ChatGPT Results Scraper

ChatGPT Results Scraper sends your questions to chatgpt.com as a logged-out visitor and returns what a person would see on screen. You get the answer in Markdown, plain text and HTML, plus the cited sources, the images, the tables, the place and product cards, a screenshot and ready-made files. It needs no OpenAI account, no API key and no browser on your side, because the pages are opened in hosted browsers that exit from the country you choose.

Ask it "best restaurant in the 15th arrondissement of Paris" with country US and you get one dataset row with the full answer as Markdown, a sources list (title, URL and domain for each cited page), a status, the time it took in milliseconds, and links to an answer- Markdown file, an item- JSON file and a screenshot- PNG of that answer. Send 200 questions in one run and you also get results.md, results.json and citations.csv covering the whole batch, which load straight into a spreadsheet.

What is ChatGPT Results Scraper?

The Actor reads answers to your own prompts. It does not log in, it does not use the OpenAI API, and it does not touch anyone's account. Each query runs in a fresh remote browser that exits from the country you choose, so the answer reflects what a visitor in that country gets.

It is useful for:

• Answer monitoring: track how ChatGPT answers the same question over time or from different countries.
• Source and citation analysis: see which domains ChatGPT cites for a topic, with citations.csv ready for pivot tables.
• Local and shopping research: collect the product cards ChatGPT shows next to an answer.
• Content and brand checks: compare the answer text with your own pages and with competitors.

What data can I extract with ChatGPT Results Scraper?

Field	What it holds
query, country	The question and the exit country used
status, error, attempts	Outcome of the query and how many engine calls it took
answerMarkdown, answerText, answerHtml	The answer in three formats
sources[]	Cited pages: position, title, url, domain, publisher, snippet
products[]	Product cards: name, price, merchant, url
tables[]	Tables of the answer: index, caption, headers, rows
images[]	Images of the answer: position, imageUrl, alt, pageTitle, publisher, context, sourcePosition, collageCount
screenshotUrl, files	Links to the screenshot and to the saved files
conversationUrl, responseTimeMs, fetchedAt	Shared conversation link, timing and timestamp
provider	Which browser network served the last attempt: primary, fallback, or fixture (local test runs only), and null for invalid_input items

How are sources and links handled?

sources lists every web page cited in the answer: the links in the text and the pages behind each citation bullet. Pages are numbered in reading order, and each citation bullet in answerMarkdown shows the numbers of its pages, so [2] is sources[1]. A bullet whose pop-up could not be read is written [?]. An answer that cites nothing has an empty sources list.

url is null when the page could not be opened without leaving the answer. domain, publisher and title are always filled for a page behind a citation bullet, with the site name in publisher and the excerpt shown in the pop-up in snippet (300 characters at most). A link of the text has publisher and snippet set to null. Pages are merged when they share a URL, and pages without URL when they share a domain and a title.

• Map links, shopping product links, links to chatgpt.com or openai.com and links that only wrap an image are left out of sources. They stay links in the answer.
• answerMarkdown and sources[].url have the utm_source tracking parameter removed. answerHtml keeps the original links exactly as ChatGPT rendered them, utm_source included.
• A shopping answer can have sources: [], because its links live in products[].url.

How are images handled?

images lists the pictures ChatGPT shows with the answer, in reading order: the image cards of the text, the pictures of an image mosaic, then the thumbnails of the citation pop-ups. An answer without pictures has an empty images list.

• imageUrl is an absolute https URL. Icons and site favicons are left out, and a URL that appears twice is listed once.
• context is card for an image card of the text, collage for a picture of a mosaic, and popover for the thumbnail of a citation pop-up entry.
• pageTitle is the title of the page the picture comes from, and publisher is the site name when the page gives one. alt is the picture's own description, and it can be empty for a pop-up thumbnail.
• collageCount is the number shown on the badge of a mosaic, which can be larger than the number of pictures listed. It is null for other pictures.
• sourcePosition is set for a popover picture. It is the position of the matching entry in sources, or null when that entry is not in sources.
• answerMarkdown shows each card and collage picture as ![alt](imageUrl) where it stands in the answer. answerText is unchanged, and answerHtml keeps src on the pictures.

