AI Search Scraper API: ChatGPT, Perplexity, AI Mode
ProxyField's AI search scraper APIs collect what AI search engines answer to your prompts: the ChatGPT, Perplexity and Google AI Mode answers for $1.50 per 1,000 successful results each, from $1.00 at volume, and the Google AI Overview, parsed inside the Google SERP API, for $1.20 per 1,000, from $0.80. You pay only for successful results; to start, contact sales.
| API | Entry | Volume, from |
|---|---|---|
| Google AI Mode API | $1.50 | $1.00 |
| ChatGPT scraper API | $1.50 | $1.00 |
| Perplexity scraper API | $1.50 | $1.00 |
Billed on successful results only.
Is there one API that collects answers and cited sources from ChatGPT, Perplexity and Google AI Overviews? Not as one request, and not from the engines themselves. With ProxyField it is one vendor and one bill: one prompt on ChatGPT, Perplexity and the AI Overview is three requests, $42.00 per 10,000 prompts at the entry prices and $28.00 at the volume prices. Every answer comes back with its cited sources as a field, in the same shape on every engine.
| Route | ChatGPT | Perplexity | Google AI Overviews |
|---|---|---|---|
| ProxyField, one bill | ChatGPT Scraper API, $1.50 per 1,000 | Perplexity Scraper API, $1.50 per 1,000 | inside the Google SERP API, $1.20 per 1,000 |
| The engines' own APIs: three keys, three bills | the OpenAI API with its web search tool returns model output with URL citations | the Perplexity Agent API returns an answer with a search_results list | Google publishes no API that returns the AI Overview; it is read from the results page |
The engines' own APIs return model output for your app, which can differ from the answer a person sees in ChatGPT, Perplexity or Google Search.
LLM web scraper: do you need your own LLM API key? No if the tool bundles a model, yes if you use an open-source library with a model provider, and none for a model you run yourself. An LLM web scraper reads web pages and pulls out fields from a plain-language instruction; ProxyField's AI search APIs are a different tool: they collect AI answers to your prompts and need no LLM key of yours.
What each AI search engine returns
ProxyField sells one API per engine; the AI Overview has no API of its own, because it is part of the Google results page.
| Engine | API | Answer text | Cited sources | Per 1,000 successful results |
|---|---|---|---|---|
| ChatGPT | ChatGPT Scraper API | the ChatGPT answer | citations, plus the web searches ChatGPT ran and any product cards | $1.50, from $1.00 |
| Perplexity | Perplexity Scraper API | the Perplexity answer | citations, plus the follow-up questions Perplexity suggests | $1.50, from $1.00 |
| Google AI Mode | Google AI Mode Scraper API | the AI Mode answer | citations | $1.50, from $1.00 |
| Google AI Overviews | Google SERP API | the Google results page as JSON, with the AI Overview, the People Also Ask block and the local pack parsed | ai_overview.citations | $1.20, from $0.80 |
| Not covered | Gemini and every other AI assistant | – | – | – |
Read moreShow less
Each engine has its own endpoint: /v1/chatgpt, /v1/perplexity and /v1/google/ai-mode, with the AI Overview inside /v1/google/search. One prompt to every engine is one request per engine, and the responses share one shape: the answer as text and as Markdown with numbered [n] markers, and citations, a list of the cited sources with position, title, URL, domain and snippet. So one parser reads the sources of every engine. Each API page shows its fields and a sample response.
# The requests the AI search APIs take (access through Contact sales): one prompt, three engines
curl https://api.proxyfield.io/v1/chatgpt -H "Authorization: Bearer <api_key>" -H "Content-Type: application/json" \
-d '{"prompt": "how often should I descale an espresso machine", "country": "gb"}'
curl https://api.proxyfield.io/v1/perplexity -H "Authorization: Bearer <api_key>" -H "Content-Type: application/json" \
-d '{"prompt": "how often should I descale an espresso machine", "country": "gb"}'
curl https://api.proxyfield.io/v1/google/ai-mode -H "Authorization: Bearer <api_key>" -H "Content-Type: application/json" \
-d '{"q": "how often should I descale an espresso machine", "gl": "gb", "hl": "en"}'The cited sources in every response (example):
{
"citations": [
{"position": 1, "title": "How to descale an espresso machine", "url": "https://www.example-coffee-guide.co.uk/descale", "domain": "www.example-coffee-guide.co.uk", "snippet": "Most home machines need descaling every one to three months."}
]
}| One prompt on three engines | Requests | Per 10,000 prompts, entry prices | Per 10,000 prompts, volume prices |
|---|---|---|---|
| ChatGPT + Perplexity + Google results page with the AI Overview | 3 | $15 + $15 + $12 = $42.00 | $10 + $10 + $8 = $28.00 |
| Add Google AI Mode | 4 | $15 + $15 + $12 + $15 = $57.00 | $10 + $10 + $8 + $10 = $38.00 |
For any other public page, the Web Scraping API returns the page, plain HTTP or with JavaScript rendering, from $0.20 per 1,000 successful results.
