What this gives you: for any of 168 countries, how much of its own news mentions Indonesia at all, whether that has moved in nine years, and how often its feed goes silent — plus which of 9 named events the world's press actually turned up for.
What it does not give you: a count of anything, an audience, or a reading of whether any of it was true. The archive behind this shrank by more than three quarters over the record, so a count would be worthless and every figure here is a share. There is also no world map, and there cannot be one from this work — the reason is further down, and it is more interesting than a map would have been.
And the checks are not finished. Two of this case's four stand PENDING rather than passed. What each one actually says is set out below, in the same plain words as everything else.
The neighbours, and then a very long tail
Set Indonesia's own press aside — a country reporting on itself is not attention from the world, and it gives Indonesia 33.8643% of its own news, which would flatten every other bar on the figure. What is left is 168 countries, and they divide unevenly enough that the shape is the finding:
- 7 countries give Indonesia more than 2% of their own news. Six of the 7 are its Southeast Asian neighbours; Singapore leads at 4.7976% — about 18.2 times the 0.2642% of the middle country in the list. The seventh is North Korea, at 2.5968%, and its feed is one of the unsteady ones: it reports nothing about Indonesia at all in 5 months of the record. Switch to the third view and it changes colour.
- 124 of 168 are below half of one per cent. For most of the world, Indonesia is a rounding error in the day's news.
- The former colonial power is 79th, at 0.2885% — behind the United States (0.4822%, 47th) and the United Kingdom (0.34%, 67th), and barely ahead of France (0.2257%, 102th). China, at 0.9839%, is the highest of any large distant economy at 17th.
Read across the whole record, articles written in English give Indonesia 0.6567% of their attention. That is roughly the world average and nothing like the neighbourhood's — which is worth holding on to, because English is the language most of the world's foreign-news commentary is written in.
Why a count of articles would have been a lie
Here is the number that decided the shape of this page. The archive these measurements are drawn from is not a fixed thing. In 2017 it took in about 627,594 articles a day; by 2026 it was taking in 142,709 — a fall of -77.3%. Nothing about the world changed to cause that. The archive changed.
- everything the archive monitored
- articles mentioning Indonesia, counted
- the same articles as a share
All three start at the same place, 2017 set to 100, so they share one scale. The grey line is the archive itself. The coral line is the raw count of articles mentioning Indonesia, which falls with it. The bright line is those same articles measured as a share of what was monitored, and it goes the other way.
So the same record answers "is the world paying more attention to Indonesia?" in two opposite ways. Counted raw, coverage fell -70.8%. Measured as a share of what the archive actually looked at, it rose 28.9%. Ranked against each other day by day, the two agree at 0.0836, where 1.0 would be perfect agreement and 0 would be none. They are not two readings of one thing. One of them is measuring the archive.
The shares had to be tested too, not just preferred
"Use a share instead" is easy to say and it is not automatically safe. If the outlets that dropped out of the archive were the ones that wrote about Indonesia, a share could drift just as badly. So two checks, both on the same 116 months:
- The middle country barely moved. Take the country that sits in the middle of the list each month. In 2017 it gave Indonesia 0.2296% of its news; in 2026, 0.2256% — while the archive underneath lost three quarters of its volume. Ranked against the size of that archive month by month, the two line up at 0.044: no relationship at all.
- The order stayed the same. Rank all 169 countries on the first three years of the record. Rank them again on the last three, drawn from an archive a quarter of the size. The two rankings agree at 0.8608. Two very different archives put nearly the same countries in nearly the same order, which is what a measurement that survives its own instrument looks like.
One exception, and it is worth naming rather than burying: Indonesia's own share moves against archive size, at -0.6521. Dividing by each country's own monitored articles removes the effect of the global archive shrinking. It does not remove a change in which outlets inside one country were monitored, and Indonesia's own feed changed a great deal. That is one more reason its bar is not in the field.
When does the world's press actually turn up?
A ranking says who reads about Indonesia on an ordinary day. It does not say what brings the rest of the world in. This case wrote down 9 dated events before looking at the data, and its review re-ran every one of them twice: once on foreign publishers only, once on Indonesian publishers only, each against its own normal level in the four weeks before.
