What 1,458 Gwangjang Reviews Say the Videos Don't

Search for Gwangjang Market and you will find someone telling you not to go. The phrasing repeats: it is dirty, it is overpriced, and Koreans do not eat there anymore. We counted how much of that exists in our advice corpus — the Reddit threads, Quora answers and YouTube comment sections where travelers pass each other information.

Of the 411 items that mention the market, 46 argue against going. The number that matters is the next one: those 46 trace back to four YouTube comment sections and one forum thread. Six of them sit under a single video, and one line was posted twice on the same day:

That is not a verdict with a thousand voices behind it. It is a handful of comment threads, loud because the format rewards a strong opening line.

The trade-off: a small number of sources does not make a claim false. Some of what those comments say does show up in the review data — at a size the videos never state.

Against those 46 items we have 1,458 rated reviews of the market itself, collected without filtering by rating so the proportions mean something. They average 4.30.

One or two stars account for 6.7% of visitors. That is the size of the disaster the videos describe, and it is worth knowing before you decide the market is either a must-see or a scam.

Sixty of the reviews use the word worth, and we read all sixty. Fifteen are about something else: a dish, a queue, a donut. Of the 45 that give a verdict on the visit itself, 34 say Gwangjang Market is worth it, 7 say so with a condition (once, if you are nearby, or to look rather than to eat) and 4 say it is not, or only barely. Eleven of those 45 verdicts come from people who rated the market three stars or lower, and seven of the eleven still say go.

The trade-off: a 4.30 average does not promise you a good meal. It tells you the odds, and the odds are not what a thumbnail saying “DO NOT GO” implies.

To find what separates a low review from a high one, we compared how often each complaint appears in the 242 reviews of one to three stars against the 1,216 of four and five. All 242 low reviews were read in full, in the language they were written in, and sorted by hand. The high reviews were searched in every language of the sample and every match was read.

Three complaints carry the low reviews: the food, the price and the dirt. One in five says the food was not good, almost as many call the market expensive, and 16.1% mention dirt, smell or hygiene. Among the high reviews each of the three stays under 4%.

Dirt and smell is the row that separates the two groups most. It is also the first word of the line the videos repeat, so its size is worth stating: 57 of the 1,458 reviews mention it, which is 3.9%. For price the figure is 87 reviews, or 6.0%. Seventeen of the 47 low reviews that call the market expensive are written in Korean, one more than in English.

The crowd is the weakest. It is the thing every video shows, and it is mentioned in 15.3% of low reviews and 9.0% of high ones. In plain counts, more happy visitors mention it (109) than unhappy ones (37). People who loved the market describe the same crush as people who did not.

One complaint is not in the video line at all: the stalls repeat. 14.5% of low reviews and 4.3% of high ones say most stalls sell the same few dishes. The same observation is a two-star review in one case and a piece of advice in another:

Three rows rest on fewer than ten high reviews, so they are shown with their counts and no multiplier. Complaints about a vendor’s manner appear in 23 low reviews and five high ones, and they read differently on each side:

Eighteen low reviews describe a problem with the bill itself: a dish they did not order, change that did not come back, or a price that was different when they paid. Two high reviews do. Only two of the eighteen are among the fifteen reviews of the next section, so these are mostly different people.

The trade-off: you can avoid the crowd by choosing your hour, read a price before you sit and look at a stall before you choose it. How the food tastes and how the vendor treats you are the two things you cannot check until you are already eating.

Fifteen of the low reviews mention overcharging and foreign visitors in the same breath. Nine of the fifteen were written in Korean, or by someone who introduces themselves as Korean. One of them, in English:

The subject comes up three times among the high reviews as well, and none of the three denies it. One five-star review notes that prices have risen and calls the markup a surprise. One, written in Korean, asks vendors to stop. The third is a regular who shrugs it off:

Both reviewers are describing the same market. What the data supports is narrow: a small group of reviewers, most of them local, warn visitors about pricing. It does not tell you how often a visitor is overcharged, because a review records who chose to write, not what happened at every stall.

