What Wispr Flow's Founder Revealed About User Tracking (2026)

TL;DR: In a June 2026 Think School podcast interview, Wispr Flow co-founder and CEO Tanay Kothari walked through the sales engine that, he says, lets a two-person team convert 1,000 enterprise deals a month. In doing so, he described — on camera, as a feature — an analytics system that ties individual dictation activity to a person's name, job title, and employer; tracks how many words each user has dictated and which applications they dictate into; pools all of it into Hex, a third-party data-analytics platform; and triggers automated outreach with, in his words, "no person involved in this loop at all." He also described a website tool that "deanonymizes" anonymous visitors down to their company and office IP address. None of this was presented as a leak. It was presented as a playbook — which is exactly why it is worth reading carefully if you dictate sensitive work through a cloud tool.

This article quotes what the founder actually said, explains what each statement implies for your privacy, and contrasts it with the on-device alternative. We are not accusing Wispr Flow of breaking any law — usage analytics and identity enrichment are common in cloud B2B software. We are showing how much granular, identity-linked data a cloud dictation product necessarily holds to work this way, and why an on-device architecture like Voibe makes that collection structurally impossible.

Disclosure: Voibe is our product. We compare Voibe to other tools using verifiable facts — in this case, Wispr Flow's CEO's own public statements on a recorded podcast, quoted with attribution. We have not independently audited Wispr Flow's systems; we are reporting what was said.

The rest of this article walks through each row with the fuller quotes and context, explains why this is inherent to the cloud dictation model rather than a one-off choice, and shows what an on-device architecture changes.

The interview was a business-strategy episode on the Think School channel, hosted by Ganesh Prasad. The framing was admiring: per the host, Wispr Flow launched its consumer product in 2025, reached millions of downloads, counts hundreds of Fortune 500 companies among its users, and runs enterprise sales with a two-person team closing "1,000 deals a month." (These figures are claims made on the podcast by the host and founder; we have not independently verified them.)

Tanay Kothari — co-founder and CEO of Wispr Flow, per the company's media kit — explained that the engine behind those numbers is data. To sell to the right person at the right company with the right message, you first have to know, at the individual level, who your power users are and how they use the product. The entire system rests on a consumer product feeding a continuous stream of identity-linked behavioral data into an analytics layer. That is the part worth slowing down on.

The most striking privacy detail is not that Wispr Flow counts words — it is that it knows where you are dictating. Describing how the system surfaces sales signals inside a company, the founder said it can determine:

"What are the biggest places they're using it? Are they using it for a lot of messaging? Are they using it in engineering to write a lot of code?"

And for understanding an existing customer before a call:

"Take a look at all the users inside the [customer] domain and... what are the different applications they're using whisper in. Look at the usage analytics to figure it out and tell me how that usage has been trending over the last few weeks."

To answer "which applications are they using Whisper in," the client app has to report the name of the application you are dictating into — Slack, VS Code, Gmail, a therapy-notes app, a legal brief — back to the cloud, attached to your account, over time. That is a behavioral fingerprint of your workday. It is one thing for a transcription tool to convert your speech to text; it is another for it to maintain a per-user, time-series record of which apps you speak into and how that changes week to week.

For comparison: an on-device tool has no mechanism to build this record. Voibe transcribes locally and discards the audio; it does not phone home with "Ayush dictated 4,000 words into Cursor this week." There is no engine to query because there is no data to collect.

The founder demonstrated the engine live. He described a typical query against the analytics layer:

"Find me all the people who are power users of whisper which means they've done more than 20,000 words and their title is VP of engineering at companies between say 500 and 5,000 people, and sort them by size of their company."

The system returned a named list with companies and word counts — "I know the total number of words that they've dictated. I know the employees of their company. And so now I have these eight people to go and message." He also described surfacing "who is the most influential person who uses whisper inside [a company]" to target deployment.

