ChatGPT Scraper API: Collect ChatGPT Answers
ProxyField's ChatGPT scraper API is in development: it is designed to send your prompt to ChatGPT and return the ChatGPT answer for $1.50 per 1,000 successful results, or $1.00 from 10,000 results in one purchase; a request that returns no result is not billed. You can collect three things: the answer text, the brands it names and in what order, and, only when ChatGPT searched the web, the sources it linked. An answer from the model's own knowledge links no sources, and the same prompt can get a different answer by wording, country and day. OpenAI's Terms of Use list, among things users may not do: "Automatically or programmatically extract data or Output". To start, contact sales.
Entry
$1.50
Volume, from
$1.00
Billed on successful results only.
Want ChatGPT to write or run a scraper for you instead? That is a different job; this page is about collecting ChatGPT's answers. See Can ChatGPT Scrape Websites? or our web scraping guide.
What data you can collect by scraping ChatGPT
Three kinds: the answer text, the brands it names, and the sources it links, and sources exist only when ChatGPT searched the web. With no web search there are no citations to collect. Every answer is one sample of an answer that varies by wording, country and day, so you also keep the run metadata that makes answers comparable.
Read moreShow less
| Data | When it exists | What it is not | In the API's design |
|---|---|---|---|
| The answer text | every successful result | not a fixed fact: the same prompt can get a different answer by wording, country and day | answer_text and answer_markdown, $1.50 per 1,000 successful results |
| Brand mentions and their order | when the answer names brands | not a ranking ChatGPT publishes: you count it from the text | counted on your side from the stored answers (the script below does it); the API returns the answer, not a brand score |
| Cited sources | only when ChatGPT searched the web for that answer; an answer from what the model already knows links none | not proof: OpenAI says ChatGPT can be inaccurate, and a linked page may be outdated or not say what the answer says, so open it before you report it | citations (position, title, URL, domain, snippet), web_search_used (did ChatGPT search) and search_queries (the searches it ran) |
| Run metadata you record | every result you store | not something ChatGPT returns: your own record of how the answer was asked, so runs can be compared | parameters echoes the prompt and country you sent; metadata.created_at and processed_at give the collection time (ISO 8601, UTC); tag holds your own label, such as a run number |
| For comparison: the official OpenAI route | when your app calls the OpenAI API with its web search tool | not the ChatGPT answer: model output for your app, which can differ from what a person sees in ChatGPT; OpenAI documents a url_citation object with the URL and title of each cited source (web search guide) | not our product: call the OpenAI API directly |
| Not available from any ChatGPT scraper | – | other users' conversations, the model's hidden reasoning, its training data | each prompt is asked in a fresh conversation and returns that one answer |
| Not returned by ProxyField's API | – | pages ChatGPT retrieved but did not show as sources | only the sources shown in the answer |
Mentioned vs cited. A brand is mentioned when the answer text names it, and cited when one of its pages is among the linked sources. A brand can be named without being linked, and linked without being named, so count the two separately. OpenAI describes when ChatGPT searches and shows sources in Searching the web with ChatGPT.
The terms, next to the data. Read OpenAI's Terms of Use, linked in the section below, and check your use with your counsel before you collect at scale.
What the API is designed to return
The API is in development; this is its design, documented in /docs/api/.
| Field | What you get |
|---|---|
answer_text, answer_markdown | The answer as plain text and as Markdown; answer_html too when you ask for it |
citations | The sources the answer links: position, title, URL, domain and snippet |
web_search_used | Whether ChatGPT searched the web for this answer |
search_queries | The web searches ChatGPT ran for the answer, when it shows them |
products | Product cards in the answer: title, price, currency, rating, review count, merchant and link |
model | The model name, when ChatGPT shows one |
Sample request and response (the design)
The design takes a POST to https://api.proxyfield.io/v1/chatgpt with your API key in an Authorization: Bearer <api_key> header and a JSON body: prompt (up to 4,000 characters), country (the country the question is asked from, any of the pool's 195+), web_search (on unless you turn it off) and include_html. It allows 20 concurrent requests per account and batches of up to 1,000 prompts, with results kept 48 hours.
