I Tracked More Than Half a Million Amazon Keyword Rankings. Something Has Changed.

I know I am publishing a post outside my typical blogging schedule, but I’ve been working on this one for a while and feel it deserves a special place.

For nearly four years, I have tracked where my books appear in Amazon search results. What began as a fairly simple attempt to understand whether readers could actually find my books eventually turned into a rather ridiculous collection of spreadsheets covering eleven titles, almost a thousand search phrases, and daily observations going back to December 2022.

I finally decided to do something with all that data because 2026 has been a shitty year for my book sales (and I don’t think I’m alone). That would be easier to accept if I had stopped publishing, stopped promoting, or simply allowed an aging catalog to fade into the background. I haven’t. I published Jezelle: Thief of Forks in April, only six months after Koen: Quills from the Raven’s Nest. I also regained the rights to Koen this summer, which led to a renewed push of the title. I am continuing to work on my existing catalog and preparing another novel for publication later this year, Sylvanus: Sword and Suns. My body of work continues to grow, yet sales across much of it have fallen sharply.

For months, I had suspected that discoverability was part of the problem. Books that once appeared regularly in Amazon searches were slipping farther down the results, while newer titles seemed extraordinarily difficult to establish at all. A bad sales year can make an author see patterns everywhere, though, so I wanted to know whether the numbers actually supported what I thought I was seeing.

After cleaning duplicate entries and obvious errors, I ended up with roughly 556,000 usable keyword-date observations stretching from December 2022 through September 2026. When I analyzed that history alongside my sales and Kindle Unlimited activity, I found something much more substantial than a few books having bad months.

I cannot see Amazon’s algorithm, and I am not claiming to have reverse-engineered it. Eleven books from one author are not a controlled study of the Kindle Store. What I can say is that the Amazon search environment my books occupy in 2026 does not behave like the one they occupied several years ago, and the clearest changes appear around late 2025 and the first half of 2026.

Amazon has been changing discovery too

Amazon has not announced a specific KDP search-algorithm change that neatly explains what I found. They protect their magic formula like a sorcerer from one of my books. Its own documentation does, however, provide some useful context. Amazon says search results are automatically determined and constantly changing, and specifically lists past sales history, popularity, availability, and the length of time an item has been listed among the factors that can influence placement. Amazon also tells KDP authors that relevant keywords work alongside factors such as sales history and Best Sellers Rank in determining search placement.

At the same time, Amazon has been publicly changing the way discovery works across its larger store. The company says its newer search systems increasingly try to understand what shoppers actually intend to buy rather than merely matching keywords, while its shopping experience has become progressively more personalized through searches, purchase history, reviews, Lists, preferences, and other customer behavior. By May 2026, Amazon said that nearly every shopping experience across the store was personalized in some way.

KDP has also introduced Book Trends Data, which measures customer engagement over the most recent 30 days using purchases across formats, Kindle and Kindle Unlimited reading, and Audible listening. A book must reach a minimum level of recent activity before the information even appears. Amazon does not say that Book Trends itself determines search ranking, so I would not make that leap, but it is another indication of how Amazon is thinking about books in terms of recent customer engagement rather than metadata alone.

None of this proves that Amazon made the specific change my data appears to detect. It does establish that the period when my books began behaving differently coincides with a time when Amazon was openly making substantial changes to search, personalization, AI-assisted discovery, and its interpretation of customer behavior.

How I measured visibility

Looking at keyword rankings over several years is trickier than it sounds because my keyword lists have changed. I might track fifty phrases for a book one year and add another hundred later. If most of those new phrases fail to rank, the overall numbers suddenly look worse even if none of the original fifty changed at all.

For the long-term comparisons, I therefore used stable groups of the same keywords and removed exact duplicates. I also needed a way to distinguish between a book ranking near the top of Amazon search results and one that barely appears at all. Simply counting both as “ranking” would obscure most meaningful movement.

I created a simple Search Visibility Index for that purpose. A keyword ranking #1 receives a score near 100. The score gradually declines as the book moves deeper into the results: a position around #150 is worth roughly 50, while a position near #300 is worth essentially zero. A keyword that does not appear in the top 300 also receives zero. I then average those scores across the fixed group of phrases being compared.

