A rise in your record count in API Monitoring > Usage usually has one of the following causes.
1. Reindexing with replaceAllObjects
replaceAllObjects builds a temporary index and then swaps it in. During the build you effectively hold two copies of your dataset — the live index and the temporary one — which can inflate your record count substantially.
Because monthly billing ignores the three days on which you held the most records, you can typically run a full reindex on up to three days a month without any additional record cost. Reindexing more often than that starts counting towards usage.
Where possible, update individual objects as they change and reserve full reindexes for a few times a month.
2. Standard replicas
Standard replicas hold their own copy of the primary index's records, so each one adds its full record count to your total.
3. Adding records to a primary index that has replicas
Following on from the above: if you add records to a primary index, every associated standard replica grows too. A single batch of new records is multiplied by the number of standard replicas.
4. Inconsistent objectIDs when indexing
If you intend to update a record but send a missing or changed objectID, Algolia creates a new record instead of overwriting the existing one. Repeated over an indexing run this can duplicate large parts of your dataset. See Does saveObjects count against my usage?
5. The Magento extension's indexing queue
Magento's indexing queue sends product, category, and page updates to Algolia asynchronously, creating temporary indices as it goes. These produce short-lived record fluctuations. See the Magento 2 quick start.
6. Committed plans measured annually
On some committed plans the record count is calculated across the whole year rather than a single month, which can make the figure look unexpectedly high. See Why do I see a high usage for my Standard or Premium Committed Plan?