# The request the ChatGPT scraper API is designed to take (in development; access through Contact sales)
curl https://api.proxyfield.io/v1/chatgpt \
-H "Authorization: Bearer <api_key>" \
-H "Content-Type: application/json" \
-d '{"prompt": "What are good lightweight trail running shoes for beginners?", "country": "us", "web_search": true}'Example response (shortened):
{
"metadata": {"id": "req_c40e8b2a9f1d6377", "api": "chatgpt", "status": "success", "billed": true},
"parameters": {"prompt": "What are good lightweight trail running shoes for beginners?", "country": "us", "web_search": true, "include_html": false},
"model": null,
"web_search_used": true,
"answer_text": "For beginners, look for a trail shoe with a moderate drop, a protective but flexible sole and lugs of about 4 to 5 mm.",
"citations": [
{"position": 1, "title": "Trail running shoes: a beginner's guide", "url": "https://www.example-running.com/trail-shoes-beginners", "domain": "www.example-running.com", "snippet": "Lug depth and drop matter more than brand."}
],
"search_queries": ["best lightweight trail running shoes beginners 2026", "trail running shoe lug depth drop beginner"],
"products": [
{"position": 1, "title": "Example Trail Runner 3 (men's)", "price": "$129.95", "currency": "USD", "merchant": "Example Sports", "url": "https://www.example-sports.com/p/trail-runner-3"}
]
}An answer that comes back with no text returns no_result and is not billed; neither is a blocked or timed-out request.
Brand monitoring in ChatGPT at scale: which tools
Three kinds of tools do it, and they differ in who counts the mentions: dashboards count them for you on a monthly plan, a scraper API gives you the raw answers to count yourself (500 prompts a day for $15 a month bought in one purchase), and open source costs your servers, IPs and upkeep.
Read moreShow less
| Item | AI-visibility dashboards | ProxyField ChatGPT scraper API (in development) | Open-source scraper |
|---|---|---|---|
| Named tools | the seven in the table below | ProxyField | projects on GitHub |
| What you get | reports and charts | the raw answer per prompt, with citations, to store and count | the raw answer, while the tool still works |
| How you pay | a monthly plan (prices on each vendor's page) | $1.50 per 1,000 successful results, $1.00 from 10,000 in one purchase; no plan | your servers, proxies and time |
| Who counts mentions | the dashboard | your script (below), sheet or BI tool | your script |
| Pick it when | you want reports without a data pipeline | you want every answer in your own database, at any volume, and your own metrics | a few prompts, one country, and you accept the upkeep |
AI-visibility dashboards and what each one tracks
What each vendor's own page says it tracks. Plans and engines change, so check before you buy.
| Dashboard | Engines it names | What it reports |
|---|---|---|
| Ahrefs Brand Radar | AI Overviews, AI Mode, ChatGPT, Perplexity, Gemini, Copilot | custom prompts refreshed daily; mentions, citations and AI share of voice |
| Peec AI | ChatGPT and other assistants | mention frequency, average position, sentiment, cited sources, competitor benchmarks; a dashboard, MCP and API |
| Profound | ChatGPT, Perplexity, Claude, Gemini, Copilot, DeepSeek, Google AI Overviews | citation analytics and a visibility score by platform |
| GEOly | 5 AI platforms, 180+ countries | asks your prompts every day and saves the full answer text; a brand leaderboard with visibility and share, and the domains and pages each answer cites; MCP |
| GEO Monitor | ChatGPT, Perplexity, Claude, Gemini, DeepSeek, Grok | mention rates, position, sentiment, sources cited for competitors but not for you, scorecards exported as CSV |
| Gumshoe | ChatGPT, Gemini, Claude, Perplexity and more | visibility and sentiment by buyer persona and topic, and a citation audit |
| Surfer AI Tracker | ChatGPT and Google AI Overviews | how your brand appears in AI answers, inside Surfer's content and SEO platform |
A dashboard decides how it counts. A scraper API leaves the formulas to you: the recipe and the script below give mention rate, position, share of voice, citation rate and average citation position from your own stored answers. A side-by-side of dashboard plans, prompt quotas and prices is in AI web scraping and AI visibility tools.
The brand-monitoring recipe
Answers vary from run to run, so you measure rates over repeated runs, with the prompt and the country held fixed.
