# Tapesearch

Aiolum story: https://aiolum.com/story/tapesearch/ · added 2026-10-26 · updated 2026-10-26

**In short.** Tapesearch (Search engine over AI-generated podcast transcripts, with alerts, AI chat, an API and an MCP server, United Kingdom): US$5K verified revenue in one month, September 2026. Evidence grade A: disclosed by a company or a filing, or verified by a payment processor. Team: 1 founder, Anjum Sayed, a data scientist; TrustMRR lists a team size of 1 person. Launched 9 January 2023; still running, revenue growing.

Search engine over AI-generated podcast transcripts, with alerts, AI chat, an API and an MCP server · United Kingdom · AI-native · solo founder · still running, revenue growing · group: Voice and audio

- Outcome: US$5K verified revenue in one month, September 2026 (evidence grade of the headline number: A; headline in US dollars: 60036)
- Launch: 9 January 2023; months from launch to the outcome: 44
- Team: 1 founder, Anjum Sayed, a data scientist; TrustMRR lists a team size of 1 person
- Outside funding: None (TrustMRR: bootstrapped)
- Main factor: Distribution
- Meters, 1 to 4: evidence strength 3, copyable pattern 2, capital needed 2
- Tags: AI-native, Solo founder, Bootstrapped, Running, B2B, Podcasts

**The product.** Tapesearch downloads new episodes of popular public podcasts every day, turns the audio into text with a speech recognition model, and makes the text searchable. A user types a phrase, sees where it was said, and jumps to that second of the audio. Paid users also download timestamped transcripts, get email alerts when a keyword is said, chat with episodes through a language model, chart how often a topic comes up, and query the data through an API or an MCP server for AI assistants.

**Why it worked.** Speech recognition became cheap and accurate enough for one person to transcribe millions of podcast episodes. The founder sells the result to people who need to know what was said: market researchers, advertisers, analysts and journalists. The headline number comes from Stripe through TrustMRR: US$5,003 of revenue in September 2026.

Evidence grades: A = disclosed by a public company, a filing or a payment processor, or verified by one; B = two or more independent reports agree, or a journalist confirmed it; C = a founder claim that no one else verified; D = reported, not verified: one founder claim or a third-party estimate with no second source; calc = computed on this page from graded facts.

## Summary

Tapesearch is a search engine for spoken words in podcasts. It runs automatic speech recognition, a model that turns recorded speech into written text, over new podcast episodes every day, and stores each transcript, the written text of an episode, with a timestamp for every line, the time in the recording where the line starts. A search for a brand or a name returns the exact moments where it was said, with a link to that point in the audio. The home page said in October 2026 that the database held "5 million and counting" transcripts, up from "2 million and counting" in July 2024.

The founder is Anjum Sayed, a data scientist in the United Kingdom. Their GitHub page lists the title Kaggle Competitions Grandmaster and says "I'm currently building Tapesearch". They posted the product on Hacker News on 9 January 2023. TrustMRR, a site that reads revenue directly from a company's payment processor, shows Stripe revenue of US$3,732 in November 2025 and US$5,003 in September 2026, US$97,658 since the start, and 184 active subscriptions. TrustMRR lists a team of 1 person and no outside funding.

This page grades every number by its evidence. The monthly revenue is grade A because TrustMRR reads it from Stripe. The user and transcript counts are the company's own claims. The page also describes the legal basis that Tapesearch states for publishing transcripts of other people's podcasts, and shows four failed search companies from the Loot Drop database.

## Pricing

Tapesearch plans read on the home page on 26 October 2026 (grade C)

| Plan | Per month, billed yearly (US$) | Per month, billed monthly (US$) | Limit |
|---|---|---|---|
| Free | 0 | 0 | Search; full transcripts of new episodes after a waiting period |
| Pro | 16.67 | 17 | Transcript search with filters, AI chat, transcript downloads (TXT, SRT, VTT, JSON), 2 email alerts |
| Insights | 30.83 | 31 | Everything in Pro, up to 50 alerts, trend charts with downloadable data |
| API Access | 60 | 60 | Everything in Insights, API search and transcript download, up to 2,000 requests a day, hosted MCP server |

Yearly billing gives 2 months free. The per-month yearly price is calculated: Pro costs US$200 a year, Insights US$370 a year and API Access US$720 a year. TrustMRR lists Pro at US$16.66 a month, the yearly price. A free account can search, but an episode page read on 25 October 2026 said that its full transcript would reach the free plan only after a waiting period ("in 16 days").

