This site contains affiliate links. We may earn a commission at no extra cost to you.

Ratings and Reviews on DLsite: Judging a Doujinshi Without Japanese

You are looking at a listing. There is a row of stars, a number next to it, and below that two or three blocks of Japanese written by people who have already read the work. Everything on that part of the page was produced by other buyers, which makes it the only part of the page not written to sell you anything — and all of it is in a language you cannot read.

The temptation is to reduce it to the star figure and move on. That is the one reading that will mislead you most often, because in this corner of the catalogue a rating is answering a different question than you think it is.

This guide covers what each number on a listing is actually counting, why a low score or no score at all is usually not a verdict, the Japanese words that recur in these reviews and what they signal, what machine translation does to them, the signals that beat the rating outright, and the order to check things in before you spend money.

What the numbers on a listing are actually counting

Four separate quantities get collapsed into “the reviews” in most people’s heads. They come from different actions and carry different amounts of information.

The star count and the written-review count are not the same population, and the gap between them is enormous. Clicking a score takes a second; writing takes a quarter of an hour. A work can carry dozens of scores and a single written review, and that single review is one person’s evening rather than a consensus.

Most doujin works carry very few of either. This is a catalogue of self-published releases from small circles, not a storefront of mass-market titles. An average over five scores is arithmetic performed on almost nothing: one strongly felt opinion moves it by a fifth of the scale. Before you read a score, read the number of people it is drawn from, and treat anything in single figures as an anecdote wearing a decimal point.

Rank and sales figures answer “how many”, not “how good”. A ranking position is a sales measurement over a window, so it favours whatever was released into a busy week with an established circle name attached. It is genuinely useful for one purpose — telling you a circle has an audience — and useless for predicting whether you will like the book.

The release date reframes all of it. A work published three days ago has had three days to accumulate scores. On a new release, an empty rating area means nothing whatsoever, and reading it as a warning will cost you works you would have enjoyed.

Why a low score, or no score, is usually not a verdict

Here is the part that changes how you buy. A rating in this genre compresses two questions into one number: was the work well made, and did it go the way the reader wanted it to go. You cannot separate those from the outside, and the second one dominates.

These are works with strongly held expectations attached. Some readers want a situation restored by the last page; some want the opposite and consider a restoration a wasted purchase. Both groups leave scores on the same five-point scale. A work that does exactly what it said on its tag row, and does it well, will collect low scores from people who wanted the other thing — and those people are not wrong about their own evening, they are simply not answering your question.

So a mixed score is often a signal about direction rather than quality. Where several low scores cluster on a work whose artwork is plainly competent, the usual explanation is a mismatch of expectation, and the fix is to find out which direction rather than to walk away. How a doujinshi ends covers the labels that declare the direction in advance.

Three more reasons a score misleads, all structural:

Selection effect. The people rating a work are overwhelmingly people who chose it, from a page that told them what it was. That population is already filtered for enthusiasm. It pushes averages up across the board, which is why a 4.5 in a small catalogue carries less information than a 4.5 somewhere with mandatory reviews.

Series position. An instalment can be rated on how it serves the sequence, not on its own merits. Scores that read as harsh are sometimes readers saying a story did not advance, which is not a complaint that applies to you if you have not read the earlier parts.

Format expectations. The same score means different things across formats, because buyers judge an image set on quantity and variety and a manga on story. Read the format field first; CG collections and manga sets out how differently the two are put together.

The one thing a poor rating does reliably flag is a work that failed on something everybody can see: pages that do not match the cover, a file problem, or a release far shorter than the price suggested. Those complaints show up as repetition — the same handful of Japanese words appearing across several reviews — and repetition is a signal you can spot without reading a single sentence.

The words that recur in reviews, and what they tell you

Doujin reviews run on a compact shorthand. Once you can recognise about fifteen strings of characters, you can extract most of the useful signal from a review block without translating any of it — you are pattern-matching, not reading.

How to use this in about a minute, without reading Japanese:

1. Copy a string from the table — you are copying characters, not parsing them
2. Open the listing and use find-in-page (Ctrl+F, or Cmd+F on a Mac)
3. Read the hit count, not the sentences. Three hits for 尻切れ across separate reviews is a pattern; one is a person
4. Run 表紙詐欺, 尻切れ and 地雷 first. Those three catch the complaints that apply regardless of taste

Repetition is the whole method. A single review tells you about one reader. The same term appearing in three of five reviews tells you about the work, and it survives the language barrier completely intact, because you are counting rather than reading.

Two of these deserve emphasis. 表紙詐欺 is the most useful complaint in the list for an overseas buyer, because it is the one failure the preview images are designed not to reveal. And 地雷 is personal by definition — it marks the thing that particular reader could not tolerate, which may be exactly the thing you came for. Note it, then find out what it was pointing at before you act on it.

