Skip to content

New AI assistant answers: ChatGPT, Gemini, Copilot and AI Mode by API

SerpKite
Get API key
Docs menu / Credits and billing

Credits and billing

SerpKite is prepaid. You buy credit packs that never expire, each call spends a known number of credits, and failed or empty calls are refunded automatically.

View as Markdown

How credits work

A credit is the unit every call is priced in. One credit buys one Google results page (up to 10 organic results) from /v1/search or any core vertical. Other endpoints and options cost more or less, as listed below.

  • Prepaid, no subscription. You buy a pack once and spend it down. There is no monthly fee and nothing renews.
  • Credits never expire. Packs stack: buy a Starter today and a Pro next year, and both balances add up.
  • Decimal amounts. Some calls cost half a credit, so balances and headers are decimal numbers such as 61499.5.
  • You only pay for results. Failed, empty and blocked requests cost nothing.

What each call costs

Endpoint or optionCreditsNotes
/v1/search, /v1/news, /v1/images, /v1/videos, /v1/maps, /v1/places, /v1/shopping, /v1/scholar, /v1/patents, /v1/webpage, /customsearch/v11Per results page (10 results).
num=100 depth bundle7Top 100 results in one call instead of 10 pages.
/v1/place1One place's details. Not billed when no place matches.
/v1/reviews1 per 10 reviewsThe first 10 reviews of a place.
/v1/autocomplete0.5
POST /v1/answer2For the model over the result page (4 when it reads pages), on top of the search (1 credit + 1 per page read: 3 at depth=fast, 8 at depth=standard with 3 sources); a failed answer refunds the model part.
POST /v1/extract1Per URL that comes back (0.5 from cache); failed URLs are free.
POST /v1/map1Per call that found URLs (free when nothing is found).
POST /v1/crawl1Per page read (0.5 from cache); the limit is reserved up front and the unused part refunded.
POST /v1/research75Per report at effort=standard (lite 30, deep 200); charged only when the report is written.
POST /v1/monitors1Per run, priced like the search or page fetch it makes (1 per 10 results, num=100 is 7); empty and failed runs are free.
POST /v1/chatgpt4Per ChatGPT answer, the same in a batch (half from the cache with max_age). Failed, refused and empty answers are free.
POST /v1/gemini3Per Gemini answer, the same in a batch (half from the cache with max_age). Failed, refused and empty answers are free.
POST /v1/copilot3Per Copilot Search answer, the same in a batch (half from the cache with max_age). Failed, refused and empty answers are free.
POST /v1/ai_mode2Per Google AI Mode answer, the same in a batch (half from the cache with max_age). Failed, refused and empty answers are free.
POST /v1/brave_ai2Per Brave AI answer, the same in a batch (half from the cache with max_age). Failed, refused and empty answers are free.
include_content=N+1 per fetched pageUp to 5 pages.
include_answer=true (/v1/search, /v1/news)+2+4 with include_content. Refunded when the model fails.
/v1/extract summary / output_schema+4 / +4 per pageNot charged when the AI step fails.
POST /v1/batches×0.5Queued, delivered by poll or webhook. AI assistant answers keep their full price.
Cache hit (max_age)×0.5Only when a fresh-enough cached result exists. Search include_content pages are charged separately.
Failed, empty or blocked request0Reserved credits are refunded automatically.
GET /v1/account, GET /v1/batches/{id}, GET /v1/status0
GET / DELETE /v1/crawl/{id}, monitor GET, PATCH, DELETE, /runs and POST /v1/monitors/{id}/run0A run started with /run is billed like a scheduled one.

