- Where
- Sidebar: Content / Subject lines
- App route
/subjects- Answers
- Which subject-line patterns travel with estimated opens
- Needs
- Analysed newsletters, at least 3 for scatter
- Updates
- Minutes after analysis, can lag about 5 minutes
What this screen is
The Subject lines screen breaks every subject in your account into countable parts. Character count, emoji, digits, currency and percent signs, exclamation and question marks, uppercase share, leftover merge-tag placeholders, a simple tone score. Each of those sits beside the open rate Newsletrix estimated for the email it came from.
The screen answers two questions. What a typical subject line looks like in this set, and which patterns travel with a higher estimated open rate. Length, emoji and punctuation are counted from the subject itself. The pattern comparisons average AI estimates, because Newsletrix receives competitor newsletters as a subscriber and can never see their real opens or clicks.
No experiment runs anywhere on this screen. The panel called A/B insights compares emails that already exist and happen to have or lack a feature, and nothing about the split was randomised. Treat what it shows as a hypothesis for your own sends.
Open it when
- Before writing a campaign subject line, to check the length band and the feature mix your competitors actually use.
- To see the highest and lowest estimated performers in the set, with the tactics each one used.
- When you are deciding whether to add an emoji, a number, a price or a question mark and want a with-versus-without comparison from your own market.
- For vocabulary. The word cloud shows which words your market repeats.
- When you suspect your subjects run too long for mobile and want the market's distribution against the 40-60 character band.
Screen anatomy
The real screen first, then a numbered map of the current layout. Panels stack in one column; the three tabs at the bottom switch only the last panel.
Subject line analysis
In-depth analysis and insights for newsletter subject lines including relevance, frequency, and sentiment analysis.
| Brand | Subject | Date | Length | Emojis | ! | ? | # | $ | % | Upper | Sentiment | Open Rate (Est.) | CTR (Est.) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Be Lenka | Last chance: 20% off barefoot boots 🔥 | Mar 14, 2026 | 37 | 1 | - | - | ✓ | ✗ | ✓ | 3.8% | Neutral | 24.3% | 3.1% |
| Reebok | LAST CHANCE: Up to 60% Off | Mar 11, 2026 | 26 | - | - | - | ✓ | ✗ | ✓ | 70.6% | Neutral | 20.0% | 4.0% |
| Alfred | The Gift That Always Fits 🎀 | Mar 9, 2026 | 27 | 1 | - | - | ✗ | ✗ | ✗ | 23.8% | Positive | 19.2% | 2.5% |
- Filter bar The Brand multi-select, "All brands" until you pick. "Clear all" appears once a brand is selected. It reaches six panels; the others ignore it.
- Key insights Up to three sentences, covering the largest sentiment share, the most repeated exact subject and the top keyword. Hidden until there is something to say.
- Metrics overview Eight tiles. Avg length carries a row against the fixed 40-60 band ("in range", "below range" or "above range").
- Avg open rate (est.) tile The only tile compared to an industry figure. An AI estimate, like every rate on this screen.
- Subject length vs. Est. open rate One dot per newsletter, a green trend line, a grey dashed industry line and an amber dashed "Ideal 50" line. The badge gives r and n.
- Word cloud - Top keywords The 30 most frequent words, sized by frequency, each with its count.
- Subject length distribution 10-character bins, the 40 and 50 bars highlighted, in-band and outside-band averages and a lift chip.
- A/B insights: what lifts open rate? Six features, each with a With bar, a Without bar, n= per side and a delta in percentage points.
- Sentiment distribution Subject lines per tone bucket; only non-empty buckets appear.
- Emoji usage A doughnut split into No Emoji, 1 Emoji, 2 Emojis and 3+ Emojis.
- Tabs and the All subject lines table All subjects (default), Top performers and Special features switch only this panel. 14 columns, newest first, 20 rows per page.
- Page header Sits above everything else. "Subject line analysis", its subtitle and a help icon that opens the "Craft Winning Subject Lines" dialog.
- Top performers tab "Top performing subjects" and "Lowest performing subjects", five each, every entry with a collapsible "Why this worked" checklist.