Then the host asked him to run it on his own organization. The founder typed a query for the Think School email domain and reported back: "only two users on the Think School domain," adding "I can see who these are" before naming them. In other words, individual dictation activity is resolvable to a specific human being, their title, and their employer, on demand. This is normal for account-based cloud software — but it is worth seeing stated so plainly. When you dictate through Wispr Flow, you are not an anonymous transcription session; you are a named row in a queryable table.

The analytics the founder demonstrated were not running inside some private, sealed Wispr Flow database. He was driving Hex (hex.tech), a third-party AI data-analytics workspace founded in 2019 and used by data teams to query and visualize company data with SQL, Python, and AI. He described it this way:

"This is our core analytics software and this is connected to every single source of data. This is connected to our users. This is connected to their users analytics. This is connected to every single marketing channel... and if somebody's talking to our team, the sales team or the customer support team, it has all the information about that as well."

When the host asked, "I just have to provide all my data to hex.tech?", the founder answered: "Yes."

The privacy point is about data sprawl. Your dictation usage does not just live in one place — it is pooled, alongside marketing-channel data and your support and sales conversations, into a separate analytics platform. Every additional system that holds your data is an additional surface for retention, access control, logging, and breach. This is the same structural lesson we cover in our voice data privacy guide: in the cloud model, "your data" is rarely in one vendor's hands — it is distributed across a chain of subprocessors and analytics tools. Wispr Flow's own published subprocessor list already names analytics, error-tracking, and CRM-sync vendors; the podcast adds Hex as the layer where it is all joined together and queried.

Beyond product usage, the founder described a tactic for turning anonymous web traffic into named sales leads:

"On your website you can add a way where it deanonymizes people who are coming. So you'll get a report... six people from Flipkart visited your website today. They visited your pricing page... laptop IP addresses, Flipkart office."

He presented this as a routine signal: even people who never sign up or log in can be mapped to their employer by correlating their IP address with corporate office networks. To be fair, IP-to-company de-anonymization is a widely used B2B growth tactic, sold by many vendors, and is not unique to Wispr Flow. But hearing it described casually as part of the standard playbook is a useful window into the mindset of the cloud-growth model: visiting a pricing page is treated as a data event tied back to where you work.

The contrast with on-device tools is again architectural. A product that requires no account, runs no per-user telemetry, and processes everything locally is not in a position to assemble this kind of cross-context profile, because it never collects the linking data in the first place. For the broader argument, see why offline dictation matters.

Finally, the founder explained how the data closes the loop into action. Once Wispr Flow learned which message works for which persona, it automated the trigger entirely:

"Whenever a new VP of engineering becomes a power user, we know what message will work on what channel that is automatically sent to them. There's no person involved here in this loop at all."

So crossing a usage threshold — by dictating enough words, in the right apps, at the right kind of company — can automatically enroll you into an outreach sequence. He was candid that this is the goal across the funnel, and even speculated that future "voice agents" could one day handle the sales conversation itself, vendor-to-vendor, with humans only handling exceptions.

There is nothing illegal about automated, behavior-triggered outreach; it is the engine of modern growth. But it underscores the throughline of the whole interview: your dictation is not just transcribed and forgotten. In this model it is measured, attributed to you, scored, pooled, and acted on automatically. Whether that is acceptable depends entirely on what you dictate — and on whether you assumed a "dictation app" was watching that closely.

It would be a mistake to read this as "one company behaving badly." Everything the founder described is a rational, well-executed version of what a cloud dictation business has to do to operate. The product is consumer-facing and free at entry; the business is enterprise sales. The bridge between them is data — knowing who uses the product, how, and where. To build that bridge, the client app must report usage back to the cloud, tied to an identity. Once that telemetry exists, pooling it into an analytics platform, enriching it with company data, and automating outreach are all natural next steps.

This is why we keep returning to the same architectural point across our coverage of Typeless, Superwhisper, and Wispr Flow: privacy is determined by architecture, not by intentions. A privacy policy can promise restraint, but the moment your audio and usage data leave your device, the capability to build the profile the founder described exists — and capabilities, once present, tend to get used, because they create real business value. The only way to make per-user dictation profiling impossible is to ensure the data is never collected. That is a property of where the processing happens, not of how trustworthy the vendor is.

