- Where
- Top of sidebar, under Overview
- App route
app.newsletrix.com/recommendations- Answers
- What should I change first?
- Needs
- One analysed newsletter
- Updates
- Minutes after each new analysis
What this screen is
Newsletrix ends every newsletter analysis with a short list of improvement suggestions, up to five per email. Each one carries a rationale and a sentence describing the expected business impact. Written one send at a time, they are easy to lose. This screen is where all of them land together.
Identical suggestions are merged. Two newsletters that produced exactly the same recommendation, the same rationale and the same expected impact become one entry, with a counter saying how many newsletters it came from. Sorting is by an impact score read off the wording of the expected impact sentence, highest first, so the strongest claims sit at the top.
Three summaries sit above the list. A quadrant separates quick wins from big bets. A block of tiles shows which suggestions repeat most often across everything you have ever collected, and a shortlist picks out the three highest-impact items on the current page.
Nothing here is measured. There are no opens, no clicks and no revenue in any of these numbers, because Newsletrix reads competitor newsletters the way an ordinary subscriber does. Every score on the page comes from the language the AI used about an email it read. Treat the whole screen as an editorial second opinion at scale. It is not evidence that anything performed.
Open it when
- You have just analysed a batch of competitor sends and want the shortlist rather than 40 separate analyses.
- You are planning next month's calendar and need three or four concrete changes to test.
- You want to know which advice keeps coming back across many newsletters, and which is a one-off note.
- You need a CSV of suggested improvements, with rationale and expected impact, to hand to a client or a designer.
- You want to see what the AI says about one specific brand, or about one specific send, by narrowing the Brand and Subject filters.
- A recommendation on a newsletter detail page looked interesting and you want to know whether it applies to more than that one email.
Screen anatomy
The screen is a single column. At the very top sits the title Newsletter recommendations with the sub-line about AI-powered insights, and a round help icon on the right that opens a fixed explainer panel. Everything below it is data. The real screen comes first, then an annotated map.
Newsletter recommendations
AI-powered insights to boost engagement and effectiveness.
Quick wins
- Add a single primary CTA above the fold
- Rewrite subject line to lead with the offer
Big bets
- Redesign the template for mobile-first tappable CTAs
Fill-ins
- Replace the generic Shop now button label with the offer
Time sinks
- Segment the list by purchase recency
- Filter card Two controls. Brand is a multi-select whose options are labelled with the brand name and the number of analysed newsletters attributed to it. Subject stays disabled and reads Select brands first until at least one brand is picked; the italic hint above it says so. A Clear all button appears in this card only while a filter is active.
- Effort × Impact quadrant Every recommendation on the current page with a known impact level, positioned by estimated effort along the horizontal axis and AI impact along the vertical one. The four counters in the header top right say how many points fell in each zone. In the example, 19 of the 20 rows on the page are plotted; the twentieth had an unclassifiable impact sentence and was dropped.
- Most common recommendations Up to six tiles showing the recommendation sentences that repeat most often across everything you have collected. The first tile is highlighted as the most frequent. Where example use cases exist, two of them are shown per tile.
- Top priority recommendations The first three high-impact rows of the current page, restated as a shortlist. Each card carries a HIGH IMPACT badge, the expected impact sentence clamped to two lines, and the newsletter count.
- Recommendation list One card per merged recommendation. The badge row gives the impact level, the number of newsletters that produced this exact text, and the numeric score. The chevron on the right expands the card in place; clicking anywhere else opens the full detail view.
- View controls Sitting just above the list: Export CSV, which downloads the rows currently on screen, and the Cards and Table toggle. The table view is denser and shows the same rows.
- Pagination footer Reads Showing X to Y of Z recommendations and appears only when there is more than one page. Twenty rows per page.
- Detail view Opens from a card, from the Details button, from a table row's View button, or from a point in the quadrant. It is the only place where the full text of every section appears, with nothing shortened.
- Empty state A lightbulb, the heading No recommendations found, and a line suggesting you adjust the filters. You will see it before your first analysis has produced any suggestions.
Every panel explained
Effort × Impact quadrant
A scatter board with four zones. Quick wins sit top left, Big bets top right, Fill-ins bottom
left and Time sinks bottom right. Each dot is one recommendation from the current page. Hover a
dot for the recommendation text, a line reading Effort N/5 · Impact N/3 and
the brands it came from. Click it to open the detail view.