## How much revenue does Tapesearch make? The numbers, graded

US$5,003 of Stripe revenue in September 2026 from a podcast-transcript search engine that one data scientist launched on Hacker News in January 2023

| Fact | Value | Grade | Note | Sources |
|---|---|---|---|---|
| Verified revenue, September 2026 | US$5,003 | A | TrustMRR monthly revenue timeline, read from Stripe through an API key, last synced 27 October 2026. October 2026 had US$5,001 by 27 October. | [1] |
| Verified revenue, last 30 days to 27 October 2026 | US$5,942 | A | TrustMRR, with growth of +25.3% against the 30 days before. | [1] |
| Verified revenue, last 12 months | US$53,358 | A | TrustMRR. | [1] |
| Verified revenue, all time | US$97,658 | A | TrustMRR. Stripe is the verified payment provider; TrustMRR shows no merchant of record. | [1] |
| Verified revenue, November 2025 | US$3,732 | A | The first month in the public TrustMRR timeline. Other months: US$4,932 in March 2026, US$5,093 in April 2026, US$3,841 in August 2026. | [1] |
| Change from November 2025 to September 2026 | +34% | calc | Calculated: 5,003 divided by 3,732 equals 1.34. | [1] |
| Run rate from the September 2026 month | about US$60,000 a year | calc | Calculated: 5,003 times 12 equals 60,036. | [1] |
| Current MRR and active subscriptions, October 2026 | US$4,251, 184 subscriptions | A | TrustMRR snapshot. Monthly recurring revenue (MRR) counts subscriptions only and spreads yearly plans over 12 months, so it is lower than the cash received in a month. | [1] |
| Average MRR per active subscription | about US$23 a month | calc | Calculated: 4,251 divided by 184 equals 23.1. | [1] |
| Profit margin, last 30 days | 80% | C | TrustMRR snapshot. Costs are not read from Stripe, so this is graded as a founder figure. | [1] |
| Team size and funding | 1 person, bootstrapped | C | TrustMRR profile text, written by the founder. | [1] |
| Transcripts in the database | 2 million (July 2024), 5 million (October 2026) | C | Home page, Wayback copy of 5 July 2024 and live page. | [9, 2] |
| Users | 3,400+ (July 2024), 17,000+ (October 2026) | C | Home page: "Trusted by 3,400+ listeners" in July 2024 and "Trusted by 17,000+ listeners" in October 2026. These are accounts, not paying customers. | [9, 2] |
| Speech recognition model | distil-whisper/distil-large-v3 | C | Named in the data of an episode page ("transcript_model"). Distil-Whisper is a smaller, faster version of OpenAI's open Whisper model. | [3] |
| Model that answers chat questions | Google Gemini | C | Privacy policy, updated 23 September 2026. | [4] |
| Time from the Show HN post to the September 2026 figure | 44 months | calc | Calculated: January 2023 to September 2026. TrustMRR gives a founded date of 1 November 2022. | [1, 8] |

How to read this table: a grade A value was read by TrustMRR from the company's Stripe account. A grade C value was written by the founder or on the company's own pages. A "calc" value was computed on this page from the rows above it.

## When did Tapesearch launch, and how did it grow? Timeline

From a Show HN post in January 2023 to US$5,003 of verified revenue in September 2026, 44 months later. Each entry carries its grade.

- **1 November 2022** (grade C [1]): Founded date given on the TrustMRR profile.
- **9 January 2023** (grade C [8]): Anjum Sayed posts "Show HN: I made a search engine that lets you find what was *said* in a podcast" with a link to tapesearch.com.
- **5 July 2024** (grade C [9]): The home page says 2 million transcripts and 3,400+ users. Two paid plans cost US$15 and US$37 a month, with 2 months free on yearly billing.
- **7 September 2024** (grade C [5]): Last update of the terms of service, which forbid scraping data from Tapesearch.
- **November 2025** (grade A [1]): US$3,732 of verified revenue, the first month in the public TrustMRR timeline.
- **April 2026** (grade A [1]): US$5,093 of verified revenue, the best month in the public timeline.
- **June to August 2026** (grade A [1]): Revenue falls to between US$3,841 and US$4,129 a month.
- **September 2026** (grade A [1, 4]): US$5,003 of verified revenue. The privacy policy, updated on 23 September 2026, lists Auth0, Stripe, Google Cloud with Gemini, Cloudflare and Akamai (Linode).
- **October 2026** (grade C [2, 7, 1]): The home page claims 5 million transcripts and 17,000+ users, and sells an API plan with a hosted MCP server for Claude and other AI assistants. TrustMRR shows 184 active subscriptions.

## Why did Tapesearch work?

### Cheap transcription at scale

Open speech recognition models such as Whisper, and smaller versions such as Distil-Whisper, made it possible to transcribe millions of hours of audio on rented servers. An episode page shows the model name distil-whisper/distil-large-v3. The work is a pipeline: fetch the RSS feeds, download the audio, transcribe it, index the text, and repeat every day at midnight UTC.