What is an LLM web scraper, and do you need your own LLM API key?
An LLM web scraper fetches a web page, gives its text to a language model with an instruction such as "return the product name, price and stock as JSON", and returns the fields the model finds. It copes with changing layouts better than fixed CSS selectors. One page in, one JSON object out:
Instruction: Return the product name, price and stock status as JSON.
Page text: ... Acme Trail Shoe ... $89.00 ... Only 3 left in stock ...{"name": "Acme Trail Shoe", "price": "89.00", "in_stock": true}Read moreShow less
Whether you need your own LLM API key depends on how the tool is offered:
| How the tool is offered | Your own LLM API key? | What you pay |
|---|---|---|
| A hosted scraper that bundles a model | no: the model's cost is in the tool's price | the tool's price per page or per extraction |
| An open-source library or framework | yes, from a model provider such as OpenAI, or a model you run yourself | page fetching, plus the model provider's per-token price |
| A self-hosted local model | no key, but your own hardware | servers or GPUs, plus page fetching |
| An AI answer scraper (ProxyField's APIs on this page) | no: you send prompts, not pages | $1.50 per 1,000 successful results, from $1.00 |
ProxyField does not sell a hosted extraction model. If you build the second kind, ProxyField's Web Scraping API fetches the pages for your model:
| Pages to fetch | Plain HTTP ($0.50 → $0.20 per 1,000) | With JavaScript rendering ($2.00 → $1.00 per 1,000) |
|---|---|---|
| 1,000 | $0.50 → $0.20 | $2.00 → $1.00 |
| 10,000 | $5.00 → $2.00 | $20.00 → $10.00 |
| 100,000 | $50.00 → $20.00 | $200.00 → $100.00 |
Add your model provider's per-token price for each page on top, then compare the total with a hosted tool's price per extraction. Open-source and hosted model-based scrapers are compared in our guides to open-source AI scrapers and hosted AI scrapers; how a model can read pages at all is in Can ChatGPT Scrape Websites?.
Do ChatGPT answers vary by country?
Yes: location, language, web search, account settings and run-to-run variation all change the answer. One different answer proves nothing; a different mention rate over 3 to 5 runs per country does. The effect is largest for location-sensitive questions, such as local businesses, because web search works from the approximate location of the request.
Read moreShow less
| Cause | What changes |
|---|---|
| Location | the approximate location of the request, and so the local results web search finds; OpenAI's help article on searching the web with ChatGPT describes how location is used |
| Language | the language of the prompt and of the account |
| Web search | which sources the engine finds and cites at that moment |
| Account settings | memory, custom instructions and past chats on a signed-in account |
| Run-to-run variation | a generated answer can differ even with identical settings |
To check answers from several countries, keep everything fixed except the country:
-
Fix the prompt and the language
The same prompts, word for word, in the same language, with the same web-search setting.
-
Use a clean session
No memory, custom instructions or chat history. Account settings are one of the five causes above, so your everyday signed-in browser is not a clean test.
-
Change only the location
Either the API's country setting (
countryon ChatGPT and Perplexity,glon Google; the countries: the proxy pool's 195+ countries), or your own browser through a residential proxy in each country. -
Split local from neutral prompts
Score location-sensitive prompts ("best plumber near me", "tenant rights here") apart from neutral ones ("explain photosynthesis"): a country effect on the neutral set is the stronger signal.
-
Repeat each prompt 3 to 5 times per country
-
Store everything prompt, country, engine, date, the full answer and the cited sources, then compare mention rates per country.
What it costs by hand today. Your own browser through ProxyField residential proxies: 195+ countries and 21,000+ cities with country, region and city targeting, sticky sessions from 30 s to 36 h for one browser session per country, and HTTP, HTTPS and SOCKS5 on gate.proxyfield.io:7000. Start with the Easy Trial: 1 GB for $2.95, no minimum purchase, no subscription, and the traffic never expires; then $2.95/GB down to $1.95/GB from 1,000 GB.
What it costs on the API. 100 prompts a day in 3 countries, run once, is 9,000 ChatGPT results a month: $13.50 at $1.50 per 1,000. With 5 runs a day: 100 prompts × 3 countries × 5 runs × 30 days = 45,000 results, $67.50 a month, or $45.00 at the $1.00 volume price.