-
Write 50 to 500 prompts in your buyers' words: "best tool for X", "X vs Y", "is X worth it", "alternatives to X".
-
Run each prompt several times a week, for example 5 runs, and keep every answer with the prompt, the run number and the date.
-
Save the answers as a CSV with the columns
prompt,run,answer_textandcitations(the cited URLs in the order the answer lists them, separated by;). -
Run the script below each week and track the numbers over time.
| Metric | Formula |
|---|---|
| Mention rate | answers that name your brand ÷ answers collected |
| Average position | the sum of your positions among the brands named ÷ answers that name you |
| Share of voice | your mentions ÷ all brand mentions in the set (each brand counted once per answer) |
| Citation rate | answers that link your site ÷ answers that link any source |
| Average citation position | the sum of your first link's place among the sources (citations[].position) ÷ answers that link you |
| Mention rate per prompt | for one prompt: runs that name your brand ÷ runs of that prompt; shows which buyer questions you win and lose |
Script: the five numbers from your stored answers
Python 3, standard library only. Set your brand, your domain and the competitors you track, then run python3 metrics.py answers.csv.
import csv, re, sys
BRAND, DOMAIN = "Acme", "acme.com"
BRANDS = [BRAND, "Globex", "Initech", "Umbrella"] # your brand first, then the competitors you track
def first_hit(text, name):
m = re.search(r"\b" + re.escape(name) + r"\b", text, re.I)
return m.start() if m else None
answers = named = pos_sum = all_brands = linked_any = linked_ours = cite_pos_sum = 0
for row in csv.DictReader(open(sys.argv[1], newline="", encoding="utf-8")):
answers += 1
hits = {b: p for b in BRANDS if (p := first_hit(row["answer_text"], b)) is not None}
all_brands += len(hits)
if BRAND in hits:
named += 1
pos_sum += sorted(hits.values()).index(hits[BRAND]) + 1
urls = [u.strip() for u in row["citations"].split(";") if u.strip()]
if urls:
linked_any += 1
ours = [i for i, u in enumerate(urls, start=1) if DOMAIN in u]
if ours:
linked_ours += 1
cite_pos_sum += ours[0] # your first link's place among the sources
print(f"answers: {answers}")
print(f"mention rate: {named / answers:.1%}")
print(f"average position: {pos_sum / named:.1f}" if named else "average position: -")
print(f"share of voice: {named / all_brands:.1%}" if all_brands else "share of voice: -")
print(f"citation rate: {linked_ours / linked_any:.1%}" if linked_any else "citation rate: -")
print(f"average citation position: {cite_pos_sum / linked_ours:.1f}" if linked_ours else "average citation position: -")On a 4-answer test file where Acme is named in 2 answers (second and first among the brands), 10 brand mentions in all, and acme.com linked in 2 of the 3 answers with sources (first in one, second in the other), it prints:
answers: 4
mention rate: 50.0%
average position: 1.5
share of voice: 20.0%
citation rate: 66.7%
average citation position: 1.5Matching names with a regular expression is a starting point: add your product names and common misspellings to BRANDS before you trust the counts.
A worked week (an illustration, not data): 100 prompts × 5 runs = 500 answers. 120 name your brand: a mention rate of 24%. Your positions add up to 264: an average position of 2.2. The 500 answers name brands 1,450 times, 120 of them yours: a share of voice of 8.3%. The week costs 500 × $1.50 ÷ 1,000 = $0.75.
What the recipe costs (30-day month):
| Prompt set | Runs | Results a month | At $1.50 per 1,000 | At $1.00, from 10,000 in one purchase |
|---|---|---|---|---|
| 100 prompts | 5 a week | about 2,150 | about $3.20 | – |
| 100 prompts | 5 a day | 15,000 | $22.50 | $15.00 |
| 500 prompts | 1 a day | 15,000 | $22.50 | $15.00 |
| 2,000 prompts | 5 a week | about 42,900 | about $64.30 | about $42.90 |
| 2,000 prompts | 1 a day | 60,000 | $90.00 | $60.00 |
Countries: the country setting picks the country each prompt is asked from, any of the proxy pool's 195+ countries. To check by hand what ChatGPT answers in another country, see ChatGPT proxies.