Image URLs point to ChatGPT's image cache. They work outside ChatGPT at the time of the run but may stop working later. Download the images you want to keep. The Actor does not download the files itself.

How to use ChatGPT Results Scraper

1. Create a free Apify account.
2. Open ChatGPT Results Scraper.
3. Type one question per line in Queries, and pick a country.
4. Click Start and wait for the run to finish.
5. Read the dataset, or open the key-value store to download the files.

How much will scraping ChatGPT cost you?

ChatGPT Results Scraper uses a pay-per-event pricing model. One event is charged for each query that returns an answer. Queries that end in an error status are not charged, and neither are invalid inputs.

See the Pricing tab for the current price. Set a maximum cost per run in the run options and the Actor stops cleanly when that limit is reached, leaving the remaining queries unprocessed.

⬇️ Input

The input is a JSON object with a list of queries and a few optional settings. See the Input tab for the full schema.

Field	Type	Default	Description
queries	array of strings	required	1 to 200 questions, each 1 to 2,000 characters
country	string	US	Exit country of the browser, picked from a menu of 20 countries (see below). Lowercase codes are accepted in API calls
saveFiles	boolean	true	Save per-query and per-run files in the key-value store
screenshot	boolean	true	Save a screenshot of each answer

Every query waits up to 120 seconds for its answer. ChatGPT answers in the language of your question. The former engine, maxWaitSeconds and language fields no longer exist. If an input still contains them, they are ignored and the run log says so.

Supported countries: United States (US), Australia (AU), Austria (AT), Belgium (BE), Brazil (BR), Canada (CA), Denmark (DK), France (FR), Greece (GR), India (IN), Italy (IT), Japan (JP), Mexico (MX), Netherlands (NL), Poland (PL), Spain (ES), Sweden (SE), Switzerland (CH), Turkey (TR) and United Kingdom (GB). The list only holds countries where the answer was checked to come back localized for that country.

A blank or over-long query does not stop the run. It produces one invalid_input item and the other queries go on.

ChatGPT Results Scraper input example

{
  "queries": [
    "best restaurant in the 15th arrondissement of Paris",
    "meilleur caviste à Bruxelles"
  ],
  "country": "FR",
  "saveFiles": true,
  "screenshot": true
}

Calling the Actor from the API

Run it synchronously and receive the dataset items in the response. Replace YOUR_APIFY_TOKEN with your own token.

curl -X POST "https://api.apify.com/v2/acts/mrbridge~chatgpt-results-scraper/run-sync-get-dataset-items" \
  -H "Authorization: Bearer YOUR_APIFY_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"queries": ["best wine shop in Brussels"], "country": "US"}'

For long batches, start the run asynchronously and read the dataset when it finishes. See the Apify API reference and the API tab of the Actor.

⬆️ Output

Each query produces one dataset item, in the order the queries finish. The item below comes from a local test run on a recorded fixture, with the HTML, the image URL and the file links shortened. The provider value is set to primary to show what a platform run returns, where the local value is fixture. On the platform the file links are https://api.apify.com/... URLs of your run's key-value store.

