How to run a 30 day content experiment and learn from it
Short answer: A 30 day content experiment asks one question, changes one thing, and compares 10 control posts with 10 variant posts on one platform. Write the question first, change only one variable, assign posts to control or variant by coin flip, and read every post at the same age. Claim a lift only if the variant median beats the control median by at least 25 percent and at least 8 of 10 variant posts beat the control median. Otherwise write "no clear difference". A scheduler such as Blotato, a software subscription that schedules posts to social accounts, keeps timing steady.
You change your posts all the time and never learn anything. You want a test small enough to run in a month and honest enough to trust.
I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.
Most "tests" on social media are not tests. You change the hook, the picture and the time in the same week, then credit whichever one you like. A real experiment is smaller. It has one question, one change, a control group and a rule for what counts as a result, written down before you start. This page gives you the design, the arithmetic and a template.
A scheduler helps by posting at the same times every week. I use Blotato as the example, and say when you do not need it.
This page may contain affiliate links from which Blur Studio earns commissions.
Step 1: define a question
A good question names one change and one result. Use this form: "Does [change] raise [result] on [platform]?"
- Does opening with a question raise saves on Instagram carousels?
- Does a text post with a short first line get more impressions than one with a long first line on LinkedIn?
- Do photo slides get more follows than short videos on TikTok?
Bad questions: "What works?" and "Is my content good?" They have no change and no result.
Write down the result you will read, for example impressions or saves. Pick one. Write down the age you will read it at, for example day 7. Both are fixed before the first post.
Step 2: design the test
| Part | Rule |
|---|---|
| Platform | One. Do not mix platforms in one test |
| Control | 10 posts as you normally make them |
| Variant | 10 posts with one change |
| Variable | One only. If you change the hook, keep the format, topic mix and time slot the same |
| Total | 20 posts in 30 days, about 5 a week |
| Assignment | Decide control or variant by coin flip for each topic, not by which one you think is better |
| Timing | Same slots in the week for both groups |
| Reading age | Every post read at the same checkpoint |
Why a coin flip? If you pick the topics you like for the variant, your pick might be what makes it win. A coin does not have favourites.
Why 20 posts? With fewer, one lucky post can decide the result. With 20 you still have a small sample, but you can apply a fair rule. If your channel posts only 2 or 3 times a week, run the test for 8 weeks, not 30 days.
Step 3: run it
- Week 0: write the question, the result, the age and the rule (below) on one page. Plan the 20 posts and flip the coin for each.
- Weeks 1 to 4: post on the planned slots. Do not change the plan mid-test. If something big happens in the news or you launch something, note it and decide later whether to drop that post and replace it.
- Each Friday: read the posts that have reached their reading age and enter them in the sheet. Do not look at the winner until all 20 are in.
- Do not change anything else. New bio, new profile picture, a paid boost or a giveaway during the test spoil the comparison.
Blotato helps with step 2, since it posts at the times you set. Its analytics are collected at set points after publishing, which on the Starter plan are day 1, 7, 14 and 30, and only for posts published through Blotato. Day 7 is a good reading age because every post has one. A reading marked pending is not zero. If your platform is LinkedIn, Blotato has no analytics, so use LinkedIn's own, and enter the same checkpoint for every post. See what Blotato analytics can show.
Step 4: read the result, with a noise rule
Write this rule before the test starts. A variant counts as a real lift only if both are true:
- Size: the variant median is at least 25 percent higher than the control median.
- Consistency: at least 8 of the 10 variant posts beat the control median.
The median is the middle number when you line the results up. Use it rather than the average, because one viral post can lift an average a long way.
Why 8 of 10? If there were no real difference, each variant post would have about a 50 percent chance of beating the control median. The chance of 8 or more out of 10 by luck is (45 + 10 + 1) / 1,024, which is 56 in 1,024, about 5.5 percent. The chance of 7 or more is 176 in 1,024, about 17 percent. That is too often to trust. So 8 is a reasonable bar for a small test, and it is a rule of thumb, not a statistical guarantee. The 25 percent size rule stops a tiny gap from being called a win.
If the variant fails either test, write "no clear difference". That is a result. It means the change did not matter much, and you can stop worrying about it.
A worked example (made-up numbers)
Question: does opening a LinkedIn post with a question raise impressions? Control posts open with a plain statement. Result: impressions at day 7.
| Group | Impressions per post | Median | Average |
|---|---|---|---|
| Control (10) | 360, 380, 390, 410, 430, 440, 450, 470, 520, 600 | 435 | 445 |
| Variant (10) | 390, 440, 450, 470, 480, 505, 520, 530, 560, 610 | 492.5 | 495.5 |
The checks: the control medians use the middle pair, 430 and 440, which is 435. The control total is 4,450, so the average is 445. The variant middle pair is 480 and 505, which is 492.5. Its total is 4,955, so the average is 495.5.