Open-source ChatGPT scraper or a paid API?
Decide in one line: neither route is official, and OpenAI's terms bar programmatic extraction, so for model output in your own app use the OpenAI API; for a few prompts in one country, open source is enough; for recurring monitoring, an API ($4.50 a month for 100 prompts a day).
An open-source scraper costs nothing to license but costs your servers, proxy traffic (from $1.95/GB) and the hours to fix it when ChatGPT changes. The API costs $1.50 per 1,000 successful results, and failed requests are not billed.
Read moreShow less
| Your situation | Pick |
|---|---|
| A few prompts, one country, you accept the maintenance | an open-source scraper from GitHub |
| You want no code and no upkeep, and accept the vendor's price per answer | a ready-made scraper hosted on a scraping marketplace |
| Recurring brand monitoring, hundreds of prompts, several countries | a ChatGPT scraper API: 100 prompts a day is $4.50 a month |
| You need model output inside your own app | the OpenAI API: model output for your app, not the ChatGPT answer people see; its web search tool is listed at $10.00 per 1,000 calls plus search content tokens at model rates (OpenAI API pricing) |
| You run an open-source tool and it needs IPs | residential proxies, $2.95/GB for 1 GB down to $1.95/GB |
What to check in a GitHub ChatGPT scraper before you trust it
Each check takes a few minutes on the repository page or in one test run, and none needs a vendor's word:
-
Last commit and open issues
If the newest commit is older than ChatGPT's last interface change, or the open issues say "no longer finds the answer", expect to patch it yourself.
-
Logged in or not
Does it need a ChatGPT account, and whose? A tool that signs in puts that account and its history at risk.
-
Browser or endpoints
Does it drive a real browser (Playwright, Selenium or Puppeteer) or call the web app's own endpoints over HTTP? Endpoint tools use less traffic and break sooner.
-
What it parses
Only the answer text, or the cited sources and the web searches too? Brand monitoring needs the sources.
-
Failed prompts
Does it save the prompts that failed, so you rerun those and not the whole set?
-
Proxies and blocks
Can you set a proxy per session, and what does it do when a datacenter IP gets a challenge page: stop, retry or save an empty answer?
-
Licence
Check that its licence allows your use, commercial use included.
What breaks, and who deals with it
Open-source ChatGPT scrapers either drive a real browser against the ChatGPT web app or call the web app's own endpoints over HTTP, and read the answer from the response.
| What breaks | Open-source scraper | ProxyField ChatGPT scraper API (design) |
|---|---|---|
| The ChatGPT interface changes and the tool stops finding the answer on the page | you wait for a fix or patch it yourself | you keep sending the same prompts; a request that returns no result is not billed |
| Browser sessions and any logins expire or get challenged | you manage the sessions and accounts | you send prompts only; no browser on your side |
| You need answers from several countries | you buy IPs in each country and route each run | you set country on each request; no IPs on your side |
| Requests from datacenter IPs get rate-limited or challenged by the site's bot protection | you move to residential IPs, rotate them and slow down | no IPs on your side |
| A run fails | it still costs servers, proxy traffic and time | not billed: you pay only for successful results |
| Cited sources parsed | depends on the tool | yes: citations, plus the web searches ChatGPT ran and any product cards |
What neither route changes: the same prompt can get a different answer by wording, country and day; sources appear only when ChatGPT searched the web; and OpenAI's terms apply to both.
If you run open source: the retry loop
Retry with a growing wait (5 s, 10 s, 20 s, capped at 60 s), take a fresh proxy session, and so a fresh IP, on each retry, and log every prompt that still fails so you rerun it, not the whole set.
import random, time
def run_with_retries(prompt, ask, waits=(5, 10, 20, 60)):
# ask(prompt, proxy_user) runs one prompt through your open-source scraper; it returns the answer or raises
for attempt, wait in enumerate(waits, start=1):
session = f"s{random.randrange(10**8)}" # a new session id asks for a new IP
proxy_user = f"<login>-country-us-session-{session}-ttl-600" # sticky for 10 min, then rotates
try:
return ask(prompt, proxy_user)
except Exception as err:
print(f"try {attempt} failed: {err}; waiting {wait} s")
time.sleep(wait)
with open("failed_prompts.txt", "a", encoding="utf-8") as f:
f.write(prompt + "\n")
return NoneThe proxy line each try uses has the form gate.proxyfield.io:7000:<login>-country-us-session-<id>-ttl-600:<password>. Add a timeout per prompt inside ask, and keep one browser profile per session.
