AI Dataset Search

What does AI Dataset Search do?

AI Dataset Search lets you query any Apify dataset using plain English instead of code. Point it at a dataset — for example the output of a scraper run — describe what you're looking for ("videos about race day at the Belgian Grand Prix", "posts by verified authors", "items with more than 100k plays"), and it returns the matching items ranked by relevance.

Under the hood, an AI model inspects the dataset's structure and turns your prompt into a search plan: fuzzy text search with Fuse.js across the most relevant fields, plus exact and numeric filters. You don't need to know the dataset's field names — the Actor figures them out for you.

Running on the Apify platform gives you API access, scheduling, integrations (Make, Zapier, webhooks), and run monitoring out of the box.

Why use AI Dataset Search?

• No query language to learn — describe what you want in natural language.
• Works with any dataset — the schema is inferred automatically from the data, including nested fields.
• Typo-tolerant — fuzzy matching finds "Belgain GP" even when you meant "Belgian GP".
• Ranked results — every result carries a match score, best matches first.
• Composable — chain it after any scraper: scrape first, then search the results with a prompt.

How to use AI Dataset Search

1. Open the Actor and go to the Input tab.
2. Paste the dataset ID or URL of the dataset you want to search — or, instead, enter an Actor (e.g. apify/instagram-scraper) to search the dataset of its latest successful run.
3. Write a search prompt describing what to find.
4. Set your OpenAI API key as the OPENAI_API_KEY environment variable (the Actor uses an OpenAI model to plan the search).
5. Click Start and open the Output tab when the run finishes.

Input

Field	Type	Description
datasetId	string	Dataset ID (e.g. wqoOfQSYVPInyFZhd) or a dataset URL from the API or Console.
actorId	string	Alternative to datasetId: an Actor ID, username/actor-name, or Store URL — the default dataset of its latest successful run is searched. Ignored when datasetId is set.
prompt	string	Natural-language description of what to find. Required.
maxItems	integer	Maximum number of items loaded from the dataset (default 5000).

Provide either datasetId or actorId.

{
    "datasetId": "wqoOfQSYVPInyFZhd",
    "prompt": "Find TikTok videos about race day at the Belgian Grand Prix",
    "maxItems": 5000
}

Output

Matching items are pushed to the default dataset, sorted by relevance. Each item is the original dataset item plus a score field (lower = better match). You can download the dataset in various formats such as JSON, HTML, CSV, or Excel.

{
    "score": 0.0016,
    "caption": "Come with us to tour F1 hospitality! #f1 #formula1 #f1spa #spagp ...",
    "url": "https://www.tiktok.com/@.../video/...",
    "author": { "uniqueId": "...", "verified": true },
    "stats": { "playCount": 518100, "diggCount": 39300 }
}

The columns after score depend entirely on the dataset you search — the Actor preserves the original items untouched.

How much does it cost to search a dataset?

Each run makes one small OpenAI API call (the search planning step) — typically a fraction of a cent with a mini-class model — plus standard Apify platform usage for the compute. Searching thousands of items completes in seconds, so runs stay well within the free tier limits of both platforms for typical use.

Tips for better results

• Be specific in the prompt. "Videos mentioning Mercedes with over 100k plays" gives the planner both a search phrase and a numeric filter to work with.
• Limit maxItems if you only need a quick answer from a huge dataset — fewer items means faster runs.
• Check the run log. The Actor logs the generated search plan (PLAN), so you can see exactly which fields were searched and which filters were applied.
• The Actor returns at most 100 top results per run.

FAQ and support

Which datasets can I search? Any dataset your Apify account can read: your own runs' outputs or publicly accessible datasets.

Does it modify the source dataset? No — the source dataset is only read; results go to the run's own output dataset.

Is my data sent anywhere? Only the dataset's field structure with one example value per field is sent to OpenAI for planning — not the full dataset contents.

Found a bug or missing a feature? Please open an issue on the Actor's Issues tab — feedback is welcome. Custom solutions (different search backends, embedding-based semantic search, other LLM providers) can be built on request.