- Special features tab Four count cards, With Emoji, With Numbers, With Personalization and With Urgency (!), each with "Avg Open Rate (Est.)".
- Empty states "No subjects found", "No data available", "No keywords available", "No length data", "Not enough data yet" and "Not enough data to plot a correlation."
Every panel explained
Key insights
Short observations you can check against the panels below.
How it is calculated. Three lines can appear, never more. The share of the largest sentiment bucket, which is that bucket divided by the sum of all buckets to one decimal. The most repeated exact subject, with the number of times it was sent. The top keyword, with the number of subject lines carrying it.
How to read it. Headline facts. The card stays hidden until there is enough on the screen to summarise.
Metrics overview
Eight tiles describing the average subject line in the selection.
How it is calculated. Total subjects counts distinct analysed newsletters. Avg length is the mean character count, rounded on the tile and one decimal in the row beneath. Avg emojis and Avg sentiment show two decimals. The three With tiles show a count and its rounded share of Total subjects. Avg open rate (est.) is the mean AI estimate, one decimal. The Avg length row compares to the fixed 40-60 band, and the Avg open rate row compares to the industry figure, showing its chip only for a material gap of at least 2 percentage points or 10% relative.
How to read it. The industry figure is a fixed 2024 published benchmark for the industry you picked in Settings (retail, 15.7%, by default), and it is flagged as possibly outdated because that figure is over 18 months old. Tiles with nothing to average show "N/A" or "0.00".
Subject length vs. Est. open rate
One dot per newsletter: subject length across, estimated open rate up.
How it is calculated. The chart plots the 500 most recent newsletters that have both a subject length and an estimated open rate. The green line is a least-squares fit, and the badge beside the title gives Pearson r and the number of dots. No line appears below 3 dots, or when every dot shares the same length or the same rate. An |r| of 0.5 or more reads "strong", 0.3 "moderate", 0.1 "weak", below that "no correlation". The grey dashed line is the industry average, the amber dashed line the ideal 50 characters.
How to read it. Mobile inboxes cut subjects off around 40-50 characters, which is what the amber line marks, and the chart asks whether staying inside that window comes with more opens. A flat line means no link in this data. Correlation is not causation, and the vertical axis is an estimate to begin with.
This chart ignores the Brand filter and always plots all brands.
Word cloud - Top keywords
The 30 most frequent words across all subject lines.
How it is calculated. Case, punctuation and emoji are ignored, so "SALE!" and "sale" are one word, though a hyphenated word stays whole. Words of 2 characters or fewer, bare numbers and 86 common English stop words (the, and, your, our, get, new and so on) never appear. Everything else counts once per occurrence, so a word used in ten sends scores 10. Size runs on a logarithmic scale, which is why the top word is bigger than the rest without being ten times bigger.
How to read it. The vocabulary your market leans on. Only English filler words are dropped, so a non-English set keeps its filler, and because every occurrence counts, one daily sender can dominate.
Subject length distribution
Subject lines per 10-character bin, and whether the mobile band pays in your data.
How it is calculated. Bins are labelled by their lower edge (0-9 is "0", 10-19 is "10"); hovering shows the count and the bin's average estimated open rate. The band runs from 40 inclusive to 60 exclusive, so only the 40 and 50 bars are highlighted. The footer compares the average estimated open rate of 40-59 character subjects with everything else; lift is in-band minus outside, one decimal, green above +0.05, red below -0.05, grey between. The chip needs an average on both sides to appear, and empty bins are left off the axis.
How to read it. Tall bars are the house style. The chip says whether the window coincides with higher estimated opens in your own set. The 40-60 rule is built in rather than learned from your data, and a bar can rest on a single newsletter.
A/B insights: what lifts open rate?
Average estimated open rate with versus without each of six features.