A few worked examples:

Request Credits
POST /v1/search with q only 1
POST /v1/search with page: 3 1 (each page is a separate call)
POST /v1/search with num: 100 7 (instead of 10 for ten pages)
POST /v1/search with include_content: 3 1 + 3 = 4
One search queued with POST /v1/batches 0.5
POST /v1/search with max_age: 3600, cache hit 0.5
POST /v1/place, place found 1 (0 when no place has that ID)
POST /v1/reviews, 10 reviews returned 1 (1 per 10 returned; 7 returned costs 1)
POST /v1/autocomplete 0.5
Cached search with include_content: 3, all pages read 0.5 + 3 = 3.5
Extract 3 live URLs, all read successfully 3
Map a site, URLs found 1
Crawl with limit: 25, 18 live pages read 25 reserved, 18 used, 7 refunded
POST /v1/search with include_answer: true 1 + 2 = 3
POST /v1/search with include_content: 3 and include_answer: true 1 + 3 + 4 = 8
POST /v1/answer with depth: "fast" 1 + 2 = 3
POST /v1/answer (standard, 3 pages read) 1 + 3 + 4 = 8
POST /v1/answer, the model fails The search part only (503 ai_unavailable)
Extract 2 URLs with output_schema, both extracted 2 × (1 + 4) = 10
Extract 2 URLs with summary: true, one summary fails 2 + 4 = 6
POST /v1/research with effort: "standard", report written 75
POST /v1/chatgpt, answer returned 4
POST /v1/brave_ai with max_age, cache hit 1
10 chatgpt prompts queued with POST /v1/batches 40 (AI answers are not halved)
POST /v1/gemini, Gemini refuses the visit (503) 0

The batch discount halves search and content-fetch charges. A search cache hit halves the search portion; include_content pages still add 1 credit each (0.5 each in a batch). Batch and cache discounts do not multiply the search portion to one quarter. Half prices are rounded up to the nearest thousandth of a credit, so nothing ever rounds down to free. See Pagination and depth, Caching, Batch requests and Page content for the details of each option.

Extract and crawl bill only successfully read pages (1 credit live, 0.5 from cache). Their failed-page lists are free. Map is free when no URLs are found. A monitor runs at the price of the request it saves; finding no new results does not make an otherwise successful nonempty check free. See Site ingestion and Scheduled monitoring.

AI features

Answers, summaries, JSON extraction and research are written by Claude (Anthropic), with typed checks by OpenAI’s Decisions API. They are priced in credits like everything else, on top of the search or page read they work from:

Option Credits When it’s charged
include_answer: true on /v1/search, /v1/news +2, or +4 with include_content When the answer is written. A failed answer is refunded and noted in meta.warnings
POST /v1/answer Search part (1 + 1 per page read) + 2 (fast) or +4 (standard) The model part only when the answer is written; meta.credits splits the two
summary: true on /v1/extract +4 per page Per summary written; a failed one is reported in summary_error
output_schema on /v1/extract +4 per page Per extraction; a failed one is reported in output_error
POST /v1/research 30 lite, 75 standard, 200 deep, flat Reserved at enqueue, charged when the report is written; refunded when it fails

A failed model step is never billed: you pay only for the search or the pages that were read. Decision checks that keep junk pages out of results (bot checks, login walls, soft 404s) cost you nothing, and those pages are not billed either.

AI assistant answers

The AI assistant endpoints ask a public assistant a prompt as a logged-out visitor and return its answer. Each answer has a flat price: ChatGPT 4, Gemini 3, Copilot Search 3, Google AI Mode 2 and Brave AI 2 credits. track and mentions cost nothing extra.

  • Failed, refused and empty answers are free, like any failed request: the reservation is refunded.
  • Cache hits cost half. With max_age, an identical request answered within the last hour (answers are cached for at most an hour) costs half the price.
  • Batches don’t halve them. Each answer holds a browser for up to a minute and a half, so it costs the same in POST /v1/batches as in realtime.
  • Monitors with an assistant endpoint pay the endpoint’s price on every run.

Google’s AI Overview comes with /v1/search as ai_overview at the price of the search.

What is never billed

  • Errors. Any 4xx or 5xx response costs 0 credits, including 503 upstream_error, 503 upstream_blocked and 503 upstream_timeout. The one exception is /v1/answer: when the search succeeded but the model failed, the call returns 503 ai_unavailable and only the search part is billed (the message says how much). See Errors.
  • Empty results. If Google returns no results for your query (meta.parse_quality: "empty"), the call is refunded.
  • Failed page fetches. With include_content, you pay only for the pages we actually fetched. Pages that time out or refuse the fetch are refunded, and so are short pages that turn out to be a bot check, a login wall, a soft 404 or an empty app shell.
  • Failed AI steps. An answer, summary, extraction or research report the model could not produce is not charged.
  • Bookkeeping endpoints. GET /v1/account, GET /v1/batches/{id} and GET /v1/status are free. Crawl polling/cancellation and monitor management/history are free too; starting or scheduling work still incurs that work’s own charge.