The open-source cost, as a formula: MB per prompt × prompts ÷ 1,000 × the price per GB, plus servers and your time. At an assumed 5 MB per prompt (a browser loads the whole web app; measure your own):
| Prompts a month | Proxy traffic | Proxy cost | On the API |
|---|---|---|---|
| 3,000 | 15 GB | $41.85 at $2.79/GB | $4.50 |
| 15,000 | 75 GB | $198.75 at $2.65/GB | $15.00 at $1.00 in one purchase ($22.50 at the entry price) |
The proxy cost is for traffic alone, before servers and your time.
Proxies for an open-source scraper
ProxyField residential proxies cost $2.95/GB for 1 GB down to $1.95/GB from 1,000 GB: 412M+ residential and 10.2M+ mobile IPs in 195+ countries and 21,000+ cities, with country, region and city targeting, per-request rotation or sticky sessions from 30 s to 36 h, and HTTP, HTTPS and SOCKS5 on gate.proxyfield.io:7000. Start with the Easy Trial, 1 GB for $2.95, with no subscription. All purchases are final; in return, your traffic never expires.
For tools that keep one session per IP, use static residential (ISP) proxies from $2.29 per IP per 30 days. For sites that accept datacenter IPs, datacenter proxies from $0.22/GB also work; test both before you scale.
ChatGPT Scraper API pricing
The ChatGPT scraper API costs $1.50 per 1,000 successful results, or $1.00 from 10,000 successful results in one purchase; a request that does not return a result costs nothing. It is in development: contact sales.
Entry
$1.50
Volume, from
$1.00
Billed on successful results only.
| Workload | Successful results a month | At $1.50 per 1,000 | At $1.00, from 10,000 in one purchase |
|---|---|---|---|
| 100 prompts a day | 3,000 | $4.50 | – |
| 500 prompts a day | 15,000 | $22.50 | $15.00 |
| 2,000 prompts a day | 60,000 | $90.00 | $60.00 |
Prices are in US dollars; pay in USDT, BTC, ETH or 10 other coins, from the same crypto balance as proxies, which never expires. There is no subscription or monthly fee. The APIs are paid from the shared balance, topped up from $2.95; there is no separate test volume. Ordinary purchases, including crypto payments, need no identity verification; we may ask for it if an account shows signs of abuse or breaks our Acceptable Use Policy.
Why scrape ChatGPT instead of calling the OpenAI API?
Because the OpenAI API returns a model's output to your app, while brand monitoring needs the answer as ChatGPT gives it to the people who use it, which can search the web and link its sources. An answer from the API can differ from the one a person gets in ChatGPT for the same question.
How to scrape ChatGPT responses
-
List your prompts in your buyers' words.
-
Send each prompt
One prompt, one request: a
POSTtohttps://api.proxyfield.io/v1/chatgptwithprompt,countryandweb_search, as in the sample above. Many prompts in one call: a batch takes up to 1,000, with a callback for each finished job. -
Store each answer with the prompt, the country and the date you collected it.
-
Compare runs with the recipe and the script above.
Doing it yourself means automating the ChatGPT web app in a real browser, keeping sessions alive and parsing each answer.
Use cases
- Brand visibility in AI answers: mention rate, position, share of voice and citation rate.
- Source research for content teams: which pages and sites ChatGPT links for your topics.
- Answer drift: how answers to the same prompts change after a launch, a news event or a model update.
- Product mentions in shopping answers: which products ChatGPT recommends for a buying question; product cards come back in their own field, with title, price, rating, merchant and link.
For several AI engines on one prompt set, see the AI Search Scraper API.
OpenAI's terms and what we collect
Read the current text of OpenAI's Terms of Use, check your use with your counsel, and see Is AI Scraping Legal? for the wider question. ProxyField's rule for every site: public pages only, and follow the site's terms. The API returns the answer a user of the ChatGPT web version sees.