If you dictate routine, low-sensitivity text and you are comfortable being a named row in a vendor's analytics, none of this needs to change your behavior. Wispr Flow is a capable product with real strengths, and behavior-driven sales is legal and common. But if you dictate anything you would not want measured, attributed, and pooled — client matters, patient notes, unreleased code, confidential strategy — the founder's own description is a reason to think about architecture.

Four checks tell you what any dictation app actually does with your voice:

1. Pull the plug. Disconnect from the internet and try to dictate. If it stops working, your audio and usage data are leaving your Mac. (Wispr Flow requires connectivity; Voibe does not.)
2. Read for "analytics." Search the privacy policy and subprocessor list for "analytics providers," "cloud servers," and named third parties. Their presence confirms where your data flows.
3. Check the account requirement. If you must log in to dictate, your activity is tied to an identity — the precondition for the per-user analytics the founder demonstrated. Voibe requires no account.
4. Watch the network. Run Little Snitch or Wireshark while dictating. A genuine on-device tool shows zero outbound traffic during transcription; a cloud tool shows connections to its servers and analytics endpoints.

For the complete version of this framework, see our voice data privacy guide and the eight-point audit in our Typeless privacy investigation. For Wispr Flow specifically, our Is Wispr Flow safe? deep-dive covers its cloud architecture, Privacy Mode defaults, subprocessors, and the 2026 Delve compliance episode.

If the podcast made you reconsider where your dictation goes, the fix is not a different cloud app with a nicer privacy policy — it is a dictation app that never collects the data in the first place. Voibe is a Mac-native dictation app built around one architectural rule: your audio and your usage never leave the device.

Voibe runs OpenAI Whisper models on Apple Silicon's Neural Engine. Press your hotkey, speak, release — the audio is transcribed locally, written into your active text field, and discarded. Mapped against the revelations in this article:

• No per-user analytics engine. Voibe does not report word counts, usage trends, or per-user activity to a server. There is no "strong analytics engine" because there is no telemetry stream.
• No record of which apps you dictate into. The app you are dictating into never gets reported anywhere. Your workday is not fingerprinted.
• No identity to attribute. Voibe requires no account. There is no "who are these users" query because there are no named user rows.
• No third-party analytics platform. Nothing is pooled into Hex or any equivalent, because nothing is transmitted.
• Nothing to de-anonymize. No account, no telemetry, no cross-context profile.

Pricing is $7.50/month or $149 lifetime (also $59/year) for unlimited dictation on Apple Silicon Macs (M1 through M4). Over three years, the $149 lifetime license runs roughly $283–$535 cheaper than Wispr Flow Pro at $144–$228/year — and the savings are the smaller story. The larger one is that there is no business model that depends on measuring how you dictate. Voibe also ships a Developer Mode for Cursor and VS Code with file and folder name resolution — a feature actively requested by Wispr Flow and Superwhisper users.

Try Voibe for free — install, grant microphone and accessibility permissions, and dictate. No account, no credit card, no audio or usage data leaving your Mac.

To his credit, Tanay Kothari was transparent. He did not hide the analytics behind his sales numbers — he opened the dashboard on camera and walked through it. That candor is exactly what makes the episode useful: it is a rare, first-person account of how much a cloud dictation product knows about its users, told by the person who built it. Word counts per user, which apps you dictate into, weekly trends, your name and title and employer, pooled into a third-party platform and wired into automated outreach — all of it presented not as surveillance but as good business.

That is the honest version of the cloud dictation bargain. You get a polished, AI-enhanced product, and in exchange your dictation becomes measurable, attributable, and actionable data. For a lot of routine writing, that trade is fine. For sensitive work — or simply if you would rather not be profiled by how you type with your voice — the only durable answer is architectural: keep the processing, and the data, on your device. A profile can't be built from data that was never collected.

Keep reading: our Is Wispr Flow safe? investigation, our full Wispr Flow review and pricing breakdown, the best on-device Wispr Flow alternatives, our cloud vs. local dictation guide, the best offline dictation apps roundup, and our complete dictation privacy hub.