Keep a Title Experiment Log to Compound Results

A title experiment log is a single file where every test lives: the variants, the hypothesis, the numbers, and the outcome. Without one, each test is an isolated event and the lessons evaporate by the next upload.

I ran mine as a messy spreadsheet for a year before it became the most useful document I own. It is not pretty. It is the reason my titles improved faster than my editing did.

What goes in the log

Six fields cover almost everything: date, video, variants tested, the variable I changed, the result, and one sentence on why I think it happened. That last field is the one that turns a log into a teacher.

I write the why before I look at the next week's data, so the guess stays honest. Later, when the pattern repeats, I can see whether my reasoning held or whether I just got lucky.

A test teaches you something once. A log teaches you the same thing every time you open it.

Log the failures too

• Failed tests map the edges of what your audience ignores, which is as useful as knowing what they like.
• A tie is a result. It means the variable did not move the metric, so stop testing it.
• Record the video it ran on. A title result is only meaningful next to the content it described.
• Note the traffic source if you can. A title that wins on browse may lose on search.

Look for patterns, not winners

Ten lines of a log matter less than the trend across them. After a quarter, I sort by variable and look for which lever keeps winning: numbers, questions, named subjects, plain claims. The winner list is noisy. The pattern is not.

Sometimes the pattern contradicts my instinct. I was convinced questions performed well, and the log showed they consistently underperformed for my audience. Without the record, I would still be running them out of habit.

Keep the log light enough to maintain

The log dies the moment it becomes a chore. One row per test, plain text, no formulas. I fill it in during the walk from my desk to the kitchen, because a task that takes three minutes is a task that survives.

I also keep a short summary at the top that I rewrite every month: the structures that won and the one that lost. It is the only part anyone else reads, and usually the only part I need.

Compound the lessons

Each logged test is small. Fifty logged tests, read together, form a map of your specific audience that no course can hand you. That compounding is the entire argument for the log.

Start the file today, even with one row. The first entry will embarrass you in a year, and that embarrassment is proof you learned something you would otherwise have forgotten.