This is not an Amazon metric, nor am I suggesting Amazon itself weights search positions this way. It is simply a consistent ruler. If the same book, measured against the same keywords, drops from 20 to 10, it has lost roughly half of its measured search visibility. It is even more granular than that. Really, only if a book appears on the first three pages of a search result is it “visible” in a real sense. Few people are going to scroll past that before deciding on a book. They certainly aren’t going to flip to page 15. But for simplicity’s sake, I think the metric is sufficient to show movement.

There is also a less clinical side to why I have enough data to do this at all. The size of this dataset represents an enormous amount of time I have spent researching keywords, testing relevant phrases, monitoring search positions, adjusting metadata, and trying to improve the organic visibility of my books. I did that because I believed search visibility was something I could gradually build through sustained effort.

That makes what I found particularly painful. If my interpretation is right, much of that effort was aimed at a discovery system that was changing underneath me. Relevant keywords still matter, and all that tracking has unexpectedly produced a dataset large enough to write this article, but the years I spent trying to squeeze incremental gains from Amazon search may no longer provide anything close to the return they once did.

The broad decline was real

Once I controlled for changing keyword lists, the decline I thought I had been seeing did not disappear. It became much clearer.

The most striking period occurred in spring 2026. Between April and May, Dawn of the Lightbearer, the best-selling title in my catalog, lost roughly 39 percent of its measured search visibility. The Mourning Son lost about 46 percent, Noonday in the North lost approximately 49 percent, Island of Stone fell roughly 39 percent, and Destiny of the Daystar dropped about 30 percent.

These were books of different ages with different sales histories and different sets of keywords. Island of Stone is not even part of the same fantasy series as most of the others. Yet they suffered significant deterioration at approximately the same time when measured against stable keyword populations.

A single book can lose momentum. A series can age, competition can increase, and sales can dry up. Those are ordinary explanations for declining visibility. When several different products all become substantially harder to find through the same searches at roughly the same time, however, a broader change in the discovery environment becomes a much more plausible explanation.

The shift also does not begin neatly in May 2026. Looking across the longer history, I can see weakening in the second half of 2025, followed by a much more pronounced deterioration in the first half of 2026. I cannot identify a precise date when Amazon changed something internally, assuming a particular change is responsible, but the pattern is difficult to reconcile with several books independently having bad years.

It also provides an obvious connection to my sales collapse. These books did not simply sell fewer copies in 2026. Across hundreds of searches where potential readers might once have encountered them, they became substantially less visible.

The older books behave differently

There is nothing surprising about sales affecting Amazon visibility. KDP authors have known that for years, and Amazon explicitly identifies past sales history and popularity as factors that can influence search. What became interesting in my data was that reader activity does not appear to affect all of my books in the same way.

Several older titles retain a surprising amount of elasticity. They can spend weeks drifting downward, receive a relatively modest amount of genuine activity, and suddenly improve across numerous searches. The improvement usually fades, but the book appears capable of recovering parts of a keyword footprint that had largely disappeared.

I see this repeatedly with books such as Dawn of the Lightbearer, The Mourning Son, and Noonday in the North. A small amount of activity can sometimes wake up many old rankings at once. Several titles released during and after 2025 have been much more difficult to move.

That difference may be more important than the overall decline in visibility because it suggests that older and newer books are not starting from the same place.

The box set is the example I keep coming back to

My Absolution of the Morning Star box set provides perhaps the clearest example. It was released in June 2025 and had what was, by the standards of my catalog, an excellent period of Kindle Unlimited reading. During July and August alone, readers consumed more than 26,000 KENP.

I would have expected that level of genuine reader activity to help a new product establish a broader organic search footprint. Instead, its visibility declined. Its Search Visibility Index fell from roughly 8.5 in June to 6.9 in July, 5.3 in August, and 3.4 in September. The percentage of tracked searches in which the box set appeared declined as well.

The point is not that reading no longer matters. The interesting part is that a relatively new book could receive substantial genuine reader activity without developing the durable search presence I had seen older titles accumulate.

I see weaker versions of the same problem with Duke Rhime of the Spire, also released in June 2025, and Koen: Quills from the Raven’s Nest, released that October. Koen has occasionally shown a strong response around reader activity, but it has been inconsistent. Neither title has developed anything resembling the broad search footprint my older books built during their earlier years.

The more I looked at the catalog as a whole, the more it appeared that these books had entered Amazon during a different discovery environment.

Older books may have accumulated something new books struggle to build

I have sometimes thought of my older books as being “grandfathered,” although I do not mean Amazon literally created a different set of rules for them. A better description may be accumulated historical search relevance.

