Exa Known Pitfalls

Exa Known Pitfalls

Overview

Real gotchas when integrating Exa's neural search API. Exa uses embeddings-based search rather than keyword matching, which creates a different class of failure modes than traditional search APIs. This skill covers the top pitfalls with wrong/right examples.

Pitfall 1: Keyword-Style Queries

Exa's neural search interprets natural language semantically. Boolean operators and keyword syntax degrade results.

import Exa from "exa-js";
const exa = new Exa(process.env.EXA_API_KEY);

// BAD: keyword/boolean style — Exa ignores AND/OR
const bad = await exa.search(
  "python AND machine learning OR deep learning 2024"
);

// GOOD: natural language statement
const good = await exa.search(
  "recent tutorials on building ML models with Python",
  { type: "neural", numResults: 10 }
);

Pitfall 2: Wrong Search Type

Using neural search for exact lookups (URLs, names) or keyword search for conceptual queries silently degrades quality.

// BAD: neural search for a specific URL/identifier
const bad = await exa.search("arxiv.org/abs/2301.00001", { type: "neural" });

// GOOD: keyword for exact terms, neural for concepts
const exactMatch = await exa.search("arxiv.org/abs/2301.00001", {
  type: "keyword",
});
const conceptual = await exa.search(
  "transformer architecture improvements for long context",
  { type: "neural" }
);

Pitfall 3: Expecting Content from search()

search() returns metadata only (URL, title, score). Content requires searchAndContents() or getContents().

// BAD: accessing .text from search() — it's undefined
const results = await exa.search("AI safety research");
const text = results.results[0].text;  // undefined!

// GOOD: use searchAndContents for text/highlights
const withContent = await exa.searchAndContents("AI safety research", {
  numResults: 5,
  text: { maxCharacters: 2000 },
  highlights: { maxCharacters: 500 },
});
console.log(withContent.results[0].text);       // actual content
console.log(withContent.results[0].highlights);  // key excerpts

Pitfall 4: Narrow Date Filters Return Empty

Date filters silently exclude results. A single-day window often returns nothing without error.

// BAD: too narrow, likely returns empty array
const bad = await exa.search("AI news", {
  startPublishedDate: "2025-03-15T00:00:00.000Z",
  endPublishedDate: "2025-03-15T23:59:59.000Z",
});

// GOOD: reasonable window with fallback
let results = await exa.search("AI news", {
  startPublishedDate: "2025-03-01T00:00:00.000Z",
  endPublishedDate: "2025-03-31T23:59:59.000Z",
  numResults: 10,
});
// Fallback if no results
if (results.results.length === 0) {
  results = await exa.search("AI news", { numResults: 10 });
}

Pitfall 5: findSimilar Takes a URL, Not a Query

findSimilar expects a URL as its first argument. Passing a query string gives meaningless results.

// BAD: passing a query string to findSimilar
const bad = await exa.findSimilar("machine learning research papers");

// GOOD: pass a URL — findSimilar finds pages semantically similar to it
const good = await exa.findSimilar("https://arxiv.org/abs/2301.00001", {
  numResults: 10,
  excludeSourceDomain: true,
});

Pitfall 6: Date Filters with company/people Categories

The company and people categories do NOT support date filters. Using them returns a 400 error.

// BAD: date filter with company category → 400 error
const bad = await exa.search("AI startups", {
  category: "company",
  startPublishedDate: "2024-01-01T00:00:00.000Z",  // not supported!
});

// GOOD: company search without date filters
const good = await exa.search("AI startups", {
  category: "company",
  numResults: 10,
});

Pitfall 7: Not Limiting Content Size

Requesting full text without maxCharacters can return massive payloads, increasing latency and cost.

// BAD: unlimited text retrieval
const bad = await exa.searchAndContents("topic", {
  numResults: 20,
  text: true,  // could return megabytes of content
});

// GOOD: limit content size
const good = await exa.searchAndContents("topic", {
  numResults: 10,
  text: { maxCharacters: 2000 },  // cap at 2000 chars per result
  highlights: { maxCharacters: 500 },
});

Pitfall 8: Creating New Client Per Request

Each new Exa() call creates a new HTTP client. Reuse a singleton for connection pooling.

// BAD: new client every request (in a route handler)
app.get("/search", async (req, res) => {
  const exa = new Exa(process.env.EXA_API_KEY);  // wasteful!
  const results = await exa.search(req.query.q);
  res.json(results);
});

// GOOD: singleton client
const exa = new Exa(process.env.EXA_API_KEY);
app.get("/search", async (req, res) => {
  const results = await exa.search(req.query.q);
  res.json(results);
});

Pitfall 9: Ignoring the requestId in Errors

Exa error responses include requestId for support debugging. Always log it.

// BAD: generic error handling
try {
  await exa.search("query");
} catch (err) {
  console.error("Search failed");  // loses diagnostic info
}

// GOOD: capture requestId
try {
  await exa.search("query");
} catch (err: any) {
  console.error("Search failed:", {
    status: err.status,
    message: err.message,
    requestId: err.requestId,  // include when contacting support
    tag: err.error_tag,
  });
}

Quick Review Checklist

• Queries are natural language, not keyword/boolean syntax
• Search type matches the query intent (neural vs keyword)
• Using searchAndContents when page content is needed
• Date filter windows are wide enough (7+ days)
• findSimilar receives URLs, not query strings
• No date filters on company or people categories
• maxCharacters set on text and highlights
• Exa client is a singleton, not created per request
• Error handling captures requestId

Resources

• Exa Search Reference
• Exa Error Codes
• Exa Contents Retrieval

Next Steps

For SDK patterns, see exa-sdk-patterns. For common errors, see exa-common-errors.