additions more visibility across ordinary relevance-based searches.
Use Balanced for a steady newness preference or Aggressive during a launch period. Choose None when product age should not affect recommendations.
Matching type
Matching type controls how closely products should match the shopper's wording and intent.
Strict
Prioritizes close matches and is less willing to show broader alternatives. This can produce fewer results, especially when customers use unusual wording or your catalog uses different product names.
Best suited to focused catalogs or stores where precision is more important than discovery.
Balanced
Balances close matches with reasonable alternatives. This is a good starting point for many stores because it supports discovery without becoming too broad.
Loose
Allows broader matching and is more forgiving of natural language, synonyms, and less precise customer requests. It can improve discovery in varied catalogs, but may occasionally include products that are only loosely related.
Best suited to conversational shopping experiences and catalogs where customers may describe the same product in many ways.
Out-of-stock product handling
This setting controls whether products with no purchasable variants can appear in recommendations.
Hide
Out-of-stock products are excluded. Customers only see products that currently have an available variant.
Choose this when avoiding unavailable recommendations is the priority.
Keep at End
Out-of-stock products may be included, but available products are always shown first. This keeps unavailable products discoverable without placing them ahead of items customers can buy now.
Choose this for catalogs where customers may still want to discover, revisit, or ask about unavailable products.
Blend
Available and unavailable products are considered together. Relevance and your other settings decide their position, so an unavailable product can appear before an available one.
Choose this when product discovery matters more than immediate availability, such as limited collections or products that are frequently restocked.
AI re-ranker instructions
After suitable products are found, Talkom AI performs a final selection step. The re-ranker instructions let you add store-specific guidance to that step.
You can use these instructions to influence:
Which qualities should be preferred when several products are suitable.
How strict or flexible the final selection should be.
Whether reasonable fallback products should be included.
How many products should normally be shown.
Good instructions are short, specific, and easy to apply. For example:
Prefer products made from sustainable materials when they match the customer's request. Return up to six products.
When several products are equally suitable, prefer our own brand.
For gift searches, show a small variety of price points while staying within the requested product type.
The instructions cannot create products or variants, change catalog information, or bypass filters such as availability and price. They only guide the final choice among the products found by search.
Avoid long or conflicting instructions. When several rules matter, state their priority clearly.
How the settings work together
Talkom AI considers the settings as parts of one recommendation process:
It finds products that match the shopper's request using the selected matching type.
It applies filters such as country, price range, sale status, and out-of-stock handling.
It uses best-selling, inventory, and new-arrival preferences to improve the order of relevant products.
It follows an explicit shopper request such as “cheapest,” “most expensive,” “newest,” or “best selling” within the relevant results.
The AI re-ranker makes the final selection using the conversation and your additional instructions.
Because several factors are considered together, a boost does not guarantee an exact position. When a shopper explicitly asks for an order such as “cheapest,” that request will usually have a stronger visible effect than general boost preferences.
Recommended way to evaluate changes
Use real questions that reflect how your customers shop, and compare the recommendations before and after changing one setting. Useful examples include:
“Show me black T-shirts.”
“What are your best-selling running shoes?”
“Show me the newest summer dresses.”
“What is the cheapest suitable option?”
“Show me similar products that are available now.”
Change one setting at a time when possible. This makes it easier to understand its effect and find the right balance for your store.
Review the settings periodically as your catalog, stock levels, and merchandising priorities change.