A book like Dawn of the Lightbearer has been available on Amazon for years. During that time, Amazon has observed years of purchases, reading activity, searches, and other customer behavior associated with it. Amazon itself says that past sales history and the length of time an item has been listed can influence search results, so it is reasonable to believe that history has some continuing value.

If those historical relationships persist, a new sale does not necessarily need to establish what Dawn is or where it belongs. The activity may simply reactivate relevance the book accumulated years earlier. A newer book has no comparable reservoir. It has to establish those relationships within the current system while Amazon simultaneously becomes more personalized and more focused on customer intent and behavior.

This leads to a distinction that I think may matter to KDP authors. An established title may need enough activity to reactivate search relevance it already possesses, while a newer title needs enough activity to create that relevance in the first place. If the threshold for doing that has risen, new books from independent authors face a much steeper climb.

I cannot prove that this is how Amazon’s internal system works, and I do not want to turn an inference into a claim about code I cannot see. What I can say is that it fits the behavior of my catalog considerably better than assuming my older books all retained an unusual ability to rebound while several newer titles independently failed to develop the same kind of search presence.

Jezelle provides the most useful counterexample

One of my newest books, Jezelle: Thief of Forks, eventually did develop substantially better Amazon visibility, but the circumstances were unusual. The book received significant exposure outside Amazon when it was included in MIT’s Summer 2026 recommended reading list, resulting in a concentrated burst of genuine purchases. Amazon search did not create that readership; the readers came from elsewhere.

Before that period, Jezelle had very little organic search presence. Its Search Visibility Index was roughly 2.9 in June. It rose to approximately 4.7 in July and 6.3 in August, while the book began appearing more frequently for relevant fantasy and grimdark searches.

The interesting observation is not that sales helped. The occasional isolated sale had previously done very little for Jezelle. What changed was the concentration of legitimate reader activity, after which the Amazon search response became substantially stronger.

That raises the possibility that the threshold for establishing organic relevance has increased. A slow trickle of purchases may no longer be enough for a new title to build meaningful momentum, while concentrated genuine demand can still get Amazon’s attention.

My data cannot tell me whether the crucial factor is sales velocity, conversion, traffic source, number of customers, the difference between purchases and KU readings, or some combination of these. The contrast between Jezelle and the box set does suggest that total activity alone does not explain what happens next. I did not promote/market the box set at all. I released it on a whim and depended solely on organic traffic, while Jezelle received an outside push. The difference there mattered.

The discovery funnel may have changed

For years, much independent-publishing advice has treated Amazon as part of the discovery process. Publish a professional book, choose appropriate categories and keywords, gather some reviews, and give Amazon enough information to connect the book with potential readers. Organic search could then provide at least some opportunity for strangers to encounter it.

My data make me question how reliably that works for a new release today.

Keywords and categories still matter, but I increasingly think of them as infrastructure rather than ignition. They tell Amazon what a book is and where it belongs. They may no longer be particularly good at generating the initial demand that gives Amazon a reason to expose the book.

Amazon’s own description of its evolving search technology reinforces that concern. The company says discovery is increasingly about understanding a shopper’s intent rather than simply matching keywords, while personalization incorporates customer searches, purchasing behavior, preferences, and other signals.

For a small KDP author, that creates an ugly feedback loop. A new book makes few sales, giving Amazon relatively little customer evidence to support greater exposure. Because it receives little exposure, fewer shoppers encounter it, which means fewer sales and still less opportunity to accumulate the behavior that may help establish relevance. This is compounded by the fact that if you make few sales, you will have few reviews, giving you little credibility. It is a terrible catch-22.

If my interpretation is correct, the discovery funnel for many new books increasingly resembles external discovery → genuine reader activity → Amazon responds → additional organic exposure, rather than relying on Amazon search → reader discovers book → sale as the starting point.

That may be manageable for authors with large mailing lists, established audiences, substantial advertising budgets, or regular media exposure—essentially those with the backing of traditional publishers. For small independent authors trying to build an audience from almost nothing, it is a significant disadvantage.

This may explain why 2026 has felt so different

This brings me back to why I started digging through these spreadsheets. My sales have collapsed this year, even though I have continued to write, publish, improve my catalog, and seek readers. I kept trying new things because I assumed there had to be something I could fix.

The data does not allow me to say Amazon caused every lost sale, but it does show that several books simultaneously became substantially less visible through searches where they had historically appeared. Lower sales are hardly surprising if fewer potential readers have the opportunity to encounter the books organically.

