A long-time small business client sends me a month of blog posts at a time. I turn them into WordPress drafts, one for each Monday. For nine months I did that by hand. For the last three, Claude did the work and I checked it. I kept the billed time for all twelve months, so here's what the client's blog posts cost before and after.

What Each Batch Used to Take

Getting the client's blog posts live started with one document and one image a week. I opened every document and pasted it into a new WordPress post. I changed bolded lines into real H2 headings. Then I filled in the Rank Math fields: focus keyword, SEO title and meta description. (Rank Math is an SEO plugin. Here's Rank Math versus Yoast if you're choosing one.)

Each image arrived as a PNG much larger than it needed to be. I converted it to a JPEG at 80 percent quality, which cut the file size a lot and still looked good. Then I set each post up as a draft for its week.

I did almost no linking. The most I added was a link from "contact" mentions to the contact page. More links would have meant charging more.

What Claude Does Now

Claude does the same work on the client's blog posts and produces the same output, Rank Math fields included. It also looks for places to add internal and external links. This client never got those before. Then I do a quality check to make sure it was done right.

Lately I queue the whole monthly batch as an autonomous run in the early morning hours, so my part is mostly that check. (Here is how those overnight runs work.)

Twelve Months of Billed Time

These are billed minutes from my TimeCamp time records, rounded to the nearest 5 minutes. Post counts come from the published posts, one per Monday.

By hand, it averaged 11.3 minutes per post. With Claude, it averages 7.3. That's about 35 percent less, and October alone was about 45 percent less. At $60 an hour, a four-post month went from roughly $45 to roughly $29, and the client also got links.

I Expected Half. The Records Say a Third.

I thought Claude would cut the time about in half. The records show about a third so far, trending toward half. I'm not going to round that up.

It doesn't always go cleanly, either. On October 21, 2026, the run "ran short because there were several questions that Claude had" for the client to answer. That's from my own time entry. It's a good example of the human step that stays.

The Client Pays Less and Gets More

This is the double-edged sword from my last post, AI Made Me Faster. Hourly Billing Made That a Pay Cut., made concrete. The client pays less and gets more. I bill less for the same deliverable. That post made the argument, and this one shows the numbers on one client. It also covers how I handle the pay cut.

If you publish on a schedule and want to know what a batch like this would cost, tell me what you publish and I'll tell you.

Questions I Expect About This

How much did Claude cut the cost of the client's blog posts?

About 35 percent per post on average, from 11.3 billed minutes to 7.3. At $60 an hour, a four-post month went from roughly $45 to roughly $29.

Does the client get less because Claude is faster?

No. They get the same finished posts, Rank Math fields included, plus internal and external links they never had before. I still do a quality check on the work.

Does Claude run the whole batch on its own?

Mostly. I queue the batch as an early morning run and check it afterward. On October 21, 2026, the run stopped short because Claude had several questions for the client to answer.