How it is calculated. The vertical axis is the AI impact label as a number, high 3, medium 2, low 1. Anything that could not be classified is dropped from the chart. The horizontal axis has nothing to do with the AI. Effort comes from the wording of the recommendation and its rationale, sorted into five fixed bands. Band 5 covers words like migrate, new program, build a, audit the. Band 4 covers DMARC, DKIM, SPF, BIMI, segmentation, redesign, integrate, attribution. Band 3 covers restructure, reorder sections, A/B test, simplify, shorten the, personalise, automation, nurture. Band 2 covers subject line, CTA copy, call to action, preheader, send time, send day. Band 1 covers small edits such as add a, remove the, swap, replace the, avoid. Where a recommendation fits more than one band the more expensive one wins, and wording that matches nothing lands on 3. The zones split at 2.5 on both axes, so only high-impact items reach the top half, and only effort 1 and 2 reach the left half.
How to read it. Work the top left first, book the top right into a sprint, batch the bottom left when you have spare capacity, and consciously decide whether the bottom right is worth anyone's week. Because the impact split falls at 2.5, a medium-impact item can never appear as a quick win, however trivial it is to do. Points are nudged slightly off their exact grid positions so overlapping recommendations stay visible.
Effort is a keyword guess, never an AI estimate. A suggestion phrased as build a welcome flow lands at effort 5 because of the words build a, even if it is a ten-minute change in your sending platform. The chart also covers only the 20 rows on the current page, and silently omits any row whose impact could not be classified, so the four counters can add up to less than the number of rows you see.
Most common recommendations
Up to six tiles naming the advice that shows up most often across your whole collection, not just the current page. Each tile shows the recommendation sentence and, where they exist, two example use cases.
How it is calculated. Grouping here is on the exact recommendation sentence together with its impact level, and the pairs that occur most often come first. Six tiles reach the screen at most. The example use cases under a tile come from one of the newsletters in the group rather than from all of them.
How to read it. Advice repeated across many newsletters is systemic, so it is usually about your template, your CTA pattern or your subject-line habits rather than about one campaign. The highlighted first tile is the single most frequent theme.
The tiles are ordered by frequency, but the frequency itself is never shown. Tile one may occur 40 times and tile six only twice. When you need the actual number, use the newsletter count on the individual rows below.
Top priority recommendations
A shortlist of up to three cards, meant as the answer to "if I only do three things this month, what are they".
How it is calculated. The first three rows on the current page whose impact level is high. The list is already sorted by impact score, so these are the strongest claims on the page. Each card footer shows the newsletter count, always written in the plural even when it is one, and a Score: value where one exists.
How to read it. Treat it as a starting point. The shortlist is drawn from the current page only, so paging changes it, and the block disappears entirely when nothing on the page is classified high.
The Score: on these cards is the average marketing effectiveness of the newsletters behind the recommendation, on a 0 to 100 scale. The Score badge on the list rows below is the impact score, on a 0 to 3 scale. They are unrelated. In practice that effectiveness value is almost always empty, so on most accounts the Score: line never appears.
The recommendation list
One card per merged recommendation, in cards view by default. The badge row gives the impact level, the newsletter count and the impact score. Below it sits the recommendation itself and the rationale clamped to two lines. Colour follows impact. Red is high, blue medium, green low, and plain grey means the impact wording could not be classified.
How it is calculated. Two suggestions merge only when the recommendation, its rationale and its expected impact sentence all match exactly. A merged row shows how many newsletters are behind it, the brands it was seen for, and their subject lines, most recent first, five of them and a note saying how many more there are. Sorting is by impact score, highest first. Repetition does not influence the ranking at all.
How to read it. The badge is the severity the AI claimed. The newsletter count says how consistently it claimed it, and the score says which wording band the expected impact fell into. Expand a card with the chevron to see the rationale in full, the expected business impact, the numbered example use cases, the affected brands and the subject lines. The table view shows the same rows with the same numbers in a denser grid, with the impact label shortened to HIGH, MEDIUM, LOW or UNKNOWN and the score printed with one decimal.
When the expected impact sentence matches none of the scoring words, the score is 0 and the badge is hidden rather than shown as zero. The advice is still there, it simply could not be ranked, and it sits at the bottom of the list.
