05 · Product audit
A careful product with rough edges in places readers rarely see.
I used the-ken.com logged out, on desktop, on 5 October 2026: searching, reading, hitting the paywall, comparing plans, and reading the page source. Most of what I found is small. Together it adds up to a few clear themes: search that loses track of time, newsletter and podcast pages that don't describe themselves to Google or AI, and a site structure that has grown over ten years without being cleaned up.
Search: good at topics, loses track of time
The semantic search launched in March 2026 is a real improvement on keyword search. I tried it with the kind of question Praveen's February post said it should handle.
Query
fintech companies that shut down before covid
Top results, in order:
- Sahay, India's fintech disruption sequel · May 2020
- Bajaj, Razorpay, Zerodha carry the Indian fintech torch · Jan 2020
- Yes Bank's collapse exposes Indian fintech's shoddy plumbing · Mar 2020
- Covid-19's impact on startups: Black Swan or something else? · Mar 2020
- An audit reckoning—the $22B Byju's juggernaut… · Jun 2022
- Navigating CRED's existential crisis · May 2020
- Covid-19 casts businesses into three concentric rings · Apr 2020
- For Zerodha, Covid is a ladder · Apr 2020
The results are on topic (fintech, Covid) but ignore "before" and "shut down". None of the top eight is about a fintech that closed, three aren't about fintech at all, and most are from early in the pandemic rather than before it.
- What's missing is a query-understanding step. "Before covid" is a date filter (before March 2020). "Shut down" is an event type. "Stories by Arundhati in 2025", the example from The Ken's own post on search, is an author filter plus a year. A small parser in front of the vector search turns those into filters, and the semantic part then ranks within them.
- Results can't be shared or bookmarked. Pressing Enter changes the address to /?query=…, but opening that address directly loads the homepage. The "Popular searches" links in the Explore panel use /?s=…, which, in my tests, opened a results page with the query filled in and no results.
- Search is where people decide to subscribe. That's Praveen's own observation: people search for their employer. A shareable results page ("Everything The Ken has written about Swiggy") is something a reader can send to a colleague, which is how corporate deals start.
Stories describe themselves. Newsletters and podcasts don't.
I checked the raw HTML of 67 stories (Feb–Oct 2026) and 7 newsletter, podcast and column pages from this week.
| Check | What I found | Why it matters |
|---|---|---|
| JSON-LD on newsletter and podcast pages | None. Story pages have it; Two by Two, Make India Competitive Again, Long and Short and the rest don't | Most of what The Ken now publishes, including the Swiggy and Air India pieces from the AI test, has no machine-readable author, date or paywall. Google asks paywalled publishers to mark the paywalled section |
| What story markup calls free | The markup declares the first paragraph free (about 28 words). Logged-out readers actually see about 320 words, roughly 13% of a 10-minute story | Worth aligning, so the markup describes what readers actually see |
| Published time | 2026-10-02 01:28:14, with no timezone and not in ISO 8601 format | Machines may read it as UTC, making stories appear 5.5 hours off |
| Modified time | 2026-10-01 19:48:55 on the same page, which is before the published time | Probably IST and UTC mixed together. Freshness signals become unreliable |
| og:type | Article, capitalised, and og:title and og:image each appear twice | Minor |
| Canonical, Twitter card, og:image | Present and correct on all 67 stories | Good |
| Headings | Five <h1> tags on every story page | Minor, but screen readers and parsers expect one |
| llms.txt | Missing (404) | A cheap way to tell AI engines what is free, what is paid, and what each product is |
The publishing engine, from the sitemaps
The Ken's sitemaps list each content type separately, which shows how output has shifted between products over the years. The dates are last-modified dates, so they are an upper bound on when things were published.
| What The Ken published | 2025 | 2026 to date |
|---|---|---|
| Stories | 259 | 207 |
| Newsletter editions (Ka-Ching!, Trade Tricks, First Principles, Two by Two, Long and Short, MICA, The Collection, Zero Shot, 90,000 Hours, The Nutgraf and others) | ~460 | ~309 |
| Daybreak episodes | 249 | 194 |
| Other podcast episodes, excluding retired shows | ~179 | ~92 |
| Morning emails | 328 | ~273 |
Stories are from the RSS feed by publication date. Everything else comes from the sitemaps by last-modified date, so those are upper bounds.
Three things the output shows
- The Ken publishes far more than one story a day. In 2025 that was roughly 1,500 pieces across formats, including morning emails, or about 6 every weekday. The reader who wrote that it's "becoming very hard to keep up" was describing a real change.
- Large companies feature more than startups. In a 67-story 2026 sample, 69% involved listed or large companies and 43% startups (a story can be both). The most-used tags of the past year are ai (32) and ipo (25). So its rivals include ET Prime and Mint, not just the startup publications.