Jezelle makes the contradiction especially clear. When the book was actually exposed outside of Amazon, it sold, and its visibility on the platform improved afterward. The increasingly difficult part appears to be reaching those first readers without Amazon’s help.

For me, that changes the diagnosis. A bad sales year and a collapse in discoverability may not be two separate problems. They may be two sides of the same problem.

What I am changing because of this

I do not think keywords are useless, nor do I think categories, descriptions, covers, reviews, and professional presentation have stopped mattering. A book still needs to give Amazon accurate information and give customers a reason to buy it.

What I am changing is how much I expect those things to generate discovery on their own.

I am less willing to assume that Amazon search will eventually rescue a new release if I wait long enough and keep refining its keywords. I am also paying much more attention to concentrated, legitimate promotion. Jezelle suggests that delivering a meaningful group of real readers to a new book within a relatively short period may accomplish more than spreading the same activity over months.

My older books may also be more valuable than their current sales suggest. If accumulated search history allows them to recover visibility more readily, they can serve as entry points into the rest of my catalog rather than being treated simply as aging backlist titles.

The biggest change, however, may be what I stop doing. I have devoted an extraordinary amount of time to keyword research and monitoring because I believed incremental improvements in Amazon search could compound into discoverability. The irony is that years of doing that gave me 556,000 observations, which allowed me to see why it may no longer be worth doing at anything close to the same intensity. If my interpretation is right, the most valuable product of all those hours may ultimately be this article and the realization that I need to spend more of that time finding readers somewhere else (and writing!).

None of this means manufacturing activity, coordinating purchases, creating artificial KU reading, misleading customers, or attempting to manipulate Amazon. I am interested in what happens when genuine readers choose a book and how Amazon responds afterward. The goal is not to game the system but to stop pouring time and money into strategies based on assumptions that may no longer be true.

There is one enormous limitation

More than half a million observations sounds like an enormous dataset, and in one sense it is. It was gathered by roughly two hours of work a day, every day, over the course of four years. That is roughly 121 days of my life, or roughly a full year of full-time work. Very few authors are foolish enough to have spent that much time collecting this much Amazon search information. Statistically, however, I still have one author and one catalog.

My books occupy overlapping areas of fantasy, dark fantasy, horror, grimdark, and related genres. Amazon’s systems are complex, the competition is constantly changing, and there are countless variables I cannot observe. I cannot take what happened to my books and declare that every KDP author experienced precisely the same thing.

What makes the dataset useful is that it spans the period before, during, and after the apparent change. It includes books released in different years, titles with very different sales histories, and enough repeated observations to reveal patterns that would be invisible if I were simply looking at this week’s bestseller rankings.

Taken together, the evidence suggests three things to me: 1. Organic search visibility across much of my catalog suffered an unusually synchronized decline beginning around late 2025 and becoming particularly severe during the first half of 2026. 2. Books with years of established Amazon history appear more capable of recovering visibility from modest activity than several books released during or after that shift. 3. New books can still gain meaningful visibility, but Jezelle suggests they may now require stronger and more concentrated genuine demand before Amazon begins helping rather than merely waiting for Amazon to create that demand.

Amazon has not confirmed that those things are connected. I cannot prove the mechanism from outside the company. But Amazon has confirmed that sales history, popularity, and listing age influence search while simultaneously describing a broader transition toward personalized, intent-based, AI-assisted discovery and recent customer engagement.

That is enough for me to take what I am seeing seriously.

I would like to know whether other authors see it too

This is ultimately why I am publishing the analysis rather than leaving it buried in my spreadsheets. I know I am not the only KDP author who has looked at 2026 sales and wondered what the hell happened.

If you have been publishing for several years and have retained historical keyword data, I would be interested in how your books performed during the second half of 2025 and the first half of 2026. Compare stable groups of the same keywords rather than changing keyword lists, and look at whether titles released earlier respond differently to reader activity than books released more recently.

If my catalog is unusual, that is useful information because it means I need to keep looking for another explanation. If authors in unrelated genres, with different audiences and completely different publishing histories, find a similar breakpoint, however, then this becomes considerably more interesting.

I cannot tell you exactly what Amazon changed. What I can say is that after nearly four years, eleven books, and roughly 556,000 keyword observations, the Amazon search environment my newest books entered does not behave like the one in which my older books established themselves.

Given what has happened to my sales this year, that distinction is no longer academic.

Cheers!