What a translator does to a review, and what it cannot do

Paste a review into a translation tool and you get fluent English. The fluency is the problem: it reads like a complete rendering of the source, and the specific parts you needed are the parts that did not survive.

Genre shorthand flattens. 逆転無し becomes a phrase about reversals that reads like a note on plot twists. 抜ける comes back as something vague and physical, or vanishes. 地雷 arrives as a literal landmine. These terms carry precise meaning inside the hobby and almost none outside it, and a general-purpose translator has no reason to know which register it is in.

Sentence fragments and slang defeat it further. Reviews are typed quickly, often without subjects, full of abbreviations and drawn-out vowels. This is the worst-case input for machine translation, and it fails silently rather than visibly — the output is grammatical, confident and partly invented.

Translate single terms, not paragraphs. One word in isolation usually comes back close enough to be useful. A paragraph of shorthand comes back as something readable that no longer matches what was written. The same rule applies to the description itself; the Japanese words on a product page covers that vocabulary.

There is one more limit worth stating plainly, because it is why you will not find translated reviews republished here. A review is a piece of writing owned by the person who wrote it. Running someone’s review through a translator and publishing the output is reproducing their work, not reporting on it. What we do instead — where we do it at all — is read the Japanese and describe, in our own words, what several readers kept saying. That is a summary of a pattern rather than a copy of anyone’s text, and it is the only form of this that is ours to publish.

The practical version for you is the same shape: look for what repeats, not for what any one review says. You do not need the sentences to see the pattern.

The signals that beat the rating

If you rank the evidence on a listing by how much it should move your decision, the star figure is not near the top.

The samples. Actual pages from the actual work, chosen by the circle. They will not show you the ending, but they show the artwork, the panel density and where the text sits — and those decide most of what your evening will feel like. Preview pages before you buy covers how to get the most from them.

The tag row. Chosen from a fixed list rather than written as prose, which is why it survives translation better than anything else on the page. Presence commits the circle to something; absence commits them to nothing.

The circle’s other listings. The strongest predictor available, and it costs nothing to check. Circles are consistent in art style, in length and in how much text they put on a page, far more consistently than any one work’s rating predicts the next. If a circle has four works and you liked one, the other three are a better bet than a stranger’s five-star release. What a doujin circle is covers how to follow that thread, and RJ numbers explained covers identifying works when titles are ambiguous across languages.

Format and length. What you are buying, and how much of it. How many pages a doujinshi is covers the trap where a description states two page counts and the smaller one comes first.

How much Japanese the work needs. This is the one no reviewer will ever mention, and the reason is structural: for the people writing those reviews, it is not a variable. Nobody rates a book on how much of it they could read. So we measure it instead — by running OCR over the free sample images and keeping only the numbers, including how confident the recognition was, because text that OCR struggles with is handwritten or drawn into the artwork, and that is exactly the text a translation app will fail on too. The measurement is published per work with the reasoning shown. The 174 works here sort into English, minimal text, OCR OK and text-heavy — 14, 66, 50 and 44 works, with the counts moving as works are added.

Start from works with an English version if you want the problem removed entirely, minimal text if you are happy to let the art carry it, OCR OK if you are willing to point a translation app at speech bubbles, and treat text-heavy as the band where a glowing review may describe an experience you cannot have.

The order to check things in

Roughly a minute, none of it requiring Japanese:

1. Look at how many people rated it, before the average. Single figures is an anecdote; ignore the decimal
2. Check the release date. A recent release with no ratings is not a warning, it is a new release
3. Find-in-page for 表紙詐欺, 尻切れ and 地雷. Count hits. Repetition across reviews is signal; one hit is a person
4. Read low scores as direction, not quality, and confirm the direction from the tags rather than from the score
5. Open the samples, and look at where the text sits as well as what the artwork is doing
6. Check the circle’s other listings. Consistency within a circle beats a rating on a single work
7. Check the text band last, because a well-reviewed work you cannot follow is still a work you cannot follow

We do not publish ratings, rankings or recommendations on this site, and that is a deliberate omission rather than a gap waiting to be filled. We have no sales data and no rating data of our own, so any ranking here would be a reproduction of someone else’s numbers or an invention. What we do have is a measurement nobody else takes, made the same way for every work, with the figure and the reasoning published so you can disagree with the threshold.

Use the storefront’s reviews for what they genuinely carry — repeated complaints, the circle’s standing, and whether a story stops rather than ends. Then decide the thing they will never tell you, which is whether the book in front of you is one you can actually read.

Listing elements are described by what they do rather than by their exact position or label, because storefront layouts change and not every listing shows every figure. The Japanese terms above are reader shorthand as it is actually used in doujin reviews; no review text is reproduced or translated here. Readability figures are measured from free sample pages and are published per work with the measurement behind them; counts are read live from the catalogue and change as works are added.