- Palu earthquake & tsunami2018-09-28
- Anak Krakatau tsunami2018-12-22
- post-election riots, Jakarta2019-05-22
- first COVID-19 cases announced2020-03-02
- Kanjuruhan stadium disaster2022-10-01 two measurements disagree
- G20 Bali summit2022-11-15
- presidential election2024-02-14
- PDNS ransomware attack2024-06-20
- Aug–Sep 2025 protests2025-08-28
- the world's press, against its own normal
- Indonesia's own press, against its own normal
A bar at one times normal means that group of publishers wrote no more about Indonesia than usual. The two are on the same scale and neither depends on the archive's size, because each is measured against itself.
The pattern is consistent and slightly uncomfortable. The world's press rose to at least twice its normal level for 6 of the 9. Indonesia's own press stayed near its normal level for 7 of 9 — which makes sense: it already writes about Indonesia every day, so a national event does not change its proportions much. The G20 Bali summit is the extreme case: foreign coverage at ×8.747 while Indonesia's own press sat at ×1.185. The world shows up for a summit far more visibly than the host country does.
One row disagrees with itself, and it is marked. For the Kanjuruhan stadium disaster, this measurement puts foreign coverage at ×0.746 — below normal — while the case's other, independent measurement of the same days put it at ×1.7854, above normal. That is 1 of 9 rows where the two ways of measuring point in opposite directions, and it is left visible rather than resolved by picking the one that reads better.
What the coverage was about, era by era
The same normalisation works one level down: for each month, what share of the Indonesia coverage used the words of a particular subject. Pick an era.
The six subjects with the largest share of Indonesia's coverage in 2017–2019. Average sentiment of that era's articles: -0.294. Left out for want of data: Coal & mining.
The six subjects with the largest share of Indonesia's coverage in 2020–2021. Average sentiment of that era's articles: -0.5797. Left out for want of data: Coal & mining.
The six subjects with the largest share of Indonesia's coverage in 2022–2023. Average sentiment of that era's articles: 0.2185. Left out for want of data: Coal & mining, Floods, Papua.
The six subjects with the largest share of Indonesia's coverage in 2024. Average sentiment of that era's articles: 0.6732. Left out for want of data: Coal & mining.
The six subjects with the largest share of Indonesia's coverage in 2025–2026. Average sentiment of that era's articles: 0.4456. Left out for want of data: Coal & mining.
The pandemic is the only subject that ever dominates — 29.56% of everything written about Indonesia in 2020–2021. Nothing else comes close, before or since. Elections are the only subject in the top six of every era, and in the most recent era the three that lead are elections, floods and protests, within 3.36% to 2.85% of each other — which is to say that recently there has been no single story.
Some subjects are missing months, and the gaps are not random. 5 of the 11 subject series have holes, and each hole is a different whole calendar year: protests are missing 2017, the new capital 2019, Papua 2022, floods 2023. Coal and mining is worse — it has 13 of 116 months and nothing after early 2018, which is a collection failure, not the end of coal news. It is therefore excluded from every era above rather than shown with an average that looks comparable and is not.
The darkest day in ten years, and what the record does not know
The other thing this case measures is tone: for each day, the average sentiment of the words in the articles — how warmly or bleakly Indonesia was being written about. It is worth being precise about what that is, because it is the number most easily misread on this whole page. It is a property of the writing, not of the country. A dark day means the articles were phrased bleakly. It does not mean something bad happened, and a bright day does not mean something good did.