Size rule: 492.5 divided by 435 is 1.132, a 13 percent lift. That is below 25 percent, so it fails. Consistency rule: the variant posts above 435 are 440, 450, 470, 480, 505, 520, 530, 560 and 610. That is 9 of 10, which passes.
The verdict: one rule passed and one failed, so there is no clear lift. The honest report says "a small hint in favour of question openers, not proven". Run a second 30 days with the same design before you change how you write. Even if it held up, a 13 percent lift is too small to justify a big rewrite.
Your experiment sheet
Copy this table. One row per post.
| # | Date | Group (control or variant) | Format | Topic | Slot | Result at day 7 | Notes (anything unusual) |
|---|---|---|---|---|---|---|---|
| 1 | |||||||
| 2 | |||||||
| 3 | |||||||
| 4 |
And the page at the front of the sheet:
| Item | Your entry |
|---|---|
| Question | |
| Platform | |
| Control (what stays as it is) | |
| Variant (the one change) | |
| Result and reading age | |
| Rule (size 25 percent, 8 of 10) | |
| Start and end dates |
Step 5: write the report
Five lines are enough. Keep it short so you actually do it.
- The question.
- What was compared: 10 control, 10 variant, the one change.
- The numbers: both medians, the lift in percent, and the count of variant posts that beat the control median.
- The verdict: real lift, no clear difference, or worse.
- The next step: adopt, repeat or drop.
Next, choose another question. A year of 12 small tests teaches you more than a year of guessing. For ideas, test your post openers, your posting times or your format mix.
Where a scheduler fits, and when not to use one
A scheduler keeps the slots the same and lets you fill 20 posts in one or two sittings. Blotato's queue and calendar do that for many platforms at once. As of November 2026, Blotato Starter is $29 a month billed monthly, with 20 connected accounts and 200 queued posts. Its AI writer can draft variants, but read every one, because the AI may change more than the one thing you meant to change.
You do not need it if you test on one platform with a few posts a week. That platform's own scheduler or a free plan is enough. Buffer, a social scheduling tool, shows a free plan on its pricing page with up to 3 channels and 10 scheduled posts per channel. You also do not need a tool if you post by hand and your analytics come from the platform.
Who should skip this
- Fewer than about 3 posts a week on the platform. Run a longer test, or do something else with your time.
- An audience of a few hundred where one post can double your numbers by chance. Test with 8 weeks and more posts.
- A time when you are launching a product or running ads. Those distort the test.
How this page was checked
The Blotato facts come from its pricing page and its help pages on analytics and social accounts, and the Buffer fact comes from its pricing page. They were last checked on 21 November 2026 and are refreshed every 90 days. The method and all example numbers are my own and made up, and the 25 percent and 8 of 10 thresholds are rules of thumb. Blur Studio has not run Blotato itself, so use its 7 day trial to test the scheduling steps before you pay.
How I'd approach this
I would run one test on your main platform. In week zero, write the question and the rule on one page, and flip the coin for 20 posts. In weeks one to four, post on the plan, enter results every Friday and leave the plan alone. In week five, read the result with the two rules and write the five line report. If the result is "no clear difference", pick another question. For help choosing the right first question, try the free tools at /free or book a free AI brainstorm call.
Questions people ask
How many posts do I need for a social media experiment?
I suggest 20, with 10 control posts and 10 variant posts. That is still a small sample, so use a clear rule, such as a 25 percent lift and 8 of 10 variant posts beating the control median.
What is a control?
Posts made the way you normally make them, so you have something to compare against. Without a control, you cannot tell a change from a lucky month.
Why only one change at a time?
If you change the hook and the picture together, you cannot tell which one mattered. One change, one question.
What if the test shows no difference?
Write "no clear difference". That is a real result. It means the change did not matter much, and you can spend your effort elsewhere.
How long should I wait before reading a post?
Pick one age for all posts, such as day 7. Blotato collects analytics at day 1, 7, 14 and 30 on the Starter plan, so day 7 works for every post. A reading marked pending is not zero.
Can I run this on LinkedIn through Blotato?
Yes for posting, but Blotato has no LinkedIn analytics, so read results in LinkedIn itself. See what Blotato cannot do on LinkedIn.
Do I need a scheduler?
No. It helps keep slots the same. A platform's own scheduler or a free plan is enough for a one platform test.
More in this topic: Social media: growth, calendars and routines for solo owners (36 guides) →