### A dataset that grows every day

Each day adds new transcripts, so the product gets more useful without new features. The home page went from 2 million transcripts in July 2024 to 5 million in October 2026. Podcasts that use the Podping notification system are transcribed within minutes, the FAQ says, so alerts arrive soon after a new episode.

### Buyers who pay to know what was said

The home page aims at market researchers, financial analysts, advertisers, academics and journalists. For these buyers, an alert when a brand is mentioned, or a list of podcasts that sell ad slots to a competitor, is worth more than a consumer subscription, so the highest plan is the API plan at US$60 a month.

### Selling to AI assistants too

The API plan includes a hosted MCP server. The Model Context Protocol (MCP) is a standard that lets an AI assistant such as Claude call an outside service. A user can ask Claude to chart how often a word was said in podcasts, and Claude queries Tapesearch.

## How it was built

These details come from the company's own pages (privacy policy, FAQ, API and MCP pages, page data) and the TrustMRR profile. They are grade C, except Stripe, which TrustMRR confirms (grade A).

- Transcription: a Distil-Whisper model (distil-whisper/distil-large-v3, named in an episode page's data). The FAQ says that for a bad transcript "we can try a different model".
- Which podcasts: the most popular podcasts by the number of Apple Podcasts ratings, plus podcasts that subscribers request. Paywalled shows, such as exclusives on Spotify, are not included.
- Chat and summaries: questions go to Google's Gemini model on Google Cloud, which also stores files and sends service emails, says the privacy policy.
- Servers and databases on Akamai (Linode), the website on Cloudflare, sign-in through Auth0 (Okta), payments through Stripe. TrustMRR lists React for the front end and Python for the back end; the page source shows Next.js, a React framework, and the Crisp chat box.
- An API at api.tapesearch.com with public documentation, and an MCP server at mcp.tapesearch.com for AI assistants, both on the US$60 plan.
- Marketing: TrustMRR lists Google Ads as the marketing channel. The privacy policy names Google Tag Manager and the Meta (Facebook) Pixel for measuring visits and sign-ups. Every podcast and episode has its own public page, which search engines can index.

Sources: [3, 2, 4, 1, 6, 7]

## Who tried this before Tapesearch, and why did they fail?

Four companies in the Loot Drop database tried to make a large body of content searchable, or to answer questions by voice, before cheap speech recognition and language models existed. The values below are from that database and are not independently verified.

The failed startups come from Loot Drop (https://www.loot-drop.io/), a free database of dead startups, retold here in our own words; Loot Drop describes its entries as AI-assisted summaries of public sources, so they carry no evidence grade.

### Topsy (2007 to 2015; funding 30 (as recorded in the database; the unit is not given); cause: Competition)

- Promise: Real-time search and analytics over tweets and other social media, to see what people say about a topic.
- Why it died: Apple bought it for its technology and shut it down, while its dependence on Twitter's data and API made it hard to stand alone.
- Record: https://www.loot-drop.io/startup/650-topsy

### Tellme (1999 to 2011; funding US$238M; cause: Competition)

- Promise: Information and transactions over the phone with voice commands, through automatic speech recognition.
- Why it died: Smartphone apps and newer voice assistants replaced phone-based voice services.
- Record: https://www.loot-drop.io/startup/612-tellme

### Quixey (2009 to 2017; funding US$134M; cause: Competition)

- Promise: Search for mobile apps by what they do, not by their names.
- Why it died: Google was stronger in search, and Quixey could not earn money fast enough before partners such as Alibaba left.
- Record: https://www.loot-drop.io/startup/478-quixey

### Neeva (2019 to 2023; funding US$77M; cause: No Market Need)

- Promise: A web search engine without ads, paid by subscription.
- Why it died: Few people would pay for general search, and ChatGPT changed what people expected from it.
- Record: https://www.loot-drop.io/startup/2288-neeva

## What changed

Tellme spent US$238 million on speech recognition for phone calls, when recognizing speech was expensive and limited to short commands. In 2023 an open model could transcribe a full hour of conversation with good accuracy on one rented machine, so one person could transcribe millions of episodes. Tapesearch does not need to understand commands; it needs cheap, accurate text from long recordings, which is what the Whisper family of models made possible.

Topsy, Quixey and Neeva show the limit of search as a business: a general search product competes with Google and depends on other companies' data. Tapesearch avoids general search. It sells a narrow dataset, spoken words in podcasts, to professional buyers who pay for alerts and an API. Topsy also shows the same dependence that Tapesearch has: it relied on Twitter's data, as Tapesearch relies on podcast feeds that it does not own.

## Can one person copy Tapesearch now?

**Verdict: The pattern, yes. The podcast index itself, only with care about the rights.**

The method is open to any engineer: pick a large body of public audio or video that nobody has made searchable, transcribe it with an open speech model, and sell search, alerts and an API to professionals. The hard parts are the running cost of transcription, the size of the index, and the legal question of publishing text made from other people's recordings.