Across 3,504 days the average sits at -0.0222 — very slightly on the bleak side of neutral, with 1,700 days below it. The extremes are far more interesting than the average.
| Day | Tone | What the record says about it |
|---|---|---|
| 2018-05-13 | -5.7912 | No label anywhere in this work. The record says only what the words were: the Terrorism query took 35.6% of that day's coverage, its highest of the whole record. |
| 2022-10-02 | -4.2395 | Kanjuruhan stadium disaster — an event written down before the data was looked at, whose own tone low fell on this day. |
| 2018-12-23 | -4.2355 | Anak Krakatau tsunami — an event written down before the data was looked at, whose own tone low fell on this day. |
| 2021-03-28 | -3.7089 | No label anywhere in this work. The record says only what the words were: the Terrorism query took 33.4% of that day's coverage, its highest of the whole record. |
| 2018-09-29 | -3.4947 | Palu earthquake & tsunami — an event written down before the data was looked at, whose own tone low fell on this day. |
| 2019-03-17 | -3.1094 | No label anywhere in this work. The record says only what the words were: the Papua query took 13.6% of that day's coverage, its highest of the whole record. |
| 2017-06-04 | -2.9074 | No label anywhere in this work. The record says only what the words were: the Terrorism query took 37.9% of that day's coverage, its highest of the whole record. |
| 2020-01-02 | -2.814 | No label anywhere in this work. The record says only what the words were: the Floods query took 40.6% of that day's coverage, its highest of the whole record. |
| 2021-12-05 | -2.69 | Nothing. No pre-registered event, no subject peak. |
| 2020-04-04 | -2.6656 | Nothing. No pre-registered event, no subject peak. |
| 2024-08-17 | +2.5636 | The 17th of August — Indonesia's Independence Day. |
| 2018-09-02 | +2.3735 | No label anywhere in this work. |
| 2024-12-01 | +2.1907 | No label anywhere in this work. |
| 2025-08-17 | +2.0896 | The 17th of August — Indonesia's Independence Day. |
| 2026-07-07 | +1.9336 | No label anywhere in this work. |
| 2019-08-17 | +1.9162 | The 17th of August — Indonesia's Independence Day. |
The 10 bleakest days on the record and the 6 brightest. Bars are the tone reading itself, on one scale.
Two things stand out and both are honest. 3 of the 6 brightest days in the archive fall on the same calendar date — the 17th of August, in 2019, 2024, 2025 — which is Indonesia's Independence Day. Three separate years landing on one date is the kind of pattern that argues the measurement is picking up something real.
And the bleakest day of the entire record, 2018-05-13 at -5.7912, has no label anywhere in this work. Only 3 of the 10 bleakest days do, and they have one only because the pipeline named that event in advance and its own tone low landed on that date. For the rest, all the record carries is which words spiked: on 2018-05-13 the terrorism query took 35.6% of the day's coverage, the highest it ever reaches. The archive does not record what happened. It records how it was written about, and there is a real difference. Not one headline or link survives anywhere in this case's output — 0 of them across every extreme day — so naming the event behind that date would have meant inventing it, and it is not named.
Why there is no map on this page
A world map shaded by each country's attention is the obvious way to draw all this, and it was checked before it was drawn. It cannot be built from this work: the country curves carry 169 country names and no code, no centre point and no outline. Borrowing outlines from somewhere else and presenting them as part of the finding is a different kind of dishonesty from getting a number wrong, and it is the more persuasive kind.
There is one geographic file — a grid of places inside Indonesia that the news mentioned, 1,695 cells of it — and that was checked as a figure in its own right and abandoned as well. Of its 25 most-mentioned cells, 6 are not what they are labelled, and between them they carry 581,062 mentions:
| Labelled | Mentions | What it actually is |
|---|---|---|
| Bali, Jawa Timur | 235,519 | the island of Bali, placed in East Java at 112.25 E — Bali is 115 E |
| Antara, Aceh | 151,190 | ANTARA is the national news agency, not a town in Aceh |
| Pancasila, Riau | 60,783 | Pancasila is the national ideology |
| Riau, Indonesia (general) | 49,173 | the province of Riau, placed in the Java Sea at 1.75 S, 106 E |
| Lampung, Kalimantan Tengah | 45,356 | Lampung is in Sumatra, not Central Kalimantan |
| Sigit, Jawa Timur | 39,041 | Sigit is a personal name, not a town in East Java |
The second-brightest cell on that map would have been a mislocated Bali. A further 14.1% of all located mentions are pinned no more finely than to a whole country, which on a map means one pile in the middle of it, and 17% of the mentions are of places outside Indonesia altogether. A ranked field of bars is a duller picture and it is the one the data supports.