How it is calculated. Emoji in subject (one or more versus none), Number in subject (a digit versus none), Personalization token (placeholder present versus not), Exclamation mark and Question hook (one or more versus none), Currency symbol (present versus not). Each side reports its count and average. Delta is With minus Without, one decimal with a "pp" suffix. A green up arrow above +0.05, a red down arrow below -0.05, a grey minus in between, and "n/a" when a side has no average. Bar lengths are relative to the largest rate in the panel, so read the numbers rather than the widths. A row disappears only when neither side has a single email in it.
How to read it. A +2.0pp row means two percentage points more estimated open rate with the feature. Read the n= before the delta.
The groups already exist in your data and differ in sender, offer and timing. No significance test runs, both averages are estimates, and a side with n=3 is still shown.
Sentiment distribution
Subject lines per tone bucket.
How it is calculated. The score is a word count. Each subject is compared, ignoring case, against a list of 32 positive and 22 negative English words. The score is (positive minus negative) divided by (positive plus negative), and exactly 0.0 when no listed word appears. Buckets run above 0.3 Very Positive, above 0.1 Positive, above -0.1 Neutral, above -0.3 Negative, and Very Negative below that. Only non-empty buckets appear.
How to read it. Most sets skew heavily to Neutral, for the reason below.
The matching is crude. Sarcasm and negation ("not great") slip past it, a word glued to punctuation counts as a different word, so "sale!" never matches "sale", and non-English text matches nothing at all. Everything it misses lands in Neutral. After a while on the page the bars can reorder by count, though the values themselves do not change.
Emoji usage
Share of subject lines by emoji count.
How it is calculated. Buckets No Emoji, 1 Emoji, 2 Emojis, 3+ Emojis. Consecutive emoji count as one, so "Hi 🎉🎉" lands in the 1 Emoji slice. A very new emoji, or one built from a combined sequence, can slip past the count.
How to read it. Pair it with the Emoji row of the A/B panel. Read the slices as close rather than exact.
All subject lines
One row per analysed newsletter: the raw evidence behind the charts.
How it is calculated. Newest first, 20 per page. The Brand cell is a blue chip or "N/A". Emojis, ! and ? each show a count, or a grey dash at zero. The #, $ and % columns show a green tick or a grey cross. Upper is the uppercase share of letters to one decimal, dashed at zero. Sentiment turns green above 0.1 and red below -0.1, grey in between, and a dash when nothing was scored. Open Rate (Est.) and CTR (Est.) carry one decimal.
How to read it. A newsletter Newsletrix has not analysed yet never appears here. One with nothing to count in its subject shows zeros and dashes instead. There is no column sorting.
Top performing and Lowest performing subjects
The five highest and five lowest estimated open rates in the account.
How it is calculated. Lowest excludes the top five, so the lists never overlap. Each entry shows the subject, a brand badge and "Open rate (Est.): x%", emerald on top and red on lowest. Top adds a purple "N emoji(s)" pill; Lowest adds an orange "Long (N chars)" pill above 70 characters.
How to read it. Structures to copy and to avoid. The ranking uses the estimate for the whole email rather than for the subject on its own, and ties fall in arbitrary order.
Why this worked
A checklist under each ranked subject, headed by a count out of 5.
How it is calculated. Five signals are counted. Length 40-60 inclusive, a digit, an emoji, a power word from a fixed English list (exclusive, free, new, save, now, today, last chance, limited, sale, special, secret, discover, introducing, flash, member, vip, premium, only, hurry, ends, bonus, gift, reward), and a percent or currency sign ($, €, £, ¥). The length chip reads "(mobile-optimal)" inside the band, "(too short)" below 40, and a cut-off warning above 60. Advisory chips add colour without adding to the count. One "!" gets an amber "Exclamation" and anything more a grey spam-risk flag, a "?" gets a green "Question hook", over 40% uppercase is flagged "shouty", and "Tone:" repeats the sentiment bucket.
How to read it. The checklist describes what the subject contains. Nothing in it is a quality score. It opens by default on the first top performer only.
Subjects with special features
Four counts with the group's average estimated open rate.
How it is calculated. With Emoji (one or more), With Numbers (a digit), With Personalization (a placeholder such as {{first_name}}, {city} or [NAME] still in the delivered subject), With Urgency (!) (one or more exclamation marks). "Avg Open Rate (Est.)" shows one decimal, or "-" with no average.