The trade-off: fifteen reviews is not a pattern you can put a probability on. It is specific, it is repeated, and it comes from people with no reason to run down their own city — which is why it is here, with its size stated.

Reviews name dishes constantly. Comparing how often a dish appears in high reviews and in low ones does not tell you whether it is good. It tells you which way the mentions lean.

Tteokbokki is the only dish we tracked that is named more often in low reviews — 6.6% against 5.6%. Six of the sixteen low reviews that mention it are written in Korean. The lean is small. What is larger is the verdict of the people who ate it: 7 of the 33 reviewers who say how the tteokbokki tasted did not like it, the highest share among the five dishes we read review by review.

Bindaetteok, the dish the market is known for, leans the other way, and so does gimbap. Mandu leans the same way more weakly. Kalguksu does too, on only four low reviews, which is too few for a multiplier. Yukhoe shows no lean at all: raw beef is mentioned at almost the same rate by people who rated the market high and people who rated it low. It is named in 118 reviews, and 80 of them are written in Japanese or Chinese.

The trade-off: these figures count mentions. They cannot tell you how many people ate each dish, so they cannot tell you which one is least likely to disappoint. A dish nobody names may be excellent and unordered.

Figures come from the SeoulFoodTour research corpus: 25,529 items after noise removal, of which 16,622 were processed for topic, sentiment and claims. The corpus keeps two kinds of evidence apart, and this article does too.

Experience — 1,458 rated reviews of Gwangjang Market collected without filtering by rating: 1,106 from Google Maps and 352 from TripAdvisor, dated March 2019 to September 2026, with 1,170 of them from 2025 or 2026. Every percentage in this article is taken from these 1,458 and from nothing else: 242 reviews of one to three stars and 1,216 of four and five.

A second pass aimed at low ratings added 283 reviews. They are used to find quotations and are never counted: a sample collected to be negative would pull every proportion down.

Advice — 411 items from Reddit, Quora and YouTube that mention the market by name, with their source thread or video recorded. That is how the 46 critical items were traced to five origins.

The worth-it count. Sixty reviews in the sample use the word worth or a Korean equivalent. Each was read in full and sorted by hand. A review counts as a verdict only when it speaks about visiting the market, not about a dish or a queue.

Language. 116 of the 1,458 reviews (8.0%) are written in Korean, and 36 reached us as machine translations. All of them are counted. Every quotation here comes from a review written in English, checked against the original record, because a translation is not someone’s words.

How rows were built. Complaint rows are defined review by review, not by a search pattern. All 242 low reviews were read in full, in the language they were written in, and each was sorted by hand into the rows it belongs to; a review can fall in several. For the 1,216 high reviews, two wide searches in every language of the sample selected 510 reviews, which were read and sorted the same way. The other 706 were not read, so the high-review counts are floors and every multiplier is a ceiling. The crowd row counts a review that calls the market crowded or packed, in any tone; a review that only calls it busy or lively is not counted. An earlier version of this table searched in English and Korean only, and every row came out too low: dirt and smell stood at six low reviews and is now 39. Dish rows search for the name of the dish in every language in the sample; an earlier version of this table searched in English and Korean only and missed most of the Japanese and Chinese reviews. A review that says only “pancake” is not counted as bindaetteok, so dish counts are floors. A multiplier is printed only where both sides have at least ten reviews; every other row shows its counts and nothing more.

The row on overcharging of foreign visitors. This row was first built with the English and Korean patterns and gave eleven low reviews. A wider search in every language of the sample, followed by reading, found four more: one in Japanese, two in Korean and one in English that named the subject in other words. The row now stands at fifteen and is defined review by review.

What this article does not do. It states no prices and no opening hours. Both change, and a review corpus records what was true on the day someone wrote it, not what is true today.