1. Pick a body of public spoken content with professional buyers: podcasts, public meetings, earnings calls, conference talks.
2. Transcribe it with an open speech recognition model on rented servers, and refresh it every day.
3. Give each item a public page, so search engines index it and bring free visitors.
4. Sell alerts, downloads and an API on monthly plans, and add an MCP server so AI assistants can use the data.
5. Read the copyright position of the content before you publish full transcripts, and offer owners a way to ask for removal.

**What is still open.** Spoken content in languages other than English, local radio, YouTube channels in one industry, and public video meetings of city councils are still poorly searchable. Each needs its own check of the rights.

## What could go wrong for Tapesearch?

- **Legal basis: other people's podcasts under "Fair Use".** Tapesearch transcribes podcasts that it does not own and publishes the transcripts on public episode pages, sells their download and serves them through an API. The only legal basis its pages state is a disclaimer on each episode page, for example on an episode of The Daily: "Generated transcripts are the property of The New York Times and are distributed freely under the Fair Use doctrine." Fair use is a defence in United States copyright law; the company is in the United Kingdom. No source that we read shows permission from podcast owners, a licence, or a removal process, and no source reports a dispute. The same company's terms of service forbid others to scrape Tapesearch's own data.
- **Revenue went down before it went up.** Verified revenue was US$5,093 in April 2026, fell to between US$3,841 and US$4,129 a month from June to August 2026, and returned to US$5,003 in September. The trend is growing against the summer months and flat against the spring.
- **Cash revenue is higher than MRR.** TrustMRR shows US$5,942 received in the last 30 days but US$4,251 of MRR. Yearly plans paid at once and one-time payments make the cash months uneven. The 184 active subscriptions are the more stable measure.
- **Depends on model and cloud suppliers.** Transcription uses an open model, but chat answers come from Google's Gemini, and the servers run on Akamai (Linode) and Cloudflare. A price change at any of them changes the margin, which TrustMRR lists at 80% for the last 30 days.
- **User counts are the company's own.** The 17,000+ users and 5 million transcripts are claims on the home page. Only the revenue and the subscription count are read from Stripe.

## Sources

1. TrustMRR: Tapesearch profile (markdown), revenue last synced 27 October 2026 — https://trustmrr.com/startup/tapesearch.md (grade A). Read the markdown profile in full on 26 October 2026. The revenue, MRR and subscription figures come from Stripe through an API key (grade A); the team size, funding, tech stack and marketing channel are founder-written profile text (grade C).
2. Tapesearch home page, with features, pricing and FAQ — https://www.tapesearch.com/ (grade C). Read in full on 26 October 2026. Source for the plans, the user and transcript counts, the podcast selection and the Podping detail.
3. Tapesearch episode page: When A.I. Comes for Your Purpose in Life, The Daily — https://www.tapesearch.com/episode/when-a-i-comes-for-your-purpose-in-life/2b5W8TKD5qWBgRk8WK7E3W (grade C). Read in full on 25 October 2026, including the page data. Source for the Fair Use disclaimer, the free-plan waiting period and the transcription model name.
4. Tapesearch privacy policy, last updated 23 September 2026 — https://www.tapesearch.com/privacy.md (grade C). Read in full (markdown version). Source for the service providers: Auth0, Stripe, Google Cloud and Gemini, Cloudflare, Akamai (Linode), Google Tag Manager and Meta Pixel.
5. Tapesearch terms of service, last updated 7 September 2024 — https://www.tapesearch.com/tos (grade C). Read in full. Source for the ban on scraping Tapesearch's data.
6. Tapesearch API Access page — https://www.tapesearch.com/api/access (grade C). Read in full. Source for the API, its documentation address and an example count of podcasts and transcripts.
7. Tapesearch MCP Server page — https://www.tapesearch.com/api/mcp (grade C). Read in full. Source for the hosted MCP server and its setup in Claude.
8. Hacker News, 9 January 2023: Show HN: I made a search engine that lets you find what was *said* in a podcast (item 34309688), by anjum48 — https://hn.algolia.com/api/v1/items/34309688 (grade C). Read in full through the Algolia API. The post has a link and no text or founder comments. Source for the public launch date.
9. Tapesearch home page, Wayback Machine copy of 5 July 2024 — https://web.archive.org/web/20240705210231/https://www.tapesearch.com/ (grade C). Read in full. Source for the 2 million transcripts, 3,400+ users and the 2024 prices.
10. GitHub profile of Anjum48 (Anjum Sayed) — https://github.com/Anjum48 (grade C). Read in full. Source for the founder's background: data scientist, Kaggle Competitions Grandmaster, "currently building Tapesearch".
11. Loot Drop database export, 1,749 failed startups — https://www.loot-drop.io/ (grade C). Local export. AI-written summaries, not independently verified. Source for the failed predecessors.