Detail view
Clicking a card, a Details button, a table row's View button or a quadrant dot opens the full record. Nothing is shortened here.
What it shows. The badge row repeats the impact badge, the newsletter count and the score, and adds an amber Priority badge when the level is high. Under it come Why this matters, Expected business impact, Example use cases numbered in full, Affected brands, Impact score and Recent newsletters with this recommendation.
How to read it. This is the full version of a recommendation, and the only place where every example use case is visible. The subject lines at the bottom are plain text rather than links, so use them as a lookup key on the Newsletters screen.
Clarify CTA value and diversify CTAs across products
AI-powered recommendation based on 7 analyzed newsletters
Metric definitions
| Metric | What it means | How it is produced | Scale |
|---|---|---|---|
| Newsletter count | How many analysed newsletters produced this exact recommendation text | How many suggestions merged into this row, where the recommendation, rationale and expected impact all match exactly | Whole number, 1 or more |
| Impact score | The number used to sort the list and shown in the Score badge | Read off the expected impact sentence. Mentions high, 3. Increase or medium, 2. Improve or low, 1. Nothing matches, 0 | 0, 1, 2 or 3 |
| Impact level | The coloured badge on the card, the table row and the detail view | A separate word test on the same sentence. High, higher, significant, substantial or improves gives high; increase or medium gives medium; relevance or low gives low; anything else is unknown. A sentence that matches more than one gets the stronger label | high / medium / low / unknown |
| Effort (quadrant horizontal axis) | Estimated cost of making the change | Read from the wording of the recommendation and its rationale, sorted into five fixed bands. Wording that fits more than one band gets the more expensive of them, and wording that fits none sits at 3 | 1 (trivial) to 5 (programme level) |
| Impact (quadrant vertical axis) | The AI impact label as a number | high becomes 3, medium 2, low 1; unknown is excluded from the chart | 1 to 3 |
| Quadrant counters | Points in each of the four zones | Both axes split at 2.5. High impact with effort 1 or 2 is a quick win, high impact with effort 3 or more is a big bet, and the low-impact half splits the same way at the bottom | Four numbers that add up to the plotted points, not to the rows on the page |
| Score on the priority cards | Average marketing effectiveness of the newsletters behind this recommendation | Average of the per-newsletter effectiveness score. Almost always empty, so the line is usually hidden | 0 to 100, one decimal |
| Example use cases | Concrete ways to apply the recommendation | Written for each recommendation and added some time after it first appears; occasionally you get more or fewer than five | Usually 5, sometimes none |
Filters and controls
| Control | What it does | Default |
|---|---|---|
| Brand | Multi-select. Narrows every panel on the screen to the selected brands. Options are labelled with the number of analysed newsletters per brand. Changing it clears the Subject filter and returns to page 1. | All brands |
| Subject | Single select. Narrows the whole screen to one specific send. Disabled until at least one brand is selected. The list holds at most 100 sends, the ones that produced the most recommendations first, and long subjects are cut to 80 characters. | Empty, and locked |
| Clear all | Resets brand, subject and the page number. Only visible while a filter is active. | Hidden |
| Cards / Table | Switches the list between expandable cards and a dense table. The switch is instant, and the choice is forgotten when you refresh the page. | Cards |
| Chevron | Expands one card in place to show the rationale, expected impact, all example use cases, brands and subject lines. | Collapsed |
| Export CSV | Downloads the rows currently on screen with the columns Recommendation, Rationale, Expected Impact, Impact Level, Occurrence Count, Brands and Avg Effectiveness Score. Disabled when the list is empty. | Current page only |
| Pagination | Twenty grouped recommendations per page. Shown only when there is more than one page. | Page 1 |
| Help icon | Opens a fixed explainer panel about how recommendations are made. It contains no data from your account. | Closed |
A few things are missing on purpose, so you can stop looking for them. There is no sort control, because the order is fixed to impact score descending. There is also no date range, so the screen always covers everything you have ever collected. No free-text search, no way to dismiss, tick off or annotate a recommendation, and no summary tiles of totals anywhere on the page. The workspace or group selector does not apply here either, so recommendations are shown across all of your tracking addresses.
How to use it
- Read the quadrant before the list The four counters tell you the shape of your backlog in one glance. A page that is mostly Fill-ins means the AI found polish, not problems. A page dominated by Big bets means your next real improvement is a template or programme change rather than a copy tweak.