How to read it. The count-only cousin of the A/B panel; counts overlap freely. A correctly personalised email arrives with the name filled in, so a high With Personalization count means broken merge tags.
Metric definitions
| Metric | What it means | How it is produced | Scale |
|---|---|---|---|
| Total subjects | Analysed newsletters in the selection | Count of distinct newsletters with an AI analysis | Integer |
| Avg length | Mean subject length | Characters including spaces and emoji; whole number on the tile, one decimal in the row | Characters |
| Avg emojis | Mean emoji runs per subject | Consecutive emoji count once; newer sequences approximate; 2 decimals | 0.00 and up |
| Avg sentiment | Mean tone score | (positive minus negative) / (positive plus negative) over 32 positive and 22 negative English words; 0.0 when none match | -1.00 to +1.00 |
| Sentiment bucket | Tone label | Above 0.3 Very Positive, above 0.1 Positive, above -0.1 Neutral, above -0.3 Negative, else Very Negative | 5 categories |
| Emoji bucket | Emoji label | 0, 1, 2, or 3 and more | 4 categories |
| With numbers | Subjects containing a digit | Count and rounded share of Total subjects | Count, 0-100% |
| With currency | Subjects containing $ € £ ¥ ₹ | Count and rounded share | Count, 0-100% |
| With percent sign | Subjects containing % | Count and rounded share | Count, 0-100% |
| Personalization tokens | Placeholders left in the subject | Counts {{name}}, {name} and [NAME] placeholders still visible in the subject | Integer per subject |
| Upper | Uppercase share of letters | Uppercase letters / all letters, one decimal; dash at zero | 0-100% |
| ! and ? | Punctuation counts | Number of "!" and "?" characters | Integer |
| Open Rate (Est.), CTR (Est.) | Newsletrix's estimate for that email | Estimated from the content of the email, never measured; averaged for tiles and comparisons | Percent, may be missing |
| Delta / lift (pp) | Gap between two group averages | With minus Without, or in-band minus outside; green above +0.05, red below -0.05 | Percentage points |
| Correlation r | Length against estimated open rate | Pearson r over up to the 500 newest newsletters; none below 3 points; 0.5 strong, 0.3 moderate, 0.1 weak | -1 to +1 |
| Mobile-optimal band | Recommended length | Fixed 40 to 60, ideal 50 (Litmus and Mailchimp research) | Characters |
| Industry benchmark | Reference open rate | Fixed 2024 figures: retail 15.68, SaaS 19.29, media 22.15, B2B 21.56, non-profit 25.17; retail by default | Percent |
| Keyword count | Word frequency in the cloud | Occurrences, ignoring case and punctuation, after dropping words of 2 characters or fewer, bare numbers and 86 stop words | Integer |
| Signals | Tactics in one subject | Length 40-60, digit, emoji, power word, percent or currency | 0-5 |
Filters and controls
| Control | What it does | Default |
|---|---|---|
| Brand | Multi-select of every brand in your newsletters. Filters the tiles, both distribution charts, the table, the A/B panel and the histogram; resets the table to page 1. Ignored by the word cloud, Top performers, Special features and the scatter. | None ("All brands") |
| Clear all | Shown only when a brand is selected; clears it and returns the table to page 1. | Hidden |
| Help icon | Opens the "Craft Winning Subject Lines" dialog. | Closed |
| Tabs | All subjects, Top performers, Special features. Switch the bottom panel only; the rest of the screen stays where it is. | All subjects |
| Pagination | 20 rows per page; the pager shows only past one page. | Page 1 |
| Benchmark industry | Not on this screen. A global preference set elsewhere in the app that picks the industry figure for the scatter line and the Avg open rate (est.) row. | Retail (15.7%) |
Several things are deliberately absent. There is no date-range filter, no search box, no column sorting, no page-size selector and no export. The optional Group label you can give a tracking address is not applied here either, so the numbers always cover every newsletter in your account.
