For the week of 27 July to 2 August 2026, APIs.io search recorded 123,498 searches across 5,595 unique terms.
That number is almost entirely useless, and explaining why is the most valuable thing on this page.
Read the tier, not the total
| Tier | Searches |
|---|---|
| internal | 118,232 |
| free | 5,266 |
The internal tier is the site calling itself — the build pipeline, the index refresh, our own pages resolving their own data. It is not demand. It is APIs.io talking to APIs.io.
Real external search for that week is 5,266. Not 123,498.
Watch what happens across five weeks:
| Week | Total | Internal | Free |
|---|---|---|---|
| 29 Jun | 28,042 | 27,289 | 499 |
| 6 Jul | 15,506 | 14,493 | 1,013 |
| 13 Jul | 38,183 | 26,911 | 11,265 |
| 20 Jul | 38,015 | 32,611 | 5,404 |
| 27 Jul | 123,498 | 118,232 | 5,266 |
The total more than tripled in the final week. External demand went down slightly, from 5,404 to 5,266. Anyone reporting that week as 340% growth would have been reporting a build job.
I am publishing this because it is the single easiest way to be wrong with your own analytics, and because we nearly were. If your platform instruments itself with the same counters it uses for customers, the counter that matters is the one with the tier filter applied.
The top query is a robot
The week’s highest-volume query was “Automation Preflight API” at 1,185 searches — more than eight times the next term. Nothing about that shape is human.
Strip the machine traffic and the real head of the distribution is small and legible: SCIM (27), Pismo (21), Comcast (18), Stripe (18), football (16), weather (16), Twilio (15). Specific companies, one standard, and two consumer categories.
That is what genuine API discovery looks like. Somebody has a company name and wants to know if it has an API. Somebody has a standard and wants to know who implements it.
The unmet list is the roadmap
279 terms went unmet — searched for, nothing good returned. That number is the output of a filter, not a raw count: 695 terms failed validation, 416 were dropped as noise, and 0 were dropped as already satisfied.
The zero-result list for the week is the honest version of a to-do:
| Query | Searches |
|---|---|
| archlending.com | 21 |
| chainfi-finance | 21 |
| Pismo | 8 |
| Insurtech | 4 |
| SoftPro Sync | 4 |
| Zoomrooms | 4 |
Two of those are domains typed straight into the search box — someone checking whether a specific company is in the catalog. Those are the highest-value misses, because the intent is unambiguous and the fix is a harvest run.
Then there is Bratwurst, Jackfruit, Cilantro, and Jeopardise at four searches each — an obvious probe pattern, and exactly the kind of thing the noise filter exists to remove before anyone treats it as a coverage gap.
What people wanted and could not have
The upgrade-demand counter records where free-tier users hit a paid boundary:
ratings (29), changes (15), me (9), security (5), scopes (4), compare (4), industries (3), areas (2), gaps (2), regions (2).
Ratings and changes are half of it. People want the score, and they want to know what moved.
The MCP number is small and rising
By surface: api 122,980, mcp 518.
MCP is a rounding error against total volume and a real number against external usage — 518 tool calls against 5,266 external searches. Roughly one in ten queries reaching the catalog through an agent rather than a browser. A year ago that column did not exist.
Takeaway
123,498 searches, of which 5,266 were people. The tier field is the difference between a demand report and a self-portrait, the top query was a bot, and the most actionable signal in the whole dataset is two domain names typed by someone checking whether their company is listed.
Search the catalog at apis.io.