- Check the repeat themes Scan Most common recommendations. Anything appearing there is a pattern across many sends, which makes it a template or process fix rather than a one-off. Those are usually worth more than a higher-scored one-off further down the list.
- Open the two or three you believe Click into the detail view. Read the rationale and the expected impact in full, then read the example use cases: they turn a sentence like clarify CTA value into something a designer can actually build.
- Narrow to one brand when you are auditing a competitor Select the brand, then pick a specific subject line if you want the advice from a single send. The quadrant, the themes and the list all follow the filter.
- Export the page and turn it into a plan Export CSV gives you the rationale and expected impact alongside each recommendation, which is what a client or a stakeholder needs to see. Remember it is one page at a time.
- Come back after the next batch Recommendations are rewritten whenever a newsletter is re-analysed, and new sends add new rows. What was a one-off last month may be a repeat theme now.
Commit to three changes per cycle, one quick win, one repeat theme and one big bet, and write down what you expect to happen before you ship them. The screen ranks suggestions by how confidently they were phrased, not by what worked for you, so your own record of results is the only thing that turns this into a real prioritisation over time.
Limits and caveats
- Nothing on this screen is measured. Every number comes from AI-written text about the content of an email. There are no opens, clicks or revenue anywhere in it.
- Impact level and impact score are word tests over natural language, so they can misfire. A sentence containing highlight matches the test for high, and below or follow match the test for low.
- Effort, and therefore the entire quadrant, is a keyword guess made from the recommendation text, not an estimate of your actual workload.
- The quadrant, the Top priority shortlist and the CSV export all cover the current page of 20 rows only, never the whole list.
- Merging is on exact text, so Shorten the subject line and Shorten subject lines stay two separate rows. The list is more granular than one you would write by hand.
- The total in the pagination line counts individual suggestions while the list shows merged rows, so whenever advice repeats, the reported total is higher than the number of rows you can actually page through, and late pages can come back empty. Rows sharing a score also have no fixed order between visits.
- Figures here can lag a few minutes behind a fresh analysis, so a newsletter analysed a moment ago may not show up straight away.
FAQ
Why does the same advice appear twice with slightly different wording?
Rows are grouped on the exact text of the recommendation, its rationale and its expected impact, all three at once. The AI rarely writes an identical sentence twice, so two phrasings of the same idea stay separate rows. The Most common recommendations tiles group on the recommendation sentence alone, so they are a little more consolidated, but still exact-match. Expect the list to be more granular than a to-do list you would write by hand.
What does the 3.0 / 10 next to Impact score mean?
It is not a score out of ten. The impact score is a 0 to 3 bucket read off the wording of the expected impact sentence. High gives 3, increase or medium gives 2, improve or low gives 1, and nothing matching gives 0. The / 10 suffix in the detail view is cosmetic, so read 3.0 as the top band.
Why do the impact badge and the impact score sometimes disagree?
The badge and the score read the same sentence but accept different words. The score looks for high, increase, medium, improve and low. The badge also accepts higher, significant, substantial and improves as high, and relevance as low. A sentence such as improves click-through therefore scores 1 but is badged HIGH IMPACT. When the two disagree, read the expected impact sentence itself and decide.
Where do the example use cases come from, and why does my account not show any?
Examples do not arrive with the recommendation. They turn up later, usually five concrete ways to apply it, and until they do the example block is simply absent from the card and from the detail view. Re-analysing a newsletter replaces its recommendations, and the examples attached to the old ones go with them until fresh ones appear.
Why does a five-minute change land in the high-effort part of the quadrant?
Effort is not estimated by the AI. It comes from the wording of the recommendation and its rationale, sorted into five fixed bands, and where a recommendation fits more than one band the more expensive one wins. A recommendation phrased as build a re-engagement flow lands at effort 5 because of the words build a, even if it takes ten minutes in your sending platform. Wording that matches no band sits at effort 3. Use the quadrant to sort, not to plan a project.
Does the CSV export contain every recommendation?
No. It exports the rows currently on screen, one page of up to 20 grouped recommendations. The columns are Recommendation, Rationale, Expected Impact, Impact Level, Occurrence Count, Brands and Avg Effectiveness Score. To get more, page through the list and export each page.