How to use it
- Read the shape Start with the tiles. Where Avg length sits against 40-60, what share of subjects uses numbers and percent signs, and whether Avg open rate (est.) shows a material gap to the industry figure.
- Find the house style In the histogram, the tallest bars are what your market writes. A green lift chip with healthy n= on both sides means the mobile window coincides with higher estimated opens; a grey chip means length is not the lever.
- Test each tactic Go row by row through the A/B panel and keep every delta backed by a few dozen newsletters per side. Treat a tiny n= or a delta inside +/-0.05pp as noise.
- Take the structure In Top performers, expand "Why this worked" on the first three entries and note which signals recur. Copy the structure rather than the wording. Read the Lowest list for the pattern to avoid, often the "Long (N chars)" pill.
- Narrow to one competitor Pick a brand and re-read the tiles, table, A/B panel and histogram. The word cloud, performer lists, special features and scatter will not have changed.
Before you schedule a send, score your own subject line against the five signals by hand. 40-60 characters, one number, at most one emoji, one power word, and a deal marker only if there really is a deal. Then check it against the A/B rows that had real sample sizes in your market.
Limits and caveats
- Every performance figure here (open rate, CTR, every lift and delta, both performer lists) is an AI estimate from the content of the email. Newsletrix never sees a competitor's real opens or clicks.
- The A/B panel and the histogram footer are observational. Nothing is randomised, no significance test runs, and no minimum sample size applies except the scatter's 3 points.
- The scatter plots at most the 500 most recent newsletters that have both a length and an estimate, so older ones are silently left off.
- Brand filtering is partial. The tiles, table, A/B panel, histogram and distribution charts follow it, while the word cloud, performer lists, special features and scatter always describe the whole account.
- Tone scoring, the power words and the filtered-out keywords are English only. Non-English markets end up with inflated Neutral counts and filler words in the cloud.
- Emoji counting collapses consecutive emoji into one and misses some modern sequences, so the figure is approximate, and the checklist can disagree with the table about the same subject.
- Most panels can lag about five minutes behind a newly arrived newsletter. One that has not been analysed never shows up at all, and one with nothing to count in its subject shows zeros and dashes in the table and stays out of the feature charts and the scatter.
FAQ
Are the open rates on the Subject lines screen real?
No. Every Open Rate (Est.) and CTR (Est.) figure is an AI estimate produced from the content of the email, not a measurement from an inbox or a sending platform. Newsletrix receives competitor newsletters as a subscriber and cannot see anyone's real opens. Use the numbers to rank emails against each other, not as reportable performance data.
Why do the word cloud, Top performers and Special features not change when I pick a brand?
In the current release those three panels, and the length-versus-open-rate scatter, always show all brands. The KPI tiles, the All subject lines table, the A/B insights panel, the length histogram and both distribution charts do follow the Brand filter. Read the unfiltered panels as market-wide context.
What exactly is the 40-60 character band?
A fixed best-practice window built into the app, attributed to Litmus and Mailchimp subject-line research, with 50 characters as the ideal. It is not derived from your data. The histogram highlights the 40 and 50 bars and compares subjects of 40 to 59 characters against everything else.
Why does the With Personalization count look wrong?
It counts merge-tag syntax still literally present in the delivered subject, such as a first-name placeholder in curly braces or a name in square brackets. A correctly personalised email arrives with the name filled in and is not counted, so a high number usually means a competitor's merge tags failed.
The sentiment chart says almost everything is Neutral. Is that right?
For most accounts, yes. Subject sentiment here is a word count. Each word is matched against 32 positive and 22 negative English words, and when none of them appear the score is exactly 0.0, which falls in Neutral (between -0.1 and 0.1). Short, brand-specific, punctuation-heavy or non-English subjects nearly always score 0.0.
The A/B panel says emoji lift open rate. Should I add emoji everywhere?
Be careful. The panel compares groups that already exist in your data; nothing was randomised, the groups differ in sender, offer and timing, and both averages are AI estimates. No minimum sample size is enforced, so check the n= counts first and treat a lift as a hypothesis to test in your own sends.