Amazon title advice, checked against 12,546 measured edits

Every title in this dataset was live on a real seller's catalog. Weeks after each change, the listing was re-examined against its own pre-change baseline. The adjustment covers market movement, price swings, stockouts, ad spend, rating changes and competitor launches, so what is left is closer to the title's own contribution than a raw before and after would be.

Every edit can be classified by where the text started and where it ended up. That single cut produces the cleanest result in the whole dataset.

Deleting text fails 11% of the time, about twice the base rate, and gets flagged for reversal at 17%, double again. Move that same text into Item Highlights instead and the failure rate is 0.6%. Eighteen times less downside, for what feels to the seller like the same decision: this does not need to be in the title anymore.

Narrow it to edits that removed a keyword and it gets worse. Out of 305 of them, 27% worked, 14% failed, and 13% backfired, which is more than three times the overall backfire rate. Almost a third were later flagged for reversal.

The post-change notes say the same thing in plainer words. Keyword removal shows up in 7.0% of the "never do this again" notes and 0.5% of the "keep doing this" notes. Nothing else in the taxonomy is that one-sided.

Some real examples, with the search volume that walked out the door:

• Dropping "Wedding" from a party supply title gave up roughly 39,000 monthly searches across wedding keywords. The product's own reviews confirmed weddings were the main use case.
• Cutting "Acid Reflux Medicine" to fix word repetition lost a ~47,000/mo term, top two for the category.
• Swapping "Pot Lid" for "Frying Pan Cover" traded a ~4,600/mo keyword for a 748/mo one.
• Replacing "Presentation Binder" with "for Presentations" gave up a ~9,600/mo exact match on the assumption Amazon would understand the paraphrase. It did not.

If deleting is risky and relocating is safe, does that show up in the money? Grouping every measured change by whether it deleted text, moved text into Item Highlights, or both:

Changes that relocated without deleting came out positive on click-through 65% of the time, with a median sales change of +3.9%. Deleting without relocating: 47%, and effectively zero on sales.

The bottom row deserves a second look. Changes that only added or reworded, with no restructuring, did worse than leaving the title alone. Median sales change of −2.5%. Tinkering is not free.

The two-field format, a title under about 75 characters with a separate Item Highlights field carrying specs and secondary keywords, is not a technique anyone chose. Amazon pushed it. It went live on 27 July 2026, and listings in this dataset were being rebuilt for it from early July.

Compare the two formats head to head and the two-field version looks dominant: 59% positive on click-through against 44%, and a median sales change of +3.8% against −3.9%. That comparison is worthless, and the reason matters, because the same trap sits in most title studies.

Almost every remaining single-field change in the window happened in the first five days of July. The two-field changes cluster around 24 July. The two groups are roughly three weeks apart, sitting on either side of a platform change. They never competed in the same market, so the gap between them measures the calendar at least as much as the format.

Here is the test that settles it. Take only the single-field changes, where Item Highlights is not involved at all, and ask whether a shorter title did better.

Length does nothing. Across ~1,250 changes the click-through rate moves by four points between the shortest and longest buckets, in favour of the longer titles. There is no short-title bucket to show you because short single-field titles barely exist: one change in the whole dataset finished under 80 characters without using Item Highlights.

So the shortening that helped these listings was never about length. It was about the second field arriving to catch what the title could no longer hold, and that field arrived because Amazon added it, not because anyone chose it.

This is what the good version of a forced shortening looks like. The title lost 119 characters and not one of the removed specs left the page.

Titles usually get optimized for clicks. So every measured change was sorted into four quadrants by what happened to click-through and conversion after adjusting for market, price and inventory.

Read the middle two rows again. Losing clicks while gaining conversion came out positive on sales 62% of the time. Gaining clicks while losing conversion, 37%. The version of that trade every seller instinctively wants is the version that loses.