We collect only the answer to your prompt, never other users' conversations: each prompt is asked in a fresh conversation and returns that one answer.
Our own rules apply too. You may use ProxyField for collecting public web data, price and SEO monitoring, ad verification, testing, market research and similar lawful work. You may not use it to access data behind logins you are not entitled to, infringe intellectual property or break the law. Some site categories are restricted to protect the network, including financial and government services. For the wider legal picture, see our guides on scraping law.
This is not legal advice.
Other AI search engines
ProxyField prices a separate scraper API for each AI engine, each in development and billed on successful results only:
| API | What it returns | Price per 1,000 successful results |
|---|---|---|
| ChatGPT scraper API (this page) | The ChatGPT answer | $1.50, from $1.00 |
| Perplexity scraper API | The Perplexity answer | $1.50, from $1.00 |
| Google AI Mode scraper API | The AI Mode answer | $1.50, from $1.00 |
| Google AI Overviews in the Google SERP API | The Google results page as JSON, with the AI Overview, the People Also Ask block and the local pack parsed | $1.20, from $0.80 |
For any other URL, the Web Scraping API returns the page, plain HTTP or with JS rendering.
Start collecting ChatGPT answers
Tell sales how many prompts you plan to run each month: the price is $1.50 per 1,000 successful results, or $1.00 from 10,000 results in one purchase.
ChatGPT Scraper API FAQ
24/7 support by live chat, including help with setup and integration.
Which tools can scrape ChatGPT answers at scale for brand monitoring?
ProxyField's ChatGPT scraper API, in development, is designed to return the raw ChatGPT answer and its cited sources for each prompt at $1.50 per 1,000 successful results, or $1.00 from 10,000 in one purchase: 500 prompts a day is $15 a month, and a short script turns the stored answers into mention rate, position, share of voice, citation rate and average citation position. Dashboards count the mentions for you on a monthly plan: Ahrefs Brand Radar, Peec AI, Profound, GEOly, GEO Monitor, Gumshoe and Surfer AI Tracker. Open-source scrapers on GitHub suit a few prompts in one country, on your own servers and IPs.
What data can I collect by scraping ChatGPT: answers, cited sources and brand mentions?
Three kinds: the answer text, the brands each answer names and in what order, and, only when ChatGPT searched the web for that answer, the sources it linked; with no web search there are no citations. ProxyField's API is designed to return the answer and its linked sources at $1.50 per 1,000 successful results, with the prompt, the country and the collection time beside each answer, and you count mentions from the stored answers. No scraper gets other users' conversations or the model's hidden reasoning, and ours does not return pages ChatGPT read but did not show. A linked source can be outdated or wrong, so open it before you report it.
Is an open-source ChatGPT scraper from GitHub good enough, or is a paid ChatGPT scraper API more reliable?
Neither route is official, and OpenAI's terms bar programmatic extraction, so for model output in your own app use the OpenAI API. For a few prompts in one country, an open-source scraper is often enough if its last commit is recent, it saves failed prompts, and you supply the browser, sessions, retries and residential IPs and fix it when ChatGPT changes. For recurring monitoring an API is simpler: ProxyField's costs $1.50 per 1,000 successful results, failed requests are not billed, and 15,000 prompts a month is $15 bought in one purchase against about $198.75 of proxy traffic alone at an assumed 5 MB per prompt.
What is the difference between a brand mention and a citation in ChatGPT?
A mention is your brand named in the answer text; a citation is one of your pages among the sources the answer links, which exist only when ChatGPT searched the web. Count them separately: mention rate is answers naming you ÷ answers collected, citation rate is answers linking your site ÷ answers linking any source.
How do I measure my brand's mention rate in ChatGPT?
Run a fixed set of prompts several times a week with the country held fixed, and divide the answers that name your brand by the answers collected; the script on this page does it from a CSV. With ProxyField's ChatGPT scraper API, 100 prompts run 5 times a week are about 2,150 results a month, about $3.20 at $1.50 per 1,000. Track average position and share of voice from the same answers.
How do I scrape ChatGPT responses?
Send each prompt through the API and store the answer it returns with the prompt, the country and the date. Doing it yourself means automating the ChatGPT web app in a real browser, keeping sessions alive and parsing each answer.
