The new role · Stewardship and ethics

The founder wants 500 posts a month

Find what a 500-posts-a-month spreadsheet leaves out (review, the ramp, the query ceiling) and replace it with a quarter-long test that has a stop rule.

critiquemediumExercise · 3 min · 58 of 60

On this page

Generated posts cost under a dollar, so the founder's spreadsheet shows content replacing paid search within months. The defects are in what the spreadsheet leaves out; this practises brand stewardship of generated output and when content costs nothing.

The situation

The invoicing app is at $3M ARR with a marketing team of three, 528 working hours a month between them at the course's 176 each. It publishes 8 written posts a month at $400 each. Paid search and social cost $50,000 a month and bring 100,000 visits. The founder proposes 500 generated posts a month instead and shares this:

cost      500 posts x $1                          = $500 a month
review    the brand steward samples 5%            = 25 posts, about 2.5 hours
visits    500 posts x 200 visits a month          = 100,000 visits a month
result    the visits paid buys for $50,000, for $500

Assume what the course assumes: a generated post costs under $1 of model cost plus a $50, 30-minute review before it is published, and the team's hour is worth $100. A post reaches 200 visits a month only at maturity, after an 18-month ramp. Keyword research found about 600 search queries the company could plausibly rank for; the 8 written posts a month each target one of them.

The question

Find the defects in the founder's spreadsheet, and say what you would run for the next quarter instead.

A strong answer · Rubric · Where answers go wrong

Write your answer first, then open

A strong answer

Three defects, each enough to sink the plan on its own.

Review is missing. The 5% sample measures the defect rate after publishing; it is not the review. Every published post gets the 30-minute read: 500 × ($1 + $50) = $25,500 a month, and 500 × 0.5 = 250 hours, nearly half the team's 528.

The visits are borrowed from month 18, and the queries run out. 200 visits a month is what a post earns at maturity; in its first quarter it earns a fraction of that. And 600 rankable queries last about five weeks at 500 posts a month. After that, new posts compete with the existing ones for the same queries, so visits per post fall rather than add up, and Google says generating many pages without adding value for users may violate its spam policy (Google Search's guidance on generative AI content on your website (opens in a new tab)), which puts the 8 posts that already rank at risk.

The comparison is wrong. Paid's 100,000 visits arrive this month and stop when the spend stops. Content is compared on cost per visit across the ramp, as content and SEO does it.

Instead, for one quarter: keep the 8 written posts and add 24 generated ones a month, each reviewed and each on an unused query. That is 24 × $51 ≈ $1,200 and 12 hours a month. At month 3, compare the generated posts' visits against the written posts' curve at the same age. Scale to 60 a month if they reach two thirds of it with no severe defect in the steward's sample; stop if the written posts lose rankings.

Rubric

Restores the $50 review on every published post and costs it: $25,500 and 250 hours a month against the team's 528.
Separates the 5% sample, which measures, from the review, which gates.
Shows the query ceiling, about five weeks of posts, and what happens to visits per post after it.
Replaces month-18 visits with the ramp and compares cost per visit, not visits, with paid.
Proposes a test with a size, a comparison against the written posts, and both a scale-up and a stop rule.
Would satisfy an interviewer who asks "what number makes you scale it?" with a figure, not "if it works".

Where answers go wrong

"Generated content is low quality, so keep writing 8 posts." It prices nothing and rules out a test that could pay. A founder hears taste, not arithmetic, and the reviewer marks it down for never naming the number that decides.
"Do it, but sample 10% instead of 5%." It still treats sampling as review: 450 of 500 posts a month are published unread, and the claims in them carry the company's name.
"500 posts will get us cited in AI answers." There is no evidence that volume earns citation, and the module 5 numbers cap the prize: on that site AI surfaces send 120 of 20,000 organic sessions a month, 0.6%, so even tripling them is a few hundred sessions a month.