What Is an API?

API stands for Application Programming Interface. MDN's documentation describes it as a set of features and rules inside a program that enable interaction with it through software, as opposed to a human user interface. It also compares an API to a contract: whoever uses it knows what they can ask for, in which format and what they will get back.

On the web, most APIs run over HTTP. A program sends a request to a specific address, called an endpoint, with the parameters it wants, and receives a response, almost always in JSON. To know who is asking, the API usually requires a key or a sign-in via OAuth, and it limits how many requests can be made in a given period.

It is worth separating the API from the tool that uses it. Search Console has a web interface for people and, on top of that, an API for programs. The data behind them is the same, but access rules, limits and what can be automated differ.

An API request always has three parts: what is being asked for, with which parameters and who is asking. In the Search Console API, for example, the searchanalytics.query method takes a date range, the dimensions to group by (query, page, country, device) and two paging parameters: rowLimit, which allows between 1 and 25,000 rows per response with a default of 1,000, and startRow, which sets the row the response starts from.

When the result does not fit in one response, you page through it. Google's guide to getting performance data says to repeat the same request, increasing startRow by 25,000 each time, until you get a response with no rows. This pattern, asking in chunks until nothing is left, is common to almost every API that returns long lists.

APIs also set quotas: how many requests can be made per minute or per day. Going over them breaks nothing, but the API returns an error until the quota resets. That is why tools that depend on an API usually fetch data day by day and keep their own copy.

Almost all SEO work at scale runs through APIs. Dashboards that combine Search Console, Google Analytics and crawl data read those sources through their APIs rather than by copying tables by hand. The same goes for measuring visibility in AI answers: our own tracking sends its questions to ChatGPT, Claude or Gemini through APIs, not by typing into the chat window.

That is why the limits of an API become limits on what you know. Anyone who calls the Search Console API without changing rowLimit gets 1,000 rows, the default, even when there are many more. And the way you ask changes the answer: with more or fewer dimensions, the number of rows and the share of traffic that stays attached to a query both vary. Before comparing figures from two tools, it is worth knowing how each one asked.

This website shows how an API can change a site's architecture. The source code of zds.es names "Astro" as its generator, checked on 8 October 2026. Behind it is our own setup: content is edited in WordPress, and Astro reads it through the WordPress REST API to build each page. For readers, nothing changes. For SEO, it does: everything that reaches Google comes from the front end, and any piece of data the front end does not request from the API does not exist on the published page.

• Read the official reference for each method before using it: parameters, defaults, limits and response format.
• Page through until you get an empty response, and check whether the total number of rows stops at a round figure.
• Request the same metric in two different ways, for instance with more or fewer dimensions, to spot cut-offs.
• Keep keys and access tokens outside the code and with the minimum permissions needed.
• Respect quotas: fetch data in daily batches and keep your own copy instead of repeating requests.
• Record which parameters were used for each dataset, so someone else can repeat the query.

• Using the default rowLimit of 1,000 rows and believing you have all the data.
• Treating a result as complete because paging ended, without checking where it stopped.
• Comparing figures from two tools that query the same API with different parameters.
• Publishing API keys in repositories, screenshots or browser code.
• Treating the web interface and the API as the same thing: they can apply different limits and aggregations.