How much does a ChatGPT scraper API cost?
ProxyField's costs $1.50 per 1,000 successful results, or $1.00 from 10,000 results in one purchase, and a request that returns no result costs nothing. 100 prompts a day is about $4.50 a month at the entry price. The API is in development: contact sales.
What is the best ChatGPT scraper?
ProxyField's ChatGPT scraper API is the per-result option: $1.50 per 1,000 successful results, with no plan. Which tool fits depends on the job: an open-source tool for a few prompts, a scraper API for recurring monitoring, a dashboard such as Ahrefs Brand Radar or Peec AI if you want reports rather than raw answers.
Is there a ChatGPT scraper on GitHub?
Yes, several open-source projects drive a real browser against the ChatGPT web app, or call its endpoints over HTTP, and save the answers. They cost nothing to license, but you run the browser, the sessions and the IPs, and you patch them when the interface changes. ProxyField's API is the paid alternative at $1.50 per 1,000 successful results.
Why scrape ChatGPT's web answers instead of calling the OpenAI API for brand monitoring?
Because the OpenAI API returns a model's output to your app, while brand monitoring needs the answer as ChatGPT gives it, which can search the web and link its sources. An answer from the API can differ from the one a person gets in ChatGPT. If you only need model output for your own app, use the OpenAI API.
Show 10 more questionsShow fewer questions
Is there one API that collects answers and cited sources from ChatGPT, Perplexity and Google AI Overviews?
ProxyField prices one API per engine, each billed on successful results only: the ChatGPT and Perplexity scraper APIs at $1.50 per 1,000, from $1.00, and AI Overviews parsed inside the SERP API for Google at $1.20 per 1,000, from $0.80. Sending 10,000 prompts to all three costs $42.00 at the entry prices ($15 + $15 + $12). The AI Search Scraper API page covers them together.
What is a ChatGPT scraper?
A ChatGPT scraper sends prompts to ChatGPT and saves the answers, so you can measure what ChatGPT says about a brand, a product or a market. ProxyField's ChatGPT scraper API is designed to do this for $1.50 per 1,000 successful results.
Why would a business scrape ChatGPT responses?
To see how ChatGPT presents your brand and your competitors, which sources it links for your topics, and how its answers change after a launch, a news event or a model update. Buyers ask ChatGPT for recommendations, so its answers shape their shortlists.
Is ChatGPT scraping suitable for large-scale projects?
Yes, through an API: you send the same prompt set regularly and compare the answers, and you pay only for successful results. 2,000 prompts a day is 60,000 results a month: $90 at $1.50 per 1,000, or $60 when bought in one purchase at the $1.00 volume price, which starts at 10,000.
Is there a free ChatGPT scraper?
Open-source ChatGPT scrapers cost nothing to license, but you pay for the servers, the proxy traffic and the fixes. ProxyField's ChatGPT scraper API is paid: $1.50 per 1,000 successful results, and requests that return no result cost nothing.
Can I scrape my own ChatGPT conversations?
To keep your own chat history, use ChatGPT's data export rather than a scraper; OpenAI explains it in Exporting your ChatGPT history and data. ProxyField's API collects answers to new prompts you send; it does not read existing conversations.
Can I scrape ChatGPT with web search on?
Yes: ProxyField's API design asks with web search on unless you set web_search to false, and each result says whether ChatGPT searched and which searches it ran. Answers that use web search draw on the sources found for that request, so they change more from run to run; repeat each prompt and read rates, not single answers.
Can I scrape ChatGPT with a web scraping API?
A web scraping API fetches a URL, but a ChatGPT answer only exists after a prompt is sent to the app. That is the job of the ChatGPT scraper API; ProxyField's Web Scraping API fetches any other public URL, from $0.20 per 1,000 successful results.
Do I need proxies to scrape ChatGPT?
Not with the API: you send prompts and get answers back. If you run an open-source scraper instead, it needs IPs, and ProxyField residential proxies cost from $1.95/GB, with country, region and city targeting.
Related pages
ChatGPT scraper API: $1.50 per 1,000 results, from $1.00
Billed on successful results only. Contact sales for the API and the volume you need. 24/7 live chat helps with setup and integration.