NewYour Exa or Tako SDK, up to 5× cheaper.

Use casesFact-checking

Check a claim against what was reported.

For fact-checkers, editors and agents that check their own output: find the coverage of a recent claim, see which articles support or contradict it, and keep the verbatim quote.

Free API keyThe request

POST /v1/searchmode: "deep" · $5.60 per 1,000 queries

The problem

Checking recent claims is still manual

A claim about something that happened this week — in a speech, a post or your own model’s output — can’t be checked against a model’s training data, and web search returns pages about the topic, not a verdict on the claim.

Checking means finding the coverage, reading it, and deciding for each article whether it supports the claim, contradicts it or doesn’t address it — with the quote that shows it.

How it works

  • 01

    Turn the claim into queries

    Two to five queries that name who, what and where. Set days or published_after around the date the claim refers to.

  • 02

    Read the coverage

    Deep mode reads 8 articles per query and returns verbatim highlights; normal reads 4.

  • 03

    A stance per article

    Put the claim in a typed question — supports, contradicts or unrelated — and get a choice with a confidence for every article, based on its text.

  • 04

    Widen from one article

    Similar takes the URL of the best match and returns every outlet that covered the same story.

API

The request

One request does it. The same parameters work over HTTP, with the SDKs and through the MCP server.

What it costs

Recommended · deep$5.60 per 1,000 queries

Deep reads 8 articles per query, with highlights: the quote that supports or contradicts the claim.

Also good · normal$2.20 per 1,000 queries

Normal ($2.20) reads the top 4: enough when the claim is narrow and well covered.

An example month

2,000 claims a month, three deep queries each

6,000 queries in deep mode ≈ $33.60

List prices from our pricing page. Each query in a multi-query request counts as one search; identical requests within 10 minutes are free.

curl https://api.typesearch.ai/v1/search \
  -H "Authorization: Bearer $TYPESEARCH_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "query": ["Acme plant closure Rosario", "Acme layoffs 2026"],
    "mode": "deep",
    "days": 14,
    "max_results": 10,
    "questions": {
      "stance": {
        "type": "choice",
        "instructions": "Claim: “Acme closed its Rosario plant this month.” What does the article say about this claim?",
        "criteria": {
          "supports": null,
          "contradicts": null,
          "unrelated": "Does not address the claim."
        }
      }
    }
  }'

What you get back

  • answers.stance: supports, contradicts or unrelated, with a confidence
  • highlights: the verbatim excerpt from each article
  • Title, URL, source and publication time of every article
  • basis: article when the answer comes from the text, headline when it doesn’t

When to use something else

  • The claim is older than 30 days: our index is a rolling window, so use a news archive.
  • The claim is about data, not news (statistics, scientific results): check the primary source.
  • You need a verdict: we return evidence and calibrated answers per article; a person or your rules decide.

One result, trimmed

{
  "title": "Acme to keep Rosario plant open, cuts one shift",
  "url": "https://diarioejemplo.example/economia/acme-rosario",
  "source": "Diario Ejemplo",
  "published_at": "2026-09-19T11:00:00-03:00",
  "score": 0.95,
  "highlights": ["The company said the plant will remain open and that one of its three shifts will be cut"],
  "answers": {
    "basis": "article",
    "values": {
      "stance": { "type": "choice", "choice": "contradicts", "confidence": 0.88 }
    }
  }
}

Questions

Does typesearch tell me whether a claim is true?

No. It finds the coverage and tells you, per article, whether it supports the claim, contradicts it or doesn’t address it, with the quote. Weighing that evidence is your call — or your agent’s rules.

Can an agent check its own answers with this?

Yes: have the agent turn each factual sentence about a recent event into a query and a typed question, and flag the sentences no article supports.

What does it cost?

One claim with three deep queries costs 3 × $5.60 per 1,000. 6,000 queries in deep mode ≈ $33.60 at list price.

Keep reading

Comparisons

Alternatives

Use cases

Products

Docs

Build it this afternoon.

Get a key with $5 in free credit, run the request above and see what comes back.

Free API keyTalk to sales