How we use a data API and LLMs to find backlink opportunities: kie.ai and evolink.ai

While running API Hop, we wanted to answer a specific question: Which backlink sources used by other AI API websites are worth looking into?

We chose kie.ai and evolink.ai and fetched the first page of 100 backlink records for each. We did not study traffic, keywords, or overall site rankings for this article. We just wanted a useful backlink list.

Here is the main finding: Of the 200 raw records, 195 remained after we removed 5 broken records. Grouping them by source hostname left just 29 distinct sources. Of evolink.ai’s 100 records, 95 came from GitHub, across just 9 GitHub pages.

In this article, we use this dataset to explain how we collected the records, counted them, used AI to organize them, and decided what to do next.

1. 100 backlink records do not mean 100 backlinks

What did the same parameters return for the two sites?

2. How did we request these two datasets?

The endpoint we used

We first read API Hop’s SimilarwebAPI documentation.

The request format was:

curl --fail-with-body --max-time 90 --get \
  'https://api.apihop.ai/v1/similarweb/websites/backlinks' \
  --header "Authorization: Bearer ${APIHOP_API_KEY}" \
  --data-urlencode 'Key=kie.ai' \
  --data-urlencode 'Page=1' \
  --data-urlencode 'PageSize=100' \
  --data-urlencode 'Status=Active' \
  --data-urlencode 'FollowType=All' \
  --data-urlencode 'sort=DomainScore' \
  --data-urlencode 'asc=false'

The actual structure of a successful response

The top-level structure of the kie.ai response is shown below. For readability, we kept only a few fields from the first record and left out the other records:

{
  "TotalRecords": 42009,
  "Records": [
    {
      "UrlFrom": "https://github.com/yitao2020/Vue-AI-web-game-asset-generator",
      "UrlTo": "https://kie.ai/",
      "Anchor": "KIE.ai",
      "DomainScore": 89,
      "IsLost": false,
      "IsBroken": false
    }
  ],
  "Page": 1,
  "PageSize": 100
}

3. How did we turn 200 records into sources to research?

Use code to remove duplicates and do the counting first

We followed a fixed order:

1. Save both raw JSON files, keeping every field returned by the service.
2. Give each raw row an ID, starting with K001 for kie.ai and E001 for evolink.ai.
3. Keep valid records and retain exclusion flags in the CSV.
4. For each competitor, remove exact duplicates by UrlFrom + UrlTo.
5. Group records by normalized source hostname and count distinct UrlFrom pages.
6. Combine hostnames across the two samples to get 29 distinct sources.

You can ask AI to help with all of these steps. For data analysis, we recommend Claude Opus5.5 and ChatGPT Astra.

The clearest difference was how concentrated the sources were

evolink.ai had far more GitHub records than kie.ai. That caught our attention, so we took a closer look: 95 links came from 9 pages. The source paths included account names close to the brand, such as EvoLinkAI, Evolink-AI, and evolinkai. This shows that its open-source project backlinks were largely concentrated in these 9 projects.

For kie.ai on SaaSHub, 26 links came from 24 pages.

GitHub was the only hostname shared by the two cleaned samples. This fits their positioning well, as developers are one of their main audiences. We think this type of backlink is worth trying.

4. How do we give AI the right material for useful results?

Let code handle the numbers and the model handle classification and analysis

To add this step to your own automation, you can convert the cleaned CSV to JSON and use this prompt:

We are researching backlink sources for an AI API product.
The input contains actual backlink samples for kie.ai and evolink.ai.

Use only the input to organize up to 15 source hostnames:
- source_hostname
- evidence_ids: must come from the input
- source_urls: must come from the input; keep the original URLs
- target_sites: kie.ai, evolink.ai, or both
- category: developer project / product directory / tutorial / community / editorial article / other
- observation: include only facts supported by the input
- suggested_action: keep suggestions separate from facts
- verification_needed: current page status, relevance, account ownership, submission page, pricing terms
- status: always use "Needs verification"

Group by source_hostname.
Do not invent domains, titles, submission pages, or prices.
Do not describe a brand's own repositories as third-party recommendations.
Do not claim that anything is free, guarantees a listing, brings traffic, or improves rankings.
When there is not enough evidence, write "Cannot determine".

We selected 10 backlinks from the actual records

The categories and priorities in the table below reflect our initial judgment of which backlinks are worth pursuing.

6. How much did it cost in total?

In this round, we made two successful backlink API calls through the native service described above. Users making the same requests through API Hop at this standard price can budget as follows:

7. Summary

Our focus was to use API Hop to turn an open-ended question into a repeatable task: Use two queries to get 200 real records, clean and group them by source hostname, then pick 10 URLs worth pursuing further. The data API provides leads, code makes the numbers reproducible, AI helps organize the text, and we make the final checks to decide whether to act.