Only 21% of changes lifted both metrics. Roughly 35% delivered one at the cost of the other. Judge a rewrite on click-through alone and you will approve the losing trade about as often as the winning one.

Here is the good version in practice. This change deliberately traded away generic traffic.

A 22% drop in click-through would get this title reverted in most operations. It was one of the most profitable changes in the dataset. The generic terms were pulling in shoppers who were never going to buy a $200 coffee scale.

Amazon's style guidance says lead with the brand, and most sellers do. In this data, brand placement is the loudest argument in the whole ledger. It shows up in 20% of the "keep doing this" notes and 23% of the "never again" notes, which is what you would expect if both placements are correct for different brands.

The rule that came out of the wins is a threshold, and it is pleasantly concrete.

• Above roughly 1,000 branded searches a month: keep the brand first. People are searching for you by name and the leading position is what matches. Validated 65 times in the win notes, including brands sitting at ~1,100/mo.
• Near zero branded search volume: move the brand to the end of the title, or out of the title into Item Highlights. Validated 188 times. One brand at 455/mo branded searches was moved out of the title and its branded keyword click-through improved anyway.

Moving the brand out of the title carried an 11:1 win to failure ratio across 323 edits, against a baseline of 8.5:1. For a brand nobody searches for, those opening characters are the best real estate on the listing and they are being spent on a word that matches no query.

One of the notes puts it bluntly: "Brand-first positioning in this commodity category. CTR remained below market benchmarks despite other improvements; the data strongly validates moving brand to end."

Of every tactic that can be isolated, moving the primary keyword into the opening of the title performed best. 63% worked, 5.8% failed, a 11:1 ratio against the 8.5:1 baseline and the highest raw success rate of any category.

The constraint that keeps coming up in the notes is the mobile visible zone. On a phone Amazon truncates the title somewhere around 60 to 80 characters, and the notes consistently target the first 40 to 60. Not "early in the title" but a specific character budget:

• "'Silver Charger Plates' as a contiguous phrase in the first 55 characters, captures the highest-volume keyword (~12,000/mo) in the mobile zone."
• "Front-loading 'Plastic Wine Glasses' in first 60 chars correlated with CTR doubling."
• "'Plastic Champagne Flutes' as contiguous phrase in first 60 chars, drove ranking improvements for 85K/mo keyword family."
• "16" size spec in first 25 characters, confirmed indexed for '16 inch cake stand' (rank 12)."

Notice how often "contiguous" turns up next to "front-loaded". Getting a keyword into the opening characters is one requirement. Keeping it as an unbroken phrase is a second one, and it is the more fragile of the two.

Sellers tend to assume that if all the words are somewhere in the title, the phrase is covered. The data disagrees, and the failure is severe. Edits touching phrase contiguity backfired at 5.2% and were flagged for reversal at 15%, both well above baseline. Contiguity appears in 6.1% of failure notes against 2.6% of win notes.

The clearest case in the dataset is a kitchen product where a restructure accidentally split the category's biggest keyword.

That last detail is the whole lesson. On the dashboard it read as a small CTR win while it was quietly pulling the listing out of the biggest keyword in its category.

Every change in this dataset gets an attribution confidence, and it is worth being precise about what that measures. It is not a judgement on the listing or on the quality of the edit. It describes how much else was happening at the same time.

A change lands in the high or medium band when the listing held steady through the measurement window: it stayed in stock, the price did not swing, no competitor launched into the category, the rating did not move and ad spend held. Under those conditions the sales change that survives the adjustments can reasonably be credited to the title. A change lands in low or very low when a stockout, a price cut or a rival's surge was large enough to swamp anything the title did. The effect might still be there. It cannot be separated from the noise.

On a listing that stayed steady, a title change is worth about 4% revenue growth at the median. The mean gets pulled to 9.3% by a handful of large wins. On one listing doing ~$204,000 a month, a single change was worth +~$19,000 a month.

