Write With AI: $400K ARR From 114,000 Readers, With Under 2% Paying

114,000 readers. $400,000 in ARR from subscriptions alone. Read the numbers in that order and it looks like the story of someone who built a big list. The picture changes the moment you work backward from how many of those 114,000 actually pay. The published price is $20 a month, so dividing $400,000 across twelve months puts paid subscribers at roughly 1,700. An annual discount would push that up, and the $350 top tier would pull it down. Either way it lands near 2,000, which is under 2% of the list. Lenny’s Newsletter, also covered here, converts 18,000 paid out of 377,000 free, or 4.8%. Write With AI converts at less than half that rate and still clears $400,000. The reason is that subscription revenue is not what carries this business.

The published figures, plotted by date

Nicolas Cole and Dickie Bush run it as a pair, and they publish subscription ARR at each milestone.

Eleven months from $200,000 to $300,000, then five months from $300,000 to $400,000. That looks like acceleration, except the following fifteen months sit at the same $400,000. On Anna David’s podcast “Behind the Book Cover,” published in June 2026, the subscription line is still described as $400,000 a year. A repeated figure is not proof of a flat line, but no new subscription milestone has been published since.

The product mix matters more than the subscription line

Their March 2025 post breaks out what moved ARR, with prices attached.

The first row is the one to sit with. A single campaign selling a $249 course to existing readers produced $90,000, which is 23% of the $400,000 subscription ARR, collected in one push. Stacking monthly subscriptions one reader at a time cannot move that fast.

They missed their own target by a factor of four

If you publish numbers, publish the inconvenient ones too. In that November 2023 post, Bush set a target of $100,000 per month by the end of 2024. The actual October 2024 figure was $300,000 ARR, or $25,000 a month, a quarter of the goal.

That reads less like a stall than like subscription revenue growing more linearly than the founders expected. At $20 a month, $100,000 in monthly revenue requires 5,000 paying readers. A list of 114,000 converting at around 2% does not get there. Run subscriptions alone and the growth rate of the list becomes the ceiling on revenue.

Two percent pay, and the other ninety-eight are monetized by products

Pulling the figures into a single reading: paid subscription works less as revenue than as the mechanism that keeps a list alive while sorting out who will pay.

Behind the Book Cover puts a $1,000,000 to $2,000,000 business on top of the $400,000 subscription line, which makes subscriptions roughly a fifth to a third of the whole. The 98% who never pay do not leave. They keep receiving one free issue a week. That is precisely why a $249 course could produce $90,000 in one campaign: the shot is aimed at all 114,000, not at the 2,000. In this model a low conversion rate is the premise, not a defect.

A third of the inflow happens inside Substack

Acquisition is published too. As of March 2025, over 35% of all subscribers arrive through Substack’s own surfaces, and Notes brings in around 1,000 a week. The October 2024 post puts 20% of paid subscribers on Substack’s recommendation engine.

The publishing pattern is two emails a week, one free and one paid. The paid issue places its paywall 25% to 30% of the way into the piece, and free subscribers receive excerpts of it. In November 2023, producing both was described as taking under four hours a week.

It is a strong setup and also a dependency. If a third of the list arrives through in-platform recommendation, a change to that algorithm hits a third of the inflow. Nobody running the newsletter controls that variable.

When two issues a week, a 23% product and 35% in-platform traffic hold up

• Paywalling one of two weekly emails works only if you can keep shipping a free issue and a paid issue in the same week. This pair reports doing both in under four hours.
• A one-off course returning 23% of annual subscription revenue works only once the list has grown to six figures, non-payers included. Run the same campaign against a small list and the result scales down with it.
• Pulling 35% of inflow from inside a platform works only if your niche puts you on the receiving end of recommendations. That slot cannot be bought, and the operator does not control it.

The limits deserve stating plainly. Both founders had audiences, courses and client work before this newsletter existed, so 114,000 was not built from zero. The $1,000,000 to $2,000,000 outside subscriptions comes from a podcast rather than their own disclosure, and no breakdown accompanies it. The ARR series consists of milestones they chose to publish, with the months in between unknown. The $90,000 campaign is one disclosed instance, and whether the same course repeated that figure a year later has not been published.

How wide the same model runs

Sorting the paid newsletters covered here by monthly revenue produces a four-order spread.

• ¥1,950 a month: a newsletter that reached $13 in paid MRR and switched to sponsorship. Two thousand free readers and sixty issues, and paid revenue stopped at $13.
• ¥5,000,000 a month: Write With AI, the case here. 114,000 readers, under 2% paying, topped up by course drops.
• ¥25,000,000 a month: Lenny’s Newsletter. 377,000 free and 18,000 paid, converting at 4.8%.

List size is not the only thing separating them. Lenny’s puts a subscriber-only Slack community at the centre of the offer and attacks the conversion rate directly. Write With AI leaves conversion low and makes up the difference in the number of one-off launches. Same category, different place to collect the money.

Related cases

• Substack Writers at Work: 2,000 subscribers in six months after two years of none. Selling the craft of growing a newsletter through a three-tier subscription.
• Lenny’s Newsletter: $2M+ a year from 377,000 free and 18,000 paid readers. The same model, tuned toward raising conversion instead.