Testing Email Subject Lines and Preview Text With AI
You do not need 10,000 subscribers to test a subject line. You need a clear question, a handful of variants, and a way to tell signal from noise. Most small business owners skip subject line testing entirely because they assume it requires the first part they do not have: a big list.
Copper Sun Content and Creative, LLC — the company that builds Brass-SEO — publishes a separate AI marketing platform called Copper Sun AI. Their blog recently ran a piece on testing email subject lines and preview text with AI, and the framework in it is worth curating here because it applies to any small list, with or without their tool. This post breaks down what it says and how to run it yourself.
Quick Navigation
- The Framework, Summarized
- Step 1: Set Your Evaluation Criteria Before You Write Anything
- Step 2: Generate Five to Eight Variants, Not Twenty
- Step 3: Write Preview Text That Adds, Not Repeats
- Step 4: Test on a Segment, Not the Full List
- Step 5: Log Results and Watch for Real Signal
- Why This Matters for a Small List
- How This Connects to Your SEO Content
- Frequently Asked Questions
The Framework, Summarized
The core method from Copper Sun's post is five steps: define your evaluation criteria first, generate five to eight subject line variants, write preview text that complements rather than repeats the subject line, test on a segment instead of your full list, then log which criteria actually predicted the winner. None of these steps require a large list or a dedicated A/B testing platform — a chat window with an AI model and a spreadsheet will run the whole thing.
The order matters more than any individual step. Most people generate subject lines first and figure out how to judge them later. The framework reverses that: criteria before copy.
Step 1: Set Your Evaluation Criteria Before You Write Anything
Before you ask an AI to write subject line options, decide what a winning subject line needs to do for this specific email. Copper Sun's post frames it directly: if you don't know how you'll evaluate the variants, you don't yet know what you're looking for.
For a small business list, that usually breaks into three questions the article recommends applying to every variant:
- Does the subject line reflect what's actually in the email?
- Does it match the audience's current context (what they already know, what they're expecting)?
- Do the subject line and preview text combination together earn the open?
Write these down before you generate anything. A subject line that scores well on curiosity but fails the "reflects what's actually in the email" test will get opens and then get you unsubscribes. That's a bad trade for a small list where every subscriber is a real prospect, not a rounding error.
Step 2: Generate Five to Eight Variants, Not Twenty
Copper Sun's framework recommends five to eight subject line variants as the productive range for most campaigns — fewer than five doesn't cover enough ground, more than eight produces diminishing returns and decision fatigue.
In practice: give an AI model your evaluation criteria from step 1, a one-sentence summary of the email's content, and your audience type. Ask for five to eight subject lines that meet the criteria, not the maximum the model can generate. A prompt built this way, with constraints stated up front, produces options you can actually compare against each other. A prompt that just says "write me some subject lines" produces options with no shared basis for comparison.
Step 3: Write Preview Text That Adds, Not Repeats
Preview text is the second subject line most people ignore, and Copper Sun's article treats it as a distinct variable, not an afterthought. The guidance: preview text should complement the subject line rather than repeat it. If the subject line establishes the topic, the preview text's job is to establish relevance — answering why this email matters right now. If the subject line's job is curiosity, the preview text's job is to add enough specificity to justify the open.
A subject line and a preview text that say the same thing in different words waste the inbox real estate. Two working examples of the pattern:
| Subject Line (topic) | Preview Text (relevance) |
|---|---|
| Your Q3 report is ready | Three numbers moved more than 10% since last quarter |
| We changed how trials work | 3 days instead of 7 — here's what that means for you |
Generate preview text as its own step, after the subject line is chosen, evaluated against the same "does this earn the open" question rather than treated as a caption for the subject line above it.
Step 4: Test on a Segment, Not the Full List
This is the part that makes the framework usable for a small list. Copper Sun's post is explicit that you don't need a full A/B split across your whole audience: send to a segment rather than a full split, and even a 200-person test group produces directional signal on open rate.
For a list under 1,000 subscribers, that means picking your two or three strongest variants from step 2, sending each to a small segment, and watching which one opens better before you commit the winner to the rest of the list. You're not running a statistically rigorous experiment. You're getting a read good enough to make a better call than guessing.
