SERP API pricing: the three billing models, what a search credit really is, and how to estimate your bill
SERP APIs are sold by monthly search credits, by the thousand requests, or purely pay as you go. Here is how each model bills, the small print that moves the number, and a worked way to size a plan before you buy.
Published by AI Knows Us · Updated 12 October 2026
SERP API pricing comes in three shapes. Some vendors sell a monthly plan with a fixed allowance of successful searches and an hourly throughput ceiling. Some sell by the thousand requests, with a free tier and a lower unit rate as volume rises. Some are pure pay as you go, where each call type carries its own price and you top up a balance. Your bill is therefore decided by three numbers: how many searches you make, how many of them succeed, and how fast you need them back.
The unit nobody checks is the one that matters most. In almost every case you are billed per search request, not per result. A query returning one hundred organic results and a query returning an empty page usually cost the same, so pulling fewer pages of results is a bigger saving than trimming the fields you ask for.
The three models, side by side
Figures below are what each vendor published on its own pricing page, checked on 12 October 2026. Pricing pages change, so confirm the current rate before you commit.
| Model | How it bills | Published example | Suits |
|---|---|---|---|
| Monthly search credits | A fixed allowance of successful searches per month, plus a cap on searches per hour | SerpApi lists a free tier at 250 searches per month and 50 per hour, then plans from USD 25 per month for 1,000 searches and 200 per hour up to USD 275 per month for 30,000 searches and 6,000 per hour | Steady monthly workloads you can forecast |
| Per thousand requests | A unit rate per 1,000 successful requests, falling as committed volume rises | Bright Data lists a free tier of 5,000 requests per month, pay as you go at USD 1.5 per 1,000 requests, and a USD 499 per month plan that includes 380,000 requests with additional requests at USD 1.3 per 1,000 | Spiky volume, and jobs that need high concurrency |
| Pay as you go per call | Each endpoint and queue type carries its own price, drawn from a prepaid balance | DataForSEO prices each API separately on a pay for what you use basis, with a stated minimum payment of USD 50 | Mixed workloads where you use several data types unevenly |
The six line items that decide the real number
- Successful versus attempted. Ask whether failed, errored and cached responses are billed. SerpApi states that only successful searches count toward the monthly allowance and that cached, errored and failed searches do not. Bright Data states that it charges only for successfully delivered data. Get this answer in writing, because a vendor that bills attempts changes your cost on every flaky day.
- Depth of results. Asking for one hundred results in one call is normally one search. Paginating through ten pages of ten results is ten searches for the same data. This is the largest single lever in most bills.
- Locale multipliers. Country, language, city level location and device are usually free parameters, but each distinct combination is a separate search. One query tracked across five cities and two devices is ten searches, not one.
- Speed tier. Low latency and premium infrastructure endpoints cost more per call than the standard queue. Batch overnight work on the standard queue and reserve the fast tier for anything a user is waiting on.
- Throughput ceiling. Monthly plans cap searches per hour as well as per month. A plan with enough monthly credits can still fail your nightly job if the hourly ceiling is below your burst rate, so size on both.
- Specialist endpoints. Maps, Shopping, Images, News, Trends, AI Overview and AI Mode endpoints are often priced apart from plain web search. Price the mix you will actually call, not the headline web search rate.
Estimate the bill in five lines
Write these five numbers down and multiply. Most teams overpay because they skip the second and third.
- Queries: the number of distinct keywords or search strings.
- Locales: country, language, city and device combinations per query.
- Frequency: refreshes per month, which is where daily tracking multiplies hard.
- Depth: pages of results per refresh, kept at one wherever a single deep call will do.
- Headroom: add a quarter for retries, backfills and the week you debug something.
Queries times locales times frequency times depth, plus headroom, gives monthly searches. Divide the peak daily figure by the hours you are willing to spread it over, and that is the hourly throughput you need to buy.
Where daily tracking quietly becomes the whole bill
Frequency is the most common source of a surprise invoice. Two hundred keywords refreshed daily across three cities is 18,000 searches a month before a single retry. The same set refreshed weekly is a little over 4,000. Decide refresh rate per keyword group instead of per account: money terms daily, the long tail monthly, anything you only report on quarterly at the start of the quarter.
Pulling one hundred results when you only report on the top ten is the other common waste. If you need position data alone, a parser that returns the first ten results is cheaper and faster than a full page parse.
What to settle before you sign
- Rollover and refunds. Ask what happens to unused credits on a downgrade, what happens when you exhaust the allowance mid month, and what the refund window is. SerpApi documents a 7 day full refund unless more than 20 percent of searches are used, an early renewal option when an allowance runs out, and leftover searches moving to an extra credits balance on a downgrade.
- Compliance posture. Vendors differ sharply here. SerpApi lists SOC 2 Type II, SOC 3 and ISO 27001 certifications and an option that stops search parameters and results being stored on its systems. If your legal team reviews subprocessors, start with this.
- Service commitments. Look for a published uptime figure and a status page with history, not a sales claim.
- Output shape. Structured JSON with stable field names costs you less in engineering time than raw HTML you parse yourself, and a vendor that changes field names without notice is an expense that never appears on the invoice.
- Exit cost. Write the integration against your own interface so swapping vendor is a day of work rather than a rewrite.
If what you actually need is tracked positions for a client report rather than raw search data in your own product, a packaged tool is cheaper than an API plus the code around it. The categories and how to pick between them are set out in the guide to SERP position tools, and the measurement side is covered in keyword rank tracking.
SERP API pricing questions engineers ask before buying
Why is SERP data sold per search instead of per result?
Because the cost to the vendor is the fetch, not the parse. Rendering a results page from the right country on the right device is the expensive part, so one request to one engine is the natural billing unit however many rows come back.
Is a free tier enough to build on?
It is enough to build and test against. Free tiers published by these vendors run from a few hundred to a few thousand requests a month, which covers integration work and a demo. Size a paid plan before you put anything in front of users, because the hourly ceiling on a free tier is low by design.
How do I avoid paying twice for the same search?
Cache every response with the query, locale, device and timestamp as the key, and serve repeats from cache until the data is older than your reporting window. Most teams find a large share of calls are duplicates generated by dashboards and retries.
Does scraping search results myself work out cheaper?
Count the whole cost before you decide: proxy bandwidth, failed fetches, parser maintenance every time a results layout changes, and engineering attention. A vendor sells you that maintenance at a predictable unit rate, which is why most teams that start with their own scraper end up buying one.
Which model is cheapest for a small agency?
Monthly credit plans usually win when volume is flat and predictable, because the per search rate at a committed tier is lower than a pay as you go rate. Per thousand pricing wins when your month has one heavy audit week and three quiet ones.
What about AI answer endpoints?
Several vendors now sell endpoints for Google AI Overviews and AI Mode alongside plain web search. They are priced and rate limited separately, so if that data is part of your product, get a quote on that endpoint specifically rather than assuming the web search rate applies.
The search result that no API returns a position for
Everything above exists to turn a results page into rows you can store. There is a growing class of answer that produces no results page at all: a buyer describes their problem to an assistant, gets back two or three named companies with a short justification, and never sees ten blue links. There is no position to parse, so the thing worth recording changes shape. It becomes whether you were named, which companies were named beside you, and which pages the answer leaned on. That is a different schema and a different collection job from the one your SERP API feeds. AI Knows Us collects those answers for a brand and keeps the citation list behind each one, so the shortlist a buyer is handed can be tracked with the same discipline you already apply to a keyword table.