Day 198: How I Used AI to Automate a Complete YouTube SEO

Yesterday was the worst single day since we started this entire $240K journey — and I think it's worth pulling apart every number so we can all learn from it.

Key Takeaways

- Sales dropped to just $407 — down from $1,200–$1,300 just two days prior.

- Cost per purchase quadrupled from an average of £10 to £43 in a single day.

- Facebook overspent the daily budget significantly as the algorithm searched for buyers and couldn't find them.

- A partial bounce-back was already visible the following morning, with cost per purchase dropping to around £17.95.

What Actually Happened Yesterday

By mid-morning on the 17th of March, something was clearly wrong. Sales had barely reached $100 by that point, and that kind of slow start is always a signal that the Facebook algorithm has pushed out into the wrong audiences.

By the end of the day, total sales sat at just $407 — the lowest figure we've recorded since starting this program. That alone would have been a tough day, but it was compounded by what happened on the ad spend side.

The Facebook Ad Overspend Problem

Our ad budget is set at £400 per day averaged across the week, meaning some days naturally spend a little more and some a little less. That's normal pacing behaviour from Facebook.

But yesterday was different. The algorithm stretched well beyond the typical variance, spending significantly over budget in an attempt to locate buyers in our audience. When it can't find them easily, it keeps looking — and keeps spending.

The result was a day where we haemorrhaged ad spend without generating the sales to support it.

Cost Per Purchase: The Numbers Don't Lie

The contrast between the 14th of March and the 17th tells the story clearly. On our strong day, cost per purchase was running at £4, £8, and £10 across individual ads. The average was around £10.

Yesterday? Individual ads came in at £43, £44, and £51. The overall average hit £43 — meaning our cost per purchase had effectively quadrupled in a single day. At that level, we were always going to be running at a loss.

Putting It in Context

It's easy to panic when you see a day like this, but it's important to zoom out. On a monthly level, we're likely still in reasonable shape — this will dent the week's figures, but one bad day doesn't erase the progress that's been made.

The algorithm has difficult days. Every advertiser running at scale experiences this. The key is not overreacting — making dramatic budget or creative changes during a bad patch can often make things worse, not better.

The Bounce-Back: Early Signs on Day 197

By the morning of Day 197, there were already signs of recovery. Ad spend was running well below budget at £143, which makes sense — after a big overspend, Facebook pulls back to rebalance the weekly average.

More encouragingly, cost per purchase had dropped back to around £17.95, and sales were already tracking above the full previous day's total by 5:20am. It's not back to our best, but it's a meaningful step in the right direction.

Split Test Update: Closing the Blog Opt-In Test

Separate from the ad drama, we're closing off a split test that's been running on the blog opt-in page. The test was comparing the new 15-minute hook against the original control.

I've already aligned the hook between the blog and the opt-in page so they're congruent, and the data — even at relatively low volume — consistently shows a higher conversion rate on the new version. That's enough to make the call and move on.

YouTube: The Next Optimisation Focus

One area that needs attention is YouTube traffic. We're getting more visits from the channel, but the opt-in rate is noticeably lower than from other traffic sources. There are two stages to fixing this.

First, improve the basic link structure and calls to action within the channel itself. Second — and more excitingly — I'm planning to test AI software to repurpose longer YouTube videos into Shorts. I've been trialling this tool on another project and plan to bring it into this one later this week.

Performance Snapshot

- Sales on worst day (16th March): $407

- Sales two days prior (14th March): ~$1,200–$1,300

- Cost per purchase on 14th: avg ~£10

- Cost per purchase on 16th: avg £43

- Ad spend on 17th (morning): £143 (well below daily average)

- Cost per purchase on 17th (morning): ~£17.95

- Daily ad budget target: £400/day averaged over the week

Closing Reflection

Bad days are part of this. What matters is whether you understand why they happened, whether you've avoided making the wrong decisions in the moment, and whether the system recovers.

Yesterday ticked all three boxes — it was painful, but it wasn't a structural problem.

The early data from Day 197 is more encouraging, and we'll know by end of day whether this was truly a one-off blip or the start of something that needs addressing.

Either way, we keep documenting it all — the good days and the disasters alike.

Resources & Next Steps

Free Top 10 Split-Tests: https://www.jonathanhowkins.com/split-testing

Join the Facebook community: https://www.facebook.com/groups/coursecreatorads

Subscribe on YouTube: https://www.youtube.com/@jonathanhowkins-coursecreator?sub_confirmation=1