DEV Community

Cover image for Which SERP API is fastest, and which one fails least? 30 days of latency and success rates
Ricardo Batista for cloro

Posted on

Which SERP API is fastest, and which one fails least? 30 days of latency and success rates

Over 30 days, from October 20 to November 18, 2026, SerpApi was the fastest of five Google SERP APIs on the same 100 keywords a day: a median of 2.1 seconds and a p90 of 4.9 seconds. DataForSEO, SearchApi and SerpApi failed least, with success rates of 99.8% to 99.9%. Oxylabs failed most, at 92.4%.

  • Latency: the median ran from 2.1 s to 7.3 s and the p90 from 4.9 s to 49.8 s, on a plain request for one results page with no AI Overview.
  • Success rate: four of the five APIs returned a result on 99% or more of requests. Most failures came on a few bad days.
  • Google AI Mode: SerpApi returned an answer on 99.8% of 12,563 requests and Bright Data on 76.6% of 12,560. From October 25, Bright Data was at 91.3%.

How do SERP API providers compare on latency and success rate?

Each API got the same 100 short US keywords once a day, one attempt each with no retries. The keywords are 2.6 words long on average, for example "life insurance", "kubernetes for beginners" and "highest mountain in africa". Latency is our stopwatch around one request, on the requests that succeeded.

API Requests Success rate Median latency p90 latency
SerpApi 3,000 99.8% 2.10 s 4.92 s
SearchApi (12 days) 1,105 99.9% 3.36 s 6.21 s
cloro 3,000 99.0% 4.72 s 11.94 s
DataForSEO 3,000 99.9% 5.32 s 11.44 s
Oxylabs 3,000 92.4% 7.31 s 49.80 s

SearchApi joined the test on November 7, so its row covers 12 days, not 30.

SerpApi was the fastest on both measures. Its p90 was lower than the median of Oxylabs. The p90 is the number to set a client timeout from: one request in ten to Oxylabs took more than 49.8 seconds, so a 30-second timeout would turn some of its successes into failures.

A Hackernoon benchmark from February 24, 2026 gives a similar order for the APIs both tests share: SerpApi at a p50 of 2.53 s, SearchApi at 2.71 s and DataForSEO at 4.54 s. It describes its sample as "thousands of calls over 30 days of usage" and does not say how it defines a successful request.

What success rate should you expect from a Google scraping API?

On plain keyword searches, expect 99% or more. Four of the five APIs were at 99.0% to 99.9% over the window. The 30-day rate hides how the failures arrive. They came in bad days more than as a steady rate:

  • DataForSEO: 3 failures in 3,000. No day under 99%.
  • SearchApi: 1 failure in 1,105, an HTTP 502.
  • SerpApi: 6 failures in 3,000. Five were HTTP 503. One was our own account running out of searches. Two days were under 99%, and none under 97%.
  • cloro: 29 failures in 3,000. 25 of them were on one day, October 30, when HTTP 500 errors took that day to 75%. The other 29 days had 4 failures together.
  • Oxylabs: 229 failures in 3,000. 214 were responses with no parsed content, and 5 were our plan's quota. 12 of the 30 days were under 95%.

"Success" here means the API returned a response with no error. SerpApi reports a failure as an error field inside an HTTP 200, DataForSEO in its task status, and an Oxylabs response counts as failed when it has no parsed results. The test does not check that the results are correct, only that they arrived. The 6 failures caused by our own plans stay in the numbers.

SerpApi vs Bright Data: which is more reliable for Google data?

Bright Data is not in the keyword test above, so for plain Google results this post has no Bright Data number. What the monitor sends to both is Google AI Mode: the same prompts to each API once a day, about 420 a day in several countries and languages.

API Requests Success rate Median latency p90 latency
SerpApi 12,563 99.8% 5.64 s 11.25 s
SearchApi (12 days) 4,632 98.4% 7.47 s 11.91 s
DataForSEO 12,570 95.6% 9.23 s 16.13 s
cloro 12,570 99.8% 12.69 s 24.29 s
Oxylabs 12,570 98.2% 25.07 s 38.25 s
Bright Data 12,560 76.6% 50.25 s 176.42 s

Oxylabs returns AI Mode through a job that the monitor polls every 3 seconds, so its latency includes up to one poll interval.

Bright Data's 30-day rate has two parts. From October 20 to 24, its AI Mode scraper and its SERP scraper timed out on almost every request: 58 answers from 2,091 AI Mode requests, and each failure was a wait of more than 300 seconds for the result. We do not know if the cause was on Bright Data's side or in our account. From October 25 to November 18, Bright Data's AI Mode rate was 91.3% (9,559 of 10,469). In both periods SerpApi was at 99.8% or more. From October 25 it was about 9 times faster at the median (5.7 s against 50.3 s).

So on the Google surface where both were measured, SerpApi was more reliable. The Hackernoon benchmark reports Bright Data at 99.99% on Google search results. Its author, Web Intelligence Hub, describes itself as a seller of proxy networks and web scrapers, and Bright Data is first in its success table. We did not test that endpoint on keywords, so we cannot confirm or refute that figure.

Which SERP API provider has responsive support during production incidents?

We did not measure support response times, and neither does the Hackernoon benchmark. What the data above shows is how often an incident reached a client in this window:

  • DataForSEO and SearchApi: no day under 99% on keyword searches.
  • SerpApi: 2 days under 99%, none under 97%.
  • cloro: 1 day under 99%, at 75%.
  • Oxylabs: 12 of 30 days under 95% on keyword searches.
  • Bright Data: 5 days in a row under 15% on AI Mode.

To judge support itself, ask each vendor which channels are open during an incident and whether the contract has a response-time SLA. Then open a ticket during the trial and time the answer.

Where does cloro fit in these results?

cloro was third of five on median latency and fourth of five on success rate in the keyword test. On Google AI Mode it tied with SerpApi on success rate (99.8%) but was more than twice as slow at the median (12.7 s against 5.6 s).

With the AI Overview, cloro is slower still. In a public benchmark of 1,000 SimpleQA Verified questions in November 2026, its p50 search latency with the AI Overview was 7.96 s, and SerpApi's was 8.20 s, because SerpApi needed a second request for the AI Overview on most questions. For plain Google results in under 3 seconds, SerpApi is the pick from this set. What cloro is built for is one API across Google Search, AI Overview, AI Mode, ChatGPT and Gemini.

How this was measured

  • Window: October 20 to November 18, 2026, in UTC. SearchApi from November 7.
  • Keyword test: 100 US keywords, a location pinned to one US city, one results page, no AI Overview or other add-ons. One request per keyword, per API, per day, with no retries. SerpApi ran with no_cache=true, so no answer came from its cache. Only the first run of a keyword on a day counts.
  • AI Mode test: the same prompt set to every API, once a day.
  • Latency: our stopwatch around one synchronous request, from the same monitor, on successful requests only. Median and p90 are percentiles over those requests.
  • Success: the API returned a response with no error, as defined above.

This comparison is published by cloro, which is one of the APIs measured. Every row comes from the same monitor and the same requests, and the publisher's own rows should be read as self-reported. The daily figures for each pair are also on the publisher's comparison pages, for example SerpApi alternatives.

Top comments (0)