ChatGPT Search Scraper

ChatGPT Search Scraper collects what ChatGPT answers to a search query, without login. Give it a list of queries (best CRM for a small business, best running shoes for beginners, Notion vs Obsidian) and get back, for each one, the full answer, every source ChatGPT searched and cited (URL, title, domain, site name, publish date, position), the citations passage by passage (which sources back which part of the answer, with character offsets), and the brands and products ChatGPT recommends, in order, with their official website. One clean row per query, ready for a spreadsheet, a dashboard or an AI pipeline.

US$ 0.08 per answer, plus US$ 0.005 once per run. No monthly fee. Error rows are free.

At a glance

• Price per answer: US$ 0.08 (US$ 80 per 1,000), plus US$ 0.005 once per run
• Engine: ChatGPT with web search, logged-out (what a visitor sees on chatgpt.com)
• Sources: Every searched or cited URL with position, title, domain, site name and publish date, flagged citedInAnswer
• Citations: Each cited passage with its text, startIndex / endIndex in the answer and the sources behind it
• Brands and products: The entities ChatGPT highlights (brands, software, products, places), in order, with category and official website
• Location: Ask from any country with a 2-letter code (US, GB, DE, BR, JP...); answers and sources change by country
• Brand check (optional): Mentioned or not, mention count, rank, cited as a source, same price
• Error rows: Free, with an errorCode

What it fixes

• Price. A ChatGPT answer with its sources and recommended brands costs US$ 0.08 here, with no charge for errors.
• The brands, not just the text. ChatGPT shows recommended brands and products as cards with an "official website" link. They come out as a ranked entities list with the official URL, so you can track who ChatGPT recommends without parsing prose.
• Inputs that just work. Country and language are free text: US, us, Japan, en-GB and pt-BR all work, and an unknown value falls back to US with a log line instead of failing the run.
• Citations you can use. Sources come de-duplicated, with tracking parameters removed, each with its type (citation, official link or plain search result) and whether ChatGPT actually cited it in the answer.
• Readable answers. answer keeps ChatGPT's Markdown (headings, lists, tables) without stray escape characters; answerText is the same answer as plain text for spreadsheets and NLP.
• Clear failures. A query that could not be answered becomes a free error row with a reason, so you never pay for an empty result.

Switching from another ChatGPT search actor? Your existing input works

The input names used by other ChatGPT and AI search actors are accepted as they are, so you can paste your current input:

You send	Read as
queries, prompts, searchQueries, questions, query, prompt, keywords, searchTerms	Search queries (a list or a single string)
country, countryCode, location, proxyCountry, gl	Country (code or name)
language, languageCode, lang, hl	Browser language

Each output row carries both query and prompt, so downstream code written for either name keeps working.

One real row

Query best password manager for families, country US (collected in the Apify cloud on 2026-10-09; answer and lists shortened here):

{
  "type": "answer",
  "query": "best password manager for families",
  "prompt": "best password manager for families",
  "answer": "# Best password managers for families\n\nFor most families, I'd start with **Bitwarden Families, 1Password Families, and Proton Pass Family**. The best choice depends on whether you want the lowest cost, the easiest experience for everyone, or a broader privacy-focused setup ...",
  "answerText": "Best password managers for families\n\nFor most families, I'd start with Bitwarden Families, 1Password Families, and Proton Pass Family. The best choice depends on whether you want the lowest cost, the easiest experience for everyone, or a broader privacy-focused setup ...",
  "answerLength": 2900,
  "sourceCount": 15,
  "citedSourceCount": 8,
  "sources": [
    {
      "position": 1,
      "url": "https://bitwarden.com/products/families/",
      "title": "Family Password Manager",
      "domain": "bitwarden.com",
      "siteName": "Bitwarden",
      "publishedDate": null,
      "citedInAnswer": true,
      "type": "citation"
    },
    {
      "position": 3,
      "url": "https://www.theverge.com/tech/883837/1password-price-increase",
      "title": "1Password is going up in price",
      "domain": "theverge.com",
      "siteName": "The Verge",
      "publishedDate": "2026-02-24",
      "citedInAnswer": true,
      "type": "citation"
    },
    {
      "...": "13 more"
    }
  ],
  "citedDomains": ["bitwarden.com", "1password.com", "theverge.com", "proton.me", "wired.com", "reddit.com"],
  "citationCount": 4,
  "citations": [
    {
      "position": 1,
      "text": "My pick for families who want strong security and flexible sharing without spending much. Everyone gets a private vault, plus you can create shared collections for household accounts. It's also open-source.",
      "startIndex": 599,
      "endIndex": 805,
      "sources": [
        {"url": "https://bitwarden.com/products/families/", "title": "Family Password Manager", "domain": "bitwarden.com", "siteName": "Bitwarden", "publishedDate": null}
      ]
    },
    {
      "...": "3 more"
    }
  ],
  "entityCount": 3,
  "entities": [
    {"position": 1, "name": "Bitwarden Families", "category": "software", "officialUrl": "https://bitwarden.com/products/families/", "imageUrl": null, "firstMentionOffset": 75},
    {"position": 2, "name": "1Password Families", "category": "software", "officialUrl": "https://1password.com/personal-family-security", "imageUrl": null, "firstMentionOffset": 95},
    {"position": 3, "name": "Proton Pass Family", "category": "software", "officialUrl": "https://proton.me/pass", "imageUrl": null, "firstMentionOffset": 119}
  ],
  "searchResultCount": 7,
  "webSearchUsed": true,
  "model": "ChatGPT (logged-out)",
  "country": "US",
  "language": "en",
  "responseSeconds": 49.3,
  "scrapedAt": "2026-10-09T04:59:47Z"
}