{
  "query": "meilleur caviste à Bruxelles",
  "status": "ok",
  "error": null,
  "provider": "primary",
  "attempts": 1,
  "country": "US",
  "conversationUrl": "https://chatgpt.com/uc/6acf5558-7c34-83ea-8d88-9893243ab1ba",
  "answerMarkdown": "Si je devais recommander les meilleurs cavistes à Bruxelles, je distinguerais les spécialistes du vin naturel, les caves traditionnelles avec de belles bouteilles et les adresses idéales pour trouver un vin original sans se ruiner.\n\n## Mon top des cavistes à Bruxelles\n\n## 1. Cave 1929\n\nRue de Rollebeek 38, 1000 Bruxelles\n\nUne cave familiale historique, avec plus de 1 500 références [...]",
  "answerText": "Si je devais recommander les meilleurs cavistes à Bruxelles [...]",
  "answerHtml": "<p>Si je devais recommander [...]</p>",
  "sources": [],
  "products": [],
  "tables": [],
  "images": [
    {
      "position": 1,
      "imageUrl": "https://images.openai.com/static-rsc-4/[...]?purpose=inline",
      "alt": "CAVE 1929 | caviste indépendant | Rue de Rollebeek 38, Bruxelles, Belgique",
      "pageTitle": "CAVE 1929 | caviste indépendant",
      "publisher": "Rue de Rollebeek 38, Bruxelles, Belgique",
      "context": "card",
      "sourcePosition": null,
      "collageCount": null
    }
  ],
  "screenshotUrl": null,
  "files": {
    "markdownUrl": "https://api.apify.com/v2/key-value-stores/STORE_ID/records/answer-meilleur-caviste-a-bruxelles-bdd9f9.md",
    "jsonUrl": "https://api.apify.com/v2/key-value-stores/STORE_ID/records/item-meilleur-caviste-a-bruxelles-bdd9f9.json",
    "tableCsvUrls": []
  },
  "filesWarning": null,
  "responseTimeMs": 15,
  "fetchedAt": "2026-10-16T09:16:13.265Z"
}

The recorded fixture has no cited links, so sources is empty here. On a live answer, a link of the text and a page behind a citation bullet look like this:

{ "position": 1, "title": "Time Out Paris", "url": "https://www.timeout.fr/paris/restaurants", "domain": "timeout.fr", "publisher": null, "snippet": null }
{ "position": 2, "title": "Webb Measures Distance to Farthest Fast Radio Burst, Suggesting Origin", "url": null, "domain": "science.nasa.gov", "publisher": "NASA Science", "snippet": "October 8, 2026, First discovered in 2007, fast radio bursts are enigmatic, millisecond-long flashes of radio emission from the distant universe." }

Notes on the structured fields:

• products[] is filled only when ChatGPT displays shopping cards for the answer.
• tables[] holds each table of the answer with its headers and rows, numbered from 1 in reading order.
• images[] holds the pictures of the answer, see above. The publisher is read from the text after the last | of the alt, so a page title that itself contains a | can be cut at the wrong place, as in the example.
• provider is primary or fallback on the platform and fixture for local test runs only. It is null for invalid_input items. Some countries are served directly by the fallback browser network, usually with a single attempt. attempts counts every engine call made for the query.
• filesWarning is set when the answer was delivered but one upload to the key-value store failed. The matching link is then null.

Downloaded files

The screenshot-<slug>.png record depends only on screenshot, and the other records on saveFiles. The Actor writes them to the run's key-value store. Open the store from the Storage tab of the run, where each kind of file has its own tab (answers, items, screenshots, tables, and the three run files), or fetch a record at https://api.apify.com/v2/key-value-stores/STORE_ID/records/KEY.

Record	Content
answer-<slug>.md	The answer in Markdown (ok items only)
item-<slug>.json	The full dataset item for that query (ok items only)
table-<slug>-<n>.csv	One CSV per table, n starting at 1 (ok items only)
screenshot-<slug>.png	Screenshot of the answer, when screenshot is on
results.md	All queries of the run, one top-level heading per query, error items included
results.json	The array of all dataset items, error items included
citations.csv	One row per source: query, position, title, url, domain, publisher, snippet. An empty cell stands for a null value

The slug is the query in lowercase with non-alphanumeric characters replaced by -, cut to 60 characters, then a dash and six hexadecimal characters that keep similar queries apart. A query with no Latin letter and no digit (for example one written only in Japanese, or only punctuation) has an empty base: the slug is q- followed by 6 hexadecimal characters, such as q-3fa9c1. Two identical queries in one run share a slug and the later one overwrites the files, though each keeps its own dataset row.

Screenshots exist only as screenshot-<slug>.png. The dataset holds the link in screenshotUrl and never the image data. A screenshot is also taken on blocked, no_answer and timeout items to help you see what happened. An item in an error status has at most its screenshot: the answer, item and table files are written for ok items only, while results.md and results.json still list the queries that ended in error.

In the CSV files, a cell that starts with =, +, -, @, a tab or a carriage return is prefixed with an apostrophe, so a spreadsheet shows it as text instead of running it. Valid negative numbers such as -12.5 are left as they are.

