Can Gong's Ask Anything Tell You How Many Customers Asked for a Feature
Not reliably, no. Ask Anything can tell you about the calls it decides to search, and for a narrow question about one deal or one contact, that's a genuinely useful answer. But "how many customers have asked for X" is a broad question across your whole call history, and Gong's own documentation on AI Ask Anything is specific that a broad query runs against "up to the top 100 calls" it finds a match in, not your full library. If your team records more than 100 calls that could plausibly mention the feature, some of the count is simply outside the search.
The second limit matters as much as the first. Ask Anything only reads what Gong recorded. Its own guidance frames answers around "what's been discussed in conversations," meaning calls and emails synced to Gong, not CRM fields and not anything happening in a different tool. A request that came up on a call and again in a Zendesk ticket, a Slack thread, or a Jira comment only shows up to Ask Anything as the call. The other two mentions of the same ask don't exist as far as Gong's AI is concerned, because Gong never saw them.
What Ask Anything is actually scoped to
Gong documents three different search scopes for Ask Anything, and they matter because the useful one for feature-request counting is the widest and least precise:
- A specific deal or account: up to 60 calls and 500 emails from the record's selected time period, per Gong's pipeline review documentation.
- A specific contact: up to 10 recent calls and 80 recent emails with that person.
- A broad question with no deal, account, or contact selected: the top 100 calls that match, drawn from wherever the underlying search decides relevance is highest.
The first two scopes are precise because they're bounded by something real: a named account, a named person. The third scope, the one you'd actually use for "how many customers asked about X," has no such anchor. It returns what its matching found most relevant among the top 100, and gives no indication of how many calls it left out or whether the ones it skipped included your biggest account.
A capped answer still looks like a complete one
The failure mode here isn't that Ask Anything gives a wrong answer. It gives an answer that reads as confident and complete, in the tone of a summary rather than a partial result, and there's no built-in flag that says "12 more calls matched this outside your search window." A team running fewer than 100 calls a month might never notice the cap exists, because their whole history fits inside it. A team running a few hundred calls a month hits it constantly, and the answer they get back looks identical either way.
There's a second, quieter problem underneath the cap. Even a perfect search across every call still counts calls, not customers. If the same account raised a request on three separate calls with three different reps, a literal tally of matching calls says three, and the real answer is one account. Getting from "three mentions" to "one customer, three touchpoints" requires resolving speakers to the same person and company across calls, which is a different job than finding the mentions in the first place.
“we need SSO before rollout”
“any update on single sign-on?”
“SSO came up twice on this call”
Linear issue, quotes attached
insight report in NotionThe DIY version
None of this means Ask Anything is broken or that a broad question is worthless. It's a fast way to sanity-check whether a topic has come up at all, and for teams under Gong's search cap, it can be close enough to a real count. The version of "count feature asks in Gong" runs the Ask Anything query for a first pass, then also checks a Gong Tracker built for the same phrase if one exists, since Trackers accumulate hits going forward in a way a one-off query doesn't, and finally searches Zendesk, Slack, or wherever else the account communicates by the customer's name rather than the feature's name, since the same ask often surfaces there in completely different words. We cover Trackers' own backfill and matching limits in more depth in why Gong's Smart Trackers miss feature requests and how to search every Gong call for a feature request, not just one, which walks through combining Trackers, Ask Anything, and the API.
That combination works for an occasional check. It stops working as a running system, because nobody re-runs three separate searches every time a request comes up again, and the running count lives in someone's memory of the last time they checked rather than anywhere durable.
The ceiling on any Gong-only count
The limit isn't Ask Anything's search cap. It's that Gong, by design, only knows about Gong. A count of customer asks that's actually accurate has to include however the same request got phrased in a support ticket, a Slack message, or an issue comment, and Gong's AI has no visibility into any of those, no matter how the question is worded.
Modem is built to count across the channels Gong can't see, and we're upfront that it isn't a better version of Ask Anything. It doesn't out-search Gong's own AI inside a single call, and answering one narrow deal-level question is a job Gong already does well. What Modem does is read transcripts from your connected Gong workspace as calls land, alongside Zendesk, Slack, email, and issue trackers, and resolve every mention of a request, whatever channel it arrived through and however it was phrased, to the same person and company. The count that comes back for "how many customers asked about threshold approvals" includes the four calls, the two tickets, and the Slack mention as one topic with seven accounts attached, not seven separate signals nobody connected. It reads Gong transcripts only, never audio, and never writes back into Gong.
What to check before you trust an Ask Anything count
Before you take a broad Ask Anything answer as a customer count, check two things. Has your call volume for the relevant period plausibly passed 100, and has anyone checked whether the same request shows up outside Gong at all. If both answers are no, the count in front of you is probably close enough to act on. If either answer is yes, treat the number as a floor, not a total, and go looking for the rest by hand until the volume makes that unsustainable.