Read that as a ceiling rather than a promise. The top two rows are not a random sample of title changes, they are the ones that happened on stable listings, and stability is doing some of the work in that number. A title change cannot lift a listing that is out of stock, and the reason those 255 changes were measurable is much the same reason they had room to succeed.

The bottom two rows make the point from the other side. Where a stockout or a price swing dominated the window, the median title change returned exactly zero. Not a small gain. Zero. If your listing is in that state, the title is not the thing to fix this week.

Note also where the 4% comes from. It is not one heroic rewrite. It is picking the edit types near the top of finding 11 and refusing the ones near the bottom, on listing after listing. Most of the money in title work is in the failures you avoid, not the wins you chase.

This one costs nothing to act on. When impressions rose after a title change, median sales rose 11% and 67% of listings gained. When impressions fell, median sales fell 8.9% and only 42% gained. Impressions tell you within days whether the edit put the listing in front of more people, which is the thing a title mostly controls.

Most sellers check click-through instead, and click-through is close to useless on its own. Correlated against the actual sales change across ~1,600 title changes:

• Sessions against sales: r = 0.655
• Adjusted conversion rate against sales: r = 0.155
• Adjusted click-through rate against sales: r = 0.089

Traffic volume dominates. That does not contradict finding 04, where conversion is the metric worth trading for at a fixed traffic level, but it does reframe what a title is for. Its first job is getting the listing into more search results. Winning the click comes second.

Which also explains the pepper grinder in finding 07. CTR went up. Impressions went down 63%. Only one of those was going to matter, and a seller watching the CTR number would have called that change a success.

The instinct with a title is to treat it as a one-off: write the good version, ship it, move on. Tracking listings across their whole optimization history says otherwise.

Only 46% of listings got their best result from the first attempt. For the other 54% the best version came later: 27% on the second round, 16% on the third, 11% after that. A single rewrite, however carefully argued, leaves value on the table for more than half of a catalogue.

That is not because the first attempt was careless. It is because a title cannot be fully judged until it has been live long enough to show which keywords it actually caught, and that evidence only exists after the change has shipped. Round two is where the first round's measurement gets spent.

The returns do run out, and the table shows where. Rounds two and three still earn their place, with the third round posting the best median sales change of any round. From the fifth onward the numbers turn hard negative.

That tail is mostly selection rather than cause. We checked whether the gap between edits explains it and it does not, since editing after three weeks performed no worse than editing after twelve. Listings reach a fifth title edit because something is wrong, and by then it is usually not the title. One note in the data puts it plainly: "Title remains unranked for virtually all tracked keywords despite 3 optimizations. The fundamental issue may be chronic low inventory suppressing organic ranking signals, not title quality."

Every edit type at once, counting only the edits whose outcome could actually be observed.

Of the ~13,000 scored edits, ~5,700 carry an intent outcome of unverifiable. A stockout, an Amazon auto-revert, or too little post-change data. Those edits still get a keep-or-revert verdict, but that verdict is a judgement call rather than a measurement. The table below drops all of them and keeps only the ~7,500 where the outcome was actually observed. Rates are higher across the board because the no-effect cases thin out, so read the ranking rather than the absolute numbers.

Moving text into Item Highlights backfires 0.2% of the time. Removing a keyword backfires 13%. Same words, about sixty times the risk, and the only difference is whether the text survived somewhere on the page.

Front-loading remains the best single edit anyone can make, at 91% worked and roughly 27 wins per failure.

Now the bottom row. Adding a feature or benefit claim sits last at 2.6:1, with a 20% backfire rate. That looks like shoppers punishing vague claims, and it is not what happened.

Opening it up: 98 of the 104 backfires are a single brand re-adding a single phrase. The phrase is "Fade Resistant", on bedding, and that brand's own rules explicitly prohibit it because the product does not reliably do it. It went back into titles 189 times across 72 ASINs.

