It depends on your implementation. With search as you type, a search is sent on every keystroke, so a four-letter query such as "polo" sends four rounds of requests.
Here is how "polo" translates into usage in three common setups:
| Implementation | Search requests | Search operations |
|---|---|---|
| One index. A single products index. | 4 | 4 |
| Two indices, separate calls. Products and suggestions, each queried with its own network call — queries are not bundled. | 8 | 8 |
| Two indices, bundled. Products and suggestions queried together in one multi-query call per keystroke — queries are bundled into a single request. | 4 | 8 |
The pattern: the operation count follows how many queries you run; the request count follows how many network calls you make. Bundling changes the second without changing the first.
What this costs you
On plans sold since March 2023, including Grow, you are billed on search requests. So in the table above, the second and third rows run exactly the same eight queries, but the third row costs half as much.
Some legacy plans bill on operations, in which case the second and third rows cost the same. See How Algolia counts search requests, operations, and records.
How this appears in your Analytics
None of these figures will match your Analytics dashboard. Analytics aggregates the keystroke sequence "p" → "po" → "pol" → "polo" into a single reported search. In the one-index case, four billed requests appear as one search; in the two-index cases, as one search per index.
See Why are the search request numbers in my Billing and Analytics pages different?
Reducing this
See How do I reduce my cost while using InstantSearch? for the two levers that actually work: bundling multi-index queries into one request, and debouncing keystrokes.