What you can do with it

• AI search monitoring: see which brands ChatGPT recommends, and in what order, for the queries that matter in your market, by country.
• GEO / AEO research: find the domains ChatGPT trusts and cites in a category, and where your content needs to earn coverage.
• Competitor tracking: check whether your brand and your competitors appear, get their official link, and follow how that changes run after run.
• Content and SEO research: compare ChatGPT's answer and sources with the classic search results for the same query.
• Datasets for analysis: build a dated record of AI answers, citations and recommendations.

Input

Field	What it does
queries	Search queries, up to 300 characters each
country	Where the queries are asked from (IP location). Default US
language	Browser language. Empty follows the country
brands	Optional: brands or domains to check in each answer (HubSpot | hubspot.com)

{
  "queries": ["What is the best CRM for a small business?", "Notion vs Obsidian for personal notes"],
  "country": "US"
}

With no input at all, the Actor runs one example query, so a test run always returns data.

This Actor is built for search queries. Content generation requests (write, generate, translate, code...) and queries over 300 characters are returned as free INVALID_INPUT rows. Queries are asked one at a time, in a fresh chat each, with a short pause between them.

Output fields

query and prompt (same value), answer (Markdown), answerText (plain text), answerLength, sourceCount, citedSourceCount, sources (position, url, title, domain, siteName, publishedDate, citedInAnswer, type), citedDomains, allDomains, citationCount, citations (position, text, startIndex, endIndex, sources), entityCount, entities (position, name, category, officialUrl, imageUrl, firstMentionOffset), images (pageUrl, title, imageUrl), searchResultCount, webSearchUsed, model, country, language, responseSeconds, scrapedAt.

When you pass brands, each row also has brands (brand, mentioned, mentionCount, firstMentionOffset, firstMentionRank, cited, citationPosition), brandsMentioned and brandsCited.

Error codes (never charged)

errorCode	Meaning
INVALID_INPUT	Query over 300 characters, or a content generation request
BLOCKED	ChatGPT did not answer after retries from fresh IPs; run again
NOT_REACHED, UNEXPECTED	The run timeout arrived, or something unforeseen broke; only that query is lost

Pricing

Event	Price
ChatGPT answer delivered	US$ 0.08
Run started (once per run)	US$ 0.005

100 queries in one run cost US$ 8.005. Error rows are free, a run with only invalid queries pays nothing, and there is no Apify compute or proxy bill on top.

Limitations

• Answers are what a logged-out visitor sees on ChatGPT. AI answers vary from one run to the next; compare several runs before drawing conclusions.
• Each answer takes about 30 to 60 seconds, one query at a time, so 100 queries take around an hour and a half. Keep the default run timeout or raise it for long lists; queries the timeout does not reach come back as free NOT_REACHED rows.
• Not every answer has brand cards or citations: ChatGPT decides when to search the web and when to show cards.
• This Actor reads public answers only and respects the sites' terms.

FAQ

Do I need a ChatGPT account, an API key or cookies? No. Queries are asked logged-out, as a visitor on chatgpt.com sees them, through the Apify proxy.

What does a run cost? US$ 0.08 per answer delivered, plus US$ 0.005 once per run. 100 queries cost US$ 8.005. Error rows are free, a run with only invalid queries pays nothing, and there is no Apify compute or proxy bill on top.

How do I see which brands ChatGPT recommends? Every row has entities: the brands and products ChatGPT highlights, in order, with category and officialUrl. To check your own brand, add it to brands (HubSpot | hubspot.com) and each row also gets brandsMentioned and brandsCited.

Can I ask from another country? Yes. Set country to a 2-letter code (US, GB, DE, BR, JP...) or a country name. Answers and cited sources change by country.

How long does a run take, and are there limits? About 30 to 60 seconds per answer, one query at a time, so 100 queries take around an hour and a half. Queries must be search queries up to 300 characters; content generation requests come back as free INVALID_INPUT rows.

Will I get the same answer every time? No. AI answers vary from one run to the next, so compare several runs before drawing conclusions.

Changelog

• 2026-10-17: FAQ in this page.
• 2026-10-09, 0.1: first release. ChatGPT answers with searched and cited sources, passage-level citations with offsets, recommended brands and products with official links, any country, optional brand check, input names of other ChatGPT search actors accepted, free error rows.