Take that one phrase out and the edit type transforms. Backfire falls from 20% to 1.7% and the ratio goes from 2.6:1 to 11:1, which moves it from the worst row in the table to near the top. Feature claims are not the problem.

That is worth sitting with if you run a catalogue of any size. The rule existed. Somebody had already worked out that the claim was not true of the product. And the title kept getting it back anyway, at a cost of roughly a fifth of every claim edit that brand made.

The distinction that does survive is verifiability. Trust and compliance claims sit near the top at 12:1, and they are checkable: a product is BPA-free or it is not, made in the USA or not, dishwasher safe or not. A claim you can point at a certificate or a spec sheet for is a good edit. A claim nobody has verified is a liability whether or not it sounds soft.

The trust claims at the top of the table do the opposite, and the good ones work twice. One note records "BPA-Free" answering the most common complaint in competitor reviews while also targeting ~64,000 monthly searches for "cutting boards non toxic". One phrase playing defence and offence at once.

Use case phrasing pays the same way when it maps to real queries rather than to a mood. "for Men & Boys" picked up "bible cover for men" at ~26,000 a month and "bible cover for boys" at ~2,500 a month, in four words. The test for any addition is whether you can point at the search volume or the certification behind it. If you cannot, it belongs in the bottom row.

One number the table above cannot show you: what a backfired edit actually costs the change it sits in.

A title change is not one edit. It is usually seven or eight of them shipped together, so the question that matters to a seller is not how a single edit scored, but what happens to the whole change when one of them goes wrong.

One bad edit is enough to take a change from +2.5% to −3.6% on sales. It does not get averaged away by the six or seven edits around it that worked. That is a swing of about six points of revenue, and it is why the ranking above is worth reading from the bottom up rather than the top down.

The same asymmetry shows up in what gets undone. 12% of all measured edits carry a recommendation to revert them. Concentrated exactly where you would now expect: 28% of keyword removals and 25% of feature claims, against 6.9% of size and count edits. Roughly one in four deletions gets put back.

One check before anyone acts on this ranking. A handful of large sellers contribute most of the edits here, so a type could sit near the top purely because whoever favoured it had a good quarter. To separate the two, the whole ranking was rerun inside each seller's catalogue on its own and compared against that seller's own average.

Four types cleared the bar in every catalogue that had enough data to test: moving text into Item Highlights, front-loading the primary keyword, adding a verifiable trust claim, and changing a size or count. Those are safe to copy. Keyword removal cleared it in under a fifth of catalogues, and rewording and soft feature claims in about a quarter, so the bottom of the table is not one team's bad luck either.

The middle rows are the ones to treat carefully. Adding keywords and moving the brand name both look respectable in aggregate but only beat the seller's own average about half to two thirds of the time. Those are edits to test on your own listings rather than copy from a table, and the ratio column alone would never have told you that.

• Never delete, always relocate. If a phrase leaves the title it should go into Item Highlights. Deleting fails eighteen times more often than moving.
• Check contiguity before you ship. Write out your top five keywords and confirm each one appears as an unbroken phrase in one field or the other. Having all the words is not the same thing.
• Look up your branded search volume before deciding where the brand goes. Under about 1,000 a month, it does not belong in the opening characters.
• Budget the first 60 characters. That is the mobile visible zone. It should hold the primary keyword as a contiguous phrase and not much else.
• Watch traffic, not click-through. Sessions track sales seven times more closely than CTR does. A CTR gain sitting on top of an impressions collapse is a listing quietly dying.
• Judge a rewrite on conversion, not clicks. Losing clicks while gaining conversion beats the reverse, 62% to 37%.
• Budget three rounds, not one. Only 46% of listings get their best title on the first attempt. Past the fourth round the returns are gone and the constraint has moved to inventory, price or reviews.
• Do not chase a character count. Length on its own does nothing. Amazon shortened your title for you in July 2026, so the only question that matters is whether the text it displaced landed in Item Highlights or vanished.