Statuses and billing

status	Meaning	Charged
ok	The answer was read	Yes
invalid_input	The query was blank or longer than 2,000 characters	No
blocked	chatgpt.com asked for a login or refused the visit	No
no_answer	The page loaded but no answer was found	No
timeout	The answer did not finish within 120 seconds	No
engine_error	The hosted browsers failed on every attempt	No

The run log holds one line per failed attempt, with the query number, the attempt number and a short reason. Expect roughly 20 to 40 seconds per question, and longer in some countries.

Error items carry empty answer strings, empty arrays and a short error message. The run itself always ends as succeeded, so a bad query never fails your whole batch. If a file upload fails but the answer was read, the item is still ok and still charged, because the answer is what you pay for.

Limits of version 1.1

• The Actor opens at most 20 citation bullets per answer and spends at most 20 seconds on them and on their pages. A bullet left closed is matched to a listed page of the same site when there is one. Otherwise it is written [?].
• The address of a page behind a citation bullet is only known when its opening can be caught without leaving the answer. Otherwise its url is null.
• Deep research and some other prompt types can ask for a login. They return blocked.
• Shopping answers are normal ok items with products[] filled. Their sources is often [], because their links are merchant links already present in products[].url.
• Availability of logged-out answers varies by country and by prompt, and chatgpt.com changes its interface often. A query that worked last week can return blocked or no_answer today.

Frequently asked questions

How does ChatGPT Results Scraper work?

For each query it opens chatgpt.com in a hosted browser, types the question, waits for the answer to finish, then reads the page and builds the item. Up to three queries run at the same time.

Do I need an OpenAI account or API key?

No. The Actor visits chatgpt.com the way a logged-out visitor does. It does not use the OpenAI API, so the answers are those of the public web interface.

Is the answer the same as the one I get in my own account?

Not necessarily. Logged-out answers can differ from signed-in ones, and they depend on the country you set.

Which country should I choose?

The one whose visitors you want to simulate, among the 20 in the menu. The default is US. ChatGPT answers in the language of your question, so there is no language setting.

Can I integrate it with other apps?

Yes. Datasets and files can be sent to Make, Zapier, n8n, Google Sheets and webhooks through the Apify integrations.

Can I use ChatGPT Results Scraper as its own API?

Yes. Call it through the Apify API or the apify-client packages for Node.js and Python, as in the example above.

Can I use ChatGPT Results Scraper via MCP?

Yes. AI assistants such as Claude or ChatGPT can call it through the Apify MCP server. Open the MCP tab of the Actor for setup, or connect to mcp.apify.com and add this Actor as a tool.

Do the image links keep working?

They point to ChatGPT's image cache and they open outside ChatGPT at the time of the run. They may stop working later, so download the images you want to keep.

Why is sources empty on a shopping answer?

The links of a shopping answer are product cards, and they are returned in products[].url instead.

Why did a query end in engine_error?

The hosted browsers could not open chatgpt.com from the chosen country after three attempts, or the connection dropped while the answer was being written, or the page did not load after a retry. This happens now and then in some countries. The item is not charged, and running the same query again later usually works. If the main network keeps failing for a country during a run, the remaining queries of that run switch to the second network.

Your feedback

We are always working on improving the Actors. If you have technical feedback or found a bug, please open a ticket in the Actor's Issues tab.

More tools, scrapers, MCP servers and data studies from the same team: mr-bridge.com.

Legal & responsible use

The Actor only reads the answers to the prompts you send. It uses no account, no login and no personal data of other people. Your prompts and the answers are your responsibility, so do not send personal data you have no right to process.

Automated access to chatgpt.com is a matter of OpenAI's terms of service. Read them and decide whether your use is acceptable for you and your organization. Results can contain personal data, which is protected by the GDPR in the European Union and by other regulations elsewhere. If you are unsure about your use, consult a lawyer. You can also read the Apify blog post on the legality of web scraping.

ChatGPT Results Scraper is an independent tool and is not affiliated with or endorsed by OpenAI.