The LLM Mentions Timeseries endpoints – Timeseries Delta and Timeseries New and Lost – return three AI search volume-based metrics: delta_ai_search_volume, new_ai_search_volume, and lost_ai_search_volume. These metrics are derived from the AI search volume – our proprietary search volume metric that estimates the popularity of queries in AI searches.

While delta_ai_search_volume, new_ai_search_volume, and lost_ai_search_volume are based on the same AI search volume data, they don’t measure the same thing. For example, if you subtract new from lost and expect to get a delta, the numbers won’t match. In this article, we’ll explain what each metric actually counts and why the math breaks down.

If you are not familiar with how the initial ai_search_volume metric is calculated for the LLM Mentions API, we recommend reading a dedicated Help Center article.

A quick note on LLM Mentions database records

To understand the difference between the three metrics, it helps to think in terms of individual records in the LLM Mentions dataset that the endpoints work with. These records are called hashes – LLM question-answer pairs distinguished by a unique combination of platform, location, language, and question. In a given time period, a hash either matches the target you specified or it doesn’t. The same hash can match in one period and drop out in the next. Eventually, the delta_ai_search_volume, new_ai_search_volume, and lost_ai_search_volume display different data about the hash records and are calculated differently. Here is how.

1. How delta_ai_search_volume is calculated

The delta_ai_search_volume metric in the Timeseries Delta endpoint returns the difference between the total AI search volume of all matching hashes in the current period and the total AI search volume of all matching hashes in the previous period.

Every matching hash counts, including those that appear in both periods. The ai_search_volume for a given hash can change over time, and that change is included in the delta. So delta_ai_search_volume reflects everything that changed: hashes that appeared in the set, hashes that were lost, and hashes that stayed but shifted up or down.

2. How new_ai_search_volume and lost_ai_search_volume are calculated

The Timeseries New & Lost endpoint provides a targeted view. The new_ai_search_volume metric captures the total AI search volume for hashes that match the target in the current period but not in the previous one. Conversely, lost_ai_search_volume measures the total search volume for hashes that previously matched but no longer do.

Hashes that match in both periods are excluded from both metrics, even if their ai_search_volume changes. The new and lost counts reflect hashes that have appeared or disappeared. Delta also captures hashes that remained but shifted in volume.

Why don’t the metrics add up

For the same target and period specified, the difference in mentions metrics can be calculated as follows:

delta_mentions = new_mentions − lost_mentions

A hash either mentions the target or it doesn’t. Since the mention count is always 0 or 1 within one hash, any change in delta_mentions comes from a hash appearing in or disappearing from the set.

The AI search volume-based metrics don’t follow the same logic:

delta_ai_search_volume ≠ new_ai_search_volume − lost_ai_search_volume

That’s because the ai_search_volume is a continuous value that changes over time within the same hash. As we mentioned before, the delta metric captures changes in search volume within a hash, whereas the new and lost AI search volume metrics do not. Thus, the delta_ai_search_volume can’t equal the difference between new_ai_search_volume and lost_ai_search_volume.

Let’s explain the logic behind each of the metrics in the practical example. The table below tracks three hashes across August and September, along with how each metric reads them.

Hash	August AI SV	September AI SV	New AI SV	Lost AI SV
1	100	150	–	–
2	200	–	–	200
3	–	300	300	–
Sum	300	450	300	200

 

Note: AI SV = AI search volume.

As we can see from the table, hash 1 matched the target in both months, but its ai_search_volume grew from 100 to 150. Hash 2 matched in August and not September, so its August value counts as lost. Hash 3 matched in September and not August, so its September value counts as new.

Now, let’s calculate the delta_ai_search_volume for September and compare it with the difference between new_ai_search_volume and lost_ai_search_volume values:

• delta_ai_search_volume = 450 − 300 = 150
• new_ai_search_volume − lost_ai_search_volume = 300 − 200 = 100

The gap of 50 comes from hash 1, which grew from 100 to 150 without joining or leaving the set. Delta AI search volume counts that growth, while new-minus-lost AI search volume doesn’t.

How to use each metric

The delta_ai_search_volume is the correct metric when measuring the overall net movement within the target, accounting for changes inside hashes that remain matched across periods.

The new_ai_search_volume and lost_ai_search_volume are suitable metrics for isolating AI search volume related to hashes that either appeared or disappeared within the period, without including in-hash shifts.

For additional details about Timeseries Delta and Timeseries New and Lost endpoints, read the article on historical data endpoints in LLM Mentions API.

If you encounter any issues, feel free to reach out to our 24/7 customer support.