More on why AI answers differ and how teams measure them: AI Web Scraping.
AI search scraper pricing
The ChatGPT, Perplexity and Google AI Mode APIs cost $1.50 per 1,000 successful results, from $1.00 at volume; the Google SERP API, which returns the AI Overview parsed, costs $1.20 per 1,000 successful results, from $0.80 at volume. You are billed on successful results only. The volume prices start at 10,000 successful results in one purchase.
What counts as a successful result: an answer or a results page delivered to you. When an engine shows no answer, the AI APIs return no_result and do not bill it; blocked and timed-out requests are not billed on any engine, and every response says whether it was billed.
What a 30-day month of tracking costs on all four engines, one successful result per prompt per engine per day:
| API | Entry | Volume, from |
|---|---|---|
| Google AI Mode API | $1.50 | $1.00 |
| ChatGPT scraper API | $1.50 | $1.00 |
| Perplexity scraper API | $1.50 | $1.00 |
Billed on successful results only.
Read moreShow less
| Prompts tracked every day | Results a month (4 engines) | At the entry prices | At the volume prices |
|---|---|---|---|
| 100 | 12,000 | $17.10 | $11.40 |
| 1,000 | 120,000 | $171.00 | $114.00 |
The arithmetic for 1,000 prompts a day: 30,000 results per engine; 3 × 30,000 × $1.50 per 1,000 = $135.00 for the three answer APIs, plus 30,000 × $1.20 per 1,000 = $36.00 for the results pages, $171.00 in all. At the volume prices, 3 × $30.00 + $24.00 = $114.00. The AI Overview costs nothing extra: it comes with the Google results page.
Prices are in US dollars; pay in USDT, BTC, ETH or 10 other coins. Balance for APIs: 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.
How to monitor your brand in AI search answers
Fix a set of real buyer prompts, run them on every engine on a regular schedule, and track how often you are named, in what position, and which sources are cited.
-
Write the prompt set
50 to 200 questions your buyers ask, in their words: "best X for Y", "X vs Y", "is X worth it", "how do I …".
-
Pick the engines
ChatGPT, Perplexity and Google AI Mode for full answers; the Google results page for the AI Overview above your rankings.
-
Run the set on a schedule
Weekly is enough to see a trend. The APIs have no scheduler of their own, so your script or tool runs the set, and brand matching is on your side, from the answer text and
citations. Sending many prompts in one call: a batch call takes up to 1,000, with a callback for each finished result. -
Store every answer with the prompt, engine, country, date and settings.
-
Compute the metrics below and compare them week over week.
| Metric | What it tells you |
|---|---|
| Mention rate | the share of answers that name your brand |
| First-place rate | how often you are the first brand named |
| Competitor share | how often each competitor is named, per engine |
| Cited domains | which sites the engines link for your topics |
| Drift | how answers change after a launch, a news event or a model update |
For rank tracking next to the AI answers, see SEO proxies.
Official APIs or scraping the answers?
Use an engine's official API to build your own product on its models; scrape the answers to measure what the engine shows its users. OpenAI, Perplexity and Google all offer developer APIs, but those return model output for your app, and it can differ from the answer a person gets in ChatGPT, Perplexity or Google Search for the same question.
Terms of each engine
Each AI service has its own terms on automated use, and each offers an official programmatic route; check your use with counsel before you collect answers at scale.
| Engine | Its terms | Official route |
|---|---|---|
| ChatGPT | OpenAI's Terms of Use govern automated or programmatic extraction of data or output | the OpenAI API |
| Perplexity | Perplexity's terms govern automated use of its service | the Perplexity developer API |
| Google AI Mode and AI Overviews | Google's terms and spam policies cover automated queries to Search | Google's developer APIs |
ProxyField's rule is the same for every engine: public pages only, and follow the site's terms.
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. What each engine's terms say is covered in Is AI Scraping Legal?, and the wider picture in our guides on scraping law. This is not legal advice.
Start collecting AI search answers
Tell sales which engines you need, how many prompts a month and in which markets. The prices are $1.50 per 1,000 successful results for each answer API and $1.20 for Google results pages, and the volume prices of $1.00 and $0.80 apply from 10,000 results in one purchase. Every API takes a POST to https://api.proxyfield.io/v1/… with a JSON body and your API key in an Authorization: Bearer <api_key> header. Rate limits: 20 concurrent requests per account; higher limits through sales.
AI Search Scraper API FAQ
24/7 support by live chat, including help with setup and integration.
What is an LLM web scraper, and do I need my own LLM API key to use one?