Step 5: Log Results and Watch for Real Signal
After each test, the article recommends logging three things: the winning variant, which evaluation criteria predicted the winner, and whether that pattern held up across multiple sends. A single test tells you what worked once. A log across 10-20 sends tells you what your specific audience actually responds to. Maybe it's curiosity. Maybe it's plain specificity. The pattern only shows up after repetition.
The article also names the failure mode to watch for: don't assume a small gap means anything. A 2-point difference in open rate is noise. A 12-point difference is a signal worth understanding. Chasing every small fluctuation between two subject lines wastes time your list size can't support anyway — save the analysis for gaps large enough to trust.
One more warning worth repeating: subject lines that overpromise inflate open rates temporarily and degrade list health over time. A clever subject line that doesn't match the email's content trains your list to stop opening you, which is the opposite of what a test is supposed to protect against.
Why This Matters for a Small List
Most subject line testing advice assumes scale — the kind of list where a 2% lift is worth thousands of dollars and justifies a dedicated testing platform. A small business list rarely has that volume, so the temptation is to skip testing altogether and write a subject line once, hope, and move on.
The framework above works precisely because it doesn't require scale. Criteria before copy is free. Five to eight variants takes ten minutes with an AI model instead of an hour of staring at a blank subject line field. A 200-person segment test is available to anyone with 200 subscribers to spare, and most small lists have that. The only thing the framework requires that a full A/B platform doesn't is discipline — deciding what "good" means before you start writing.
How This Connects to Your SEO Content
Subject lines and title tags solve the same problem from two different channels: get someone to click on a promise, then deliver on it. If you've read our Brass-SEO email copywriting framework, the overlap should look familiar — the same "does this reflect what's actually here" discipline that keeps a subject line accurate also keeps a title tag from overpromising and getting penalized with a lower click-through rate once readers bounce back to the results page.
Both channels give you real, first-party signal you can act on. Search Console shows you the searches that led to a click as well as the ones that didn't, the same way a subject line test shows you which promise a reader was willing to open. If you already connect your Google Search Console and Google Analytics 4 accounts to Brass-SEO, the same instinct — set criteria, test a few variants, watch for real signal — applies directly to your title tags and meta descriptions. Check your dashboard for pages with high impressions and low click-through rate; those are your subject line tests waiting to happen on the search results page instead of the inbox.
Frequently Asked Questions
Do I need a large email list to test subject lines?
No. Copper Sun's framework tests on a segment rather than a full A/B split — a 200-person test group produces directional signal on open rate. For most small business lists, that means testing your top two or three variants on a portion of subscribers before sending the winner to everyone else, rather than needing thousands of recipients to get a meaningful read.
How many subject line variants should I generate for a test?
Five to eight. Fewer than five doesn't give you enough variation to compare against your evaluation criteria. More than eight produces diminishing returns and makes it harder to pick a clear winner. Generate the variants after you've defined what you're testing for, not before.
What's the difference between a subject line and preview text, and do I need to test both?
The subject line is what shows in the inbox as the message title; preview text is the snippet that follows it, usually pulled from the email's first line unless you set it manually. They should not repeat each other. If the subject line states the topic, the preview text should add relevance or specificity that justifies opening the email. Testing them as separate variables, rather than treating preview text as an afterthought, is part of the framework.
How do I know if a subject line test result is meaningful or just random noise?
Watch the size of the gap, not just which variant "won." A 2-point difference in open rate between two subject lines is noise on a small list — it could flip the other way next time. A 12-point difference is worth paying attention to and worth logging as a pattern. Running the same type of test across 10-20 sends, and logging which evaluation criteria predicted the winner each time, is what turns individual test results into a usable pattern for your specific audience.
Is Brass-SEO the same product as Copper Sun AI?
No. Both are built by Copper Sun Content and Creative, LLC, but they serve different jobs. Brass-SEO connects to your Google Search Console and Google Analytics 4 accounts and answers plain-English questions about your SEO performance for $25/month. Copper Sun AI is a separate marketing platform built for teams running full campaigns, including the subject line and preview text testing features referenced in this post. You can apply the framework above with any AI chat tool — it doesn't require Copper Sun AI specifically.