Skip to main content
Adaptive Intent improves NeuralSearch results on high-volume, high-value queries: the searches that drive the most engagement on your site. Without Adaptive Intent, NeuralSearch relies on general semantic similarity. This means that terms that are specific to your data, such as brand names, niche terminology, or regional product names, might not match the records users expect. Adaptive Intent solves this by learning from the records your users actually click and convert on.

​
How Adaptive Intent works

For each query that receives sufficient engagement, Adaptive Intent combines records that users consistently click and convert on in response to that query into a bag. This bag is represented as a vector, weighted by engagement signal strength:
  • Conversions count 10× more than clicks, so records with stronger engagement have more influence
  • The bag is refreshed regularly as new engagement data arrives
  • At query time, NeuralSearch uses the bag instead of the query vector generated from the query text alone, so results reflect users’ actual intent rather than only semantic similarity
Queries without enough engagement data continue to use NeuralSearch as usual.

​
Where Adaptive Intent helps most

Adaptive Intent addresses problems that standard NeuralSearch struggles with:
  • Brand names and niche terminology. Generic language models might not recognize terms specific to your data: sporting goods brands, regional product names, specialty vocabulary. Adaptive Intent learns these from your data without manual work.
  • Ambiguous queries. Short or vague queries such as “bottle” or “hydro” can match unintended records when relying on generic semantic similarity. Adaptive Intent resolves ambiguity from users’ past behavior.
  • Conversion misalignment. Semantically relevant results aren’t always the ones that convert. Adaptive Intent bridges this gap by learning which results users engage with for each query.

​
Before you begin

To use Adaptive Intent, your index must:
  • Use NeuralSearch
  • Send click and conversion events to Algolia
Adaptive Intent activates automatically for queries that meet a minimum engagement threshold. At least 60 engagement points (1 point for each click and 10 points for each conversion) across at least three records. Queries below this threshold fall back to standard NeuralSearch with no degradation.
If you’re using Dynamic Re-Ranking or Algolia A/B testing, you’re probably already sending the required events.

​
Enable Adaptive Intent from the dashboard

  1. Go to the Algolia dashboard and select your Algolia application and index.
  2. Click the NeuralSearch tab.
  3. Open Settings.
  4. Turn on Adaptive Intent.
Adaptive Intent trains on your engagement data and activates for qualifying queries automatically.

​
Enable Adaptive Intent with the API

Adaptive Intent is controlled through the adiConfig extension in the Search API. To enable it on a per-query basis or within an A/B test:
Command line
curl --request POST \
  --url https://{ALGOLIA_APPLICATION_ID}.algolia.net/1/indexes/{INDEX_NAME}/query \
  --header 'content-type: application/json' \
  --header 'x-algolia-api-key: {ALGOLIA_ANALYTICS_API_KEY}' \
  --header 'x-algolia-application-id: {ALGOLIA_APPLICATION_ID}' \
  --data '{
    "query": "your search query",
    "adiConfig": {
      "enabled": true
    }
  }'

​
Monitor and manage Adaptive Intent

Use the Algolia dashboard to check which queries use Adaptive Intent and compare their results with other retrieval strategies.

​
Review trained queries in Query Explorer

To see the impact of Adaptive Intent on individual queries, go to the NeuralSearch tab, click Compare, and open the Query Explorer. For each trained query, the Query Explorer shows:
  • Number of documents. How many records are in the training set for that query.
  • Similarity score across training documents. A high score may indicate overfitting, while a lower score suggests a more diverse training set.
  • Training status: whether Adaptive Intent is active for that query.
Turn Adaptive Intent on or off for any query with the toggle in the Query Explorer. When you turn off Adaptive Intent for a query, it uses standard NeuralSearch vector retrieval.

​
Compare retrieval strategies

To see how Adaptive Intent affects a specific query’s results:
  1. Go to the Compare tab in the NeuralSearch section of the Algolia dashboard.
  2. Enter a query that has Adaptive Intent training.
  3. Use the toggles on each side to switch between keyword search, vector search, and NeuralSearch with Adaptive Intent.
This lets you compare all three retrieval strategies side by side for the same query.

​
Run an A/B test with Adaptive Intent

To measure Adaptive Intent’s impact before you enable it for all traffic, run an A/B test.

​
Launch an Adaptive Intent test

Before starting the test, check that Adaptive Intent in your NeuralSearch settings is set to Disabled. This ensures the control variant doesn’t use Adaptive Intent.
Command line
curl --request POST \
  --url https://analytics.algolia.com/2/abtests \
  --header 'content-type: application/json' \
  --header 'x-algolia-api-key: {ALGOLIA_ANALYTICS_API_KEY}' \
  --header 'x-algolia-application-id: {ALGOLIA_APPLICATION_ID}' \
  --data '{
    "name": "NeuralSearch vs NeuralSearch with Adaptive Intent",
    "variants": [
      {
        "index": "{INDEX_NAME}",
        "trafficPercentage": 50,
        "description": "NeuralSearch (control)"
      },
      {
        "index": "{INDEX_NAME}",
        "trafficPercentage": 50,
        "customSearchParameters": {
          "adiConfig": {
            "enabled": true
          }
        },
        "description": "NeuralSearch with Adaptive Intent (variant)"
      }
    ],
    "endAt": "2026-12-31T00:00:00.000Z"
  }'
For guidance on test duration and metrics, see NeuralSearch A/B testing.

​
Limitations

  • Adaptive Intent requires click and conversion events. Queries with insufficient engagement fall back to standard NeuralSearch.
  • You can’t manually edit bags, such as by adding or removing records or adjusting weights.
Last modified on June 16, 2026