ProxyField's AI search scraper APIs collect what AI engines answer, from $1.50 per 1,000 successful results, and need no LLM API key of yours. An LLM web scraper is the other kind: it fetches pages and has a language model pull fields such as names and prices out of them as JSON. You need no key if the tool bundles a model, a model provider's key if you use an open-source library, and none for a model you run yourself. If you build one, ProxyField's Web Scraping API fetches the pages for $0.50 per 1,000 over plain HTTP ($0.20 at volume) or $2.00 with JavaScript rendering ($1.00), before your model's cost.
What is an AI search scraper API?
An API that sends prompts to AI search engines such as ChatGPT, Perplexity and Google AI Mode and returns their answers, so you can measure what they say about a brand or a topic. ProxyField's cost $1.50 per 1,000 successful results each.
How do I scrape AI answers from ChatGPT, Perplexity and Google?
Send each prompt to the API for that engine and store every answer with the prompt, engine and date. For Google's AI Overview, request the Google results page through the SERP API, which returns it parsed. Every engine returns its cited sources in the same citations field, so one parser reads them all.
Is there one scraping API for ChatGPT, Perplexity, Gemini and Google AI?
No single request, but one vendor and one bill: ProxyField sells one API per engine, ChatGPT, Perplexity and Google AI Mode at $1.50 per 1,000 successful results and Google AI Overviews inside the SERP API at $1.20. One prompt on ChatGPT, Perplexity and the AI Overview is 3 requests, $42.00 per 10,000 prompts ($28.00 at volume); with AI Mode it is 4 requests, $57.00 ($38.00). There is no Gemini API.
How do I scrape Gemini?
ProxyField does not offer a Gemini scraper. For Google's AI answers in Search, the Google AI Mode scraper API returns the AI Mode answer, and the Google SERP API returns the AI Overview parsed.
How much does it cost to scrape AI search answers?
$1.50 per 1,000 successful results for ChatGPT, Perplexity or Google AI Mode, from $1.00 at volume, and $1.20 per 1,000 Google results pages with the AI Overview parsed. 100 prompts a day on all four come to $17.10 a month at the entry prices; 1,000 prompts a day come to $171.00, or $114.00 at the volume prices.
Why scrape the AI engines' web answers instead of using their official APIs?
Because the official APIs return model output for your app, while monitoring needs the answer the engine shows its users, which can search the web and cite sources. If you are building your own product on a model, use its official API.
Do I need an account with each AI engine to collect its answers?
You send prompts to ProxyField's APIs and get the answers back, from $1.50 per 1,000 successful results. You do not need an LLM API key of your own.
Which programming languages can I use to collect AI search answers?
Any language your team already uses to call web APIs and store results, such as Python or Node.js.
What's the rate limit for collecting AI search answers?
Tell sales your monthly prompt volume per engine and how fast you need the answers; billing is per successful result, $1.50 per 1,000 for each answer API.
Show 6 more questionsShow fewer questions
Which AI search engines can I scrape?
ChatGPT, Perplexity and Google AI Mode, each with its own API, plus Google AI Overviews inside the Google SERP API. Gemini and other AI assistants are not covered.
What is the best search API for AI agents?
ProxyField's Google SERP API gives an agent live search results as structured data: the Google results page as JSON, with the AI Overview and the People Also Ask block parsed, for $1.20 per 1,000 successful results, from $0.80; for the pages behind the results, the Web Scraping API returns any public URL from $0.20 per 1,000. Compare any search API on the fields it returns and its price per 1,000 successful results.
Can I build my own AI search engine with a scraping API?
You can build the retrieval part: search results from a SERP API, pages from a web scraping API, then your own index and model on top. ProxyField sells those two inputs, not the model or the search index.
Can I track my brand in AI search answers by country?
Yes, if you keep the prompts, language and session fixed and change only the location: 100 prompts a day in 3 countries, 3 runs each, is 27,000 ChatGPT results, $40.50 a month at $1.50 per 1,000. The APIs take the country as a setting: country for ChatGPT and Perplexity, gl for Google; by hand, ProxyField residential proxies target 195+ countries, starting with the Easy Trial, 1 GB for $2.95.
Do AI search answers change from run to run?
Yes: wording, location, web search and model updates all change them. Run a fixed prompt set 3 to 5 times per market and read trends, such as your mention rate per week, rather than single answers.
Is it legal to scrape AI search engines?
Each engine's terms govern automated use of its service, so review them with your counsel before you collect answers at scale. ProxyField allows public-data and SEO monitoring and forbids access to data behind logins you are not entitled to. This is not legal advice.
Related pages
AI search APIs from $1.50 per 1,000 results
Billed on successful results only. Contact sales for the API and the volume you need. 24/7 live chat helps with setup and integration.