Reservation and refund

When a call starts, SerpKite reserves its maximum possible cost from your balance. When the call finishes, it settles to the real cost and refunds the difference. If the call fails, the whole reservation is refunded. You can see this in two places:

  • The response headers: X-Credits-Used is the final cost (0 for a refunded call) and X-Credits-Remaining is the balance after settlement. See Response headers.
  • The credit ledger in the dashboard (Billing). It is append-only: every purchase, grant, usage and refund is a row, and refunds are separate compensating rows rather than edits. Ledger reasons are grant, purchase, usage and refund. Usage is aggregated per day.

If your balance is lower than the reservation, the call is rejected with 402 insufficient_credits before anything runs.

Free tier

  • 1,000 credits when you sign up and verify your email.
  • 1,500 more the first time you link a GitHub or Google account.
  • 1,000 credits every month for verified accounts, from the 1st of the month after you sign up (00:00 UTC).
  • 5 requests per second, shared by the account’s keys.

No card is required. Free and purchased credits share one balance, and there is no daily limit: free credits can be spent as fast as the rate limit allows.

Free credits are for one account per person. A new account signed up from the same device or network as another account gets no free credits, and its requests get 403 account_restricted until it buys a pack (team members you invite are exempt). The same pause applies to accounts whose requests run in step with several other accounts, and to accounts on a disposable email address; it uses request timing, never query text. If that’s a mistake, write to support@serpkite.com. See Privacy and data retention for what this stores.

Credit packs

PackPriceCreditsPer 1,000Rate limit
Free$01,000 on signup, +1,500 for linking GitHub or Google, 1,000 every month5 req/s
Starter$1010,000$1.0020 req/s
Growth$5062,500$0.8050 req/s
Pro$300500,000$0.60100 req/s

Buying a pack also raises the rate limit of all your keys to that pack’s rate. The rate comes from the largest pack you have ever bought, so it doesn’t drop when your balance runs down.

Committed volume above these packs (from $0.18 per 1,000) is available on Enterprise plans.

Paying

Packs are one-time purchases through Paddle, our merchant of record. Paddle handles the checkout, card processing, VAT and sales tax, and issues the invoice. Listed prices include VAT or GST in the EU, UK and most other countries; in the US, Canada and India, sales tax is added on top at checkout.

  1. Open Billing

    In the dashboard, go to Billing and pick a pack.

  2. Check out

    You are taken to the Paddle checkout. Pay by card or any method Paddle offers in your country.

  3. Credits land on your balance

    When Paddle confirms the payment, the credits are added to your balance, usually within seconds. The grant is idempotent: a retried webhook never credits you twice.

Every paid order is listed under Billing → Orders with a link to the invoice PDF hosted by Paddle. Enterprise customers can pay by invoice.

To top up automatically when the balance runs low, see auto-recharge. To cap how much you spend per month, see Spend controls.

Tracking spend

  • Per request: meta.credits_used in the body and the X-Credits-Used, X-Credits-Remaining and X-Cost-USD headers. X-Cost-USD is the dollar value of the credits used, at the per-credit price of the largest pack you have bought (0 on a free account).
  • Per account and key: GET /v1/account returns the balance and this month’s credits and requests; the dashboard’s Usage page breaks usage down by day, endpoint and key and exports CSV.
  • Alerts: email alerts for low balance and for a percentage of your monthly cap. See Spend controls.

Refunds of purchases

Unused credits don’t expire, so there is rarely a reason to refund a pack. If you bought a pack by mistake and haven’t used any of its credits, email support@serpkite.com with the order ID from Billing → Orders within 14 days of purchase and we refund it through Paddle. Refunded credits are removed from your balance.

Free credits and partly used packs are not refundable, except where the law requires it. Unused purchased credits are forfeited when you delete your account, so ask for a refund first. The full rules are in the Refund Policy.

Last updated: 2026-11-06