- 93% of stories come out at 8:00 IST. That's a strong habit to build on. It is also the one fixed moment when a renewal reminder, a "your company was mentioned" alert or a morning audio version can reach readers who already expect something from The Ken.
Pages, paywall and pricing
Small inconsistencies that a careful reader, who is exactly The Ken's customer, will notice.
A recommendation card shows another story's summary
Under "Recommended for you", the card for Wealth managers love job hopping. An upstart wants to fix that showed the summary of a different story, the JS Deepak memoir piece. The cards are probably filled from a misaligned list.
"Top Subscriber Comments" cycles through two comments
The carousel repeats the same two comments three times. With a subscriber-only community and comments on most stories, there's more to show.
Three different device limits
The pricing page, the story paywall and the FAQ each describe Premium's device limit differently: 2 browsers plus 2 apps, "Upto three devices", and 2 web plus 1 phone plus 1 tablet.
"No discounts", next to a 50% student discount
The FAQ says "Sorry, no" to discounts. The same page offers 50% off for students, and alumni pages offer 10%. The testimonials also contain two typos: "in-dept" and "saying relevant".
A sales pitch before the first paragraph
A Two by Two edition opens with "This year, Praveen is doing a select few speaking engagements…" before the story starts. Speaking is a real revenue line, but placing it there costs some reading focus.
The free-trial funnel
"Story unlocked just for you", with a countdown timer, an email sign-up that unlocks one subscriber story a day for 30 days, and Google sign-in. It's clear and generous, and built with care.
Ten years of taxonomy
The site's structure has grown with each product launch. That's normal. It matters now because search, recommendations and company pages all depend on it.
The company list is out of date
The Explore panel's 47 companies lead with Byju's and include Dunzo, plus Hotstar, Facebook and Sequoia Capital (now Peak XV in India) under their old names, plus GoJek, Grab and Shopee from the closed Southeast Asia edition. Duplicates: Jio and Reliance Jio, Reliance and Reliance Industries. Airtel appears as both Airtel and Bharti Airtel in the footer.
Two spellings for many URLs
/kaching/ and /ka-ching/, /tradetricks/ and /trade-tricks/, /newsletter/ and /newsletters/, /long_and_short/ and /long-and-short/. Story URLs and section URLs follow different conventions.
The same topic under several tags
Stories average 10 tags each. The past year has ipo and initial public offering, hdfc and hdfc bank, evs and electric vehicles, jio and reliance jio, airtel and bharti airtel, and five separate education tags. Each split weakens recommendations and "more on this" modules.
Small things a crawler notices
A comment link uses #commment-, with three m's. A story slug ends …-without-npcil while its headline doesn't. Robots.txt lists GPTBot three times.
The stack, from outside
| Layer | What's visible |
|---|---|
| CMS | WordPress with a custom theme and plugins (customisation, corporate subscription, tokenisation) |
| Edge | Cloudflare (email obfuscation, content signals in robots.txt) |
| Analytics | Google Tag Manager and Mixpanel |
| Front end | jQuery 3.7, Owl Carousel, self-hosted fonts (Reckless, Ivar Text, Archivo) |
| Apps | React Native, Android 4.1.31 (30 Sep 2026), iOS and iPad |
| Search and recommendations | In-house semantic search and hybrid recommendations on The Ken's own servers |
A detail from 2018. In its early years The Ken told subscribers it was deleting their data, collecting only a name and an email, and running "zero trackers". Today's pages load Google Tag Manager and Mixpanel. That's a sensible trade for a company that needs to understand reading behaviour, and it's also the data a renewal radar would be built on (project 1).
A short "what we measure and why" page would carry that 2018 privacy stance forward, and suits a brand built on trust.
The fix list, ordered by effort
| Fix | Effort | Effect |
|---|---|---|
| Make /?query= URLs load results, and point Popular searches at them | Hours | Search results become shareable |
| Put one device limit on every surface. Correct the discount FAQ and testimonial typos. Remove The Nutgraf from plan descriptions and Popular searches | Hours | Trust at the moment of payment |
| ISO 8601 timestamps with timezone; lowercase og:type; remove duplicate tags | Hours | Clean freshness signals |
| Extend the story-page NewsArticle and paywall markup to newsletter and podcast templates | A day | Paywall compliance, rich results, clearer signals to AI engines |
| Fix the recommendation card mismatch; show more subscriber comments | A day | Homepage credibility |
| Refresh the Explore company list from what readers actually search for | Days | Discovery for new visitors |
| Query understanding (dates, authors, companies) in front of semantic search, with a test set | Weeks | Search that answers real questions |
| Clean up company tags across the archive | Weeks | The basis for company pages and corporate sales |