How good is the underlying data, according to people who checked?
Nobody has published a study of Indonesia's media profile in this archive, so there is no result here to compare against. What has been checked, repeatedly, is the archive itself — and a reader should weigh these before weighing anything above.
- Of the records this archive labels as protests, an audit found only about 21% of the valid ones described a real protest.
- Hand-checking a year of records put the accuracy of the event type at about 56% in English and 53% in other languages, with about 21% of records duplicating another.
- Against a hand-coded reference, its event counts agree at 0.222 on a 0-to-1 scale; against a rival machine archive, 0.317.
- Its tone score agrees with other sentiment dictionaries at 0.82 when a whole day is averaged, but only 0.5 on a single article.
- A different archive of the same kind agrees with a careful reference at 0.64 for a national yearly series, and 0.26 once you ask about a particular place in a particular month.
Read together those say one thing consistently: this kind of data is usable in aggregate and unreliable in the particular. That is why this page reports shares and eras rather than events, and why it will not tell you what happened on a given day in a given province. The instrument was never accurate enough for that question, and no amount of care on this end changes it.
What this cannot do
This is the part a government or an NGO should read before quoting any figure from it.
- Every attention figure is a share, and it has to be. The archive shrank 4.4 times over the record. Any absolute count — articles about Indonesia, mentions of a province, stories about an election — is uninterpretable across years, and the two counts on the figure above prove it by disagreeing with each other completely.
- Coverage is measured, monitoring is not. For 81 of 168 countries the press is recorded as mentioning Indonesia in no month at all at least once, and 23 are blank more often than not. There is no denominator published per country, so it is impossible to tell a country whose press genuinely ignored Indonesia from one whose outlets were barely monitored. East Timor, which shares a land border, is blank in 112 of 116 months and ranks 139th. That is not a finding about East Timor.
- Tone is about the writing. It is an average sentiment score of the words in the articles. Bleak coverage and a bleak reality are different things, and the record only has the first.
- One country is missing from the figure entirely, and nobody knows which. 1 of the 170 country series this case publishes arrived with a blank name — real data, 116 months of it, averaging 0.1536% — with nothing to say whose press it is. A bar that cannot be labelled, searched for or explained is worse than no bar, so it was dropped and this sentence exists instead of it.
- The two halves of the record only partly agree. Two independent measurements of the same coverage, over the same 3,504 days, line up at 0.635 where 1.0 would be perfect. Split by language, articles written in English agree at only 0.207 while translated ones agree at 0.719 — so the agreement is carried almost entirely by the press that is not in English.
- Who counts as "the world's press" is a hand-made list. The foreign-against-domestic split on this page comes from classifying 22,657 publishing sites by origin, using a curated list of 98 Indonesian domains plus anything ending in Indonesia's country suffix. That list covers 68.4% of the material; an Indonesian outlet on a generic domain and not on the list is counted as foreign, so the domestic side is understated by an unknown amount.
The four checks, and where they actually stand
This case set four checks before the analysis and publishes all four as they came out. The overall verdict is PENDING, which is neither a pass nor a failure — it means two of them cannot be decided from what was collected.
| Check | Stands at | In plain words |
|---|---|---|
| Shares, not raw counts | PASS | The archive these curves come from shrank 4.4 times over the record, so every attention figure on this page is a share and never a count. every attention chart is a share; raw counts appear once, in the methodology. |
| Events written down in advance | PENDING | Of 9 events written down before the data was looked at, 8 show up unprompted, 0 contradict the record and 1 cannot be decided. undecided: PDNS ransomware attack — the word series it needs was never fetched. |
| Two sources agree | PASS | Two independent measurements of the same thing agree at 0.635, where 1.0 would be perfect agreement; the check required 0.6. English-only articles agree at just 0.207 and translated ones at 0.719 — the agreement is carried by the press that is not in English. |
| Indonesia's press against the world's | PENDING | The split between what Indonesia's own press writes and what the rest of the world's press writes was never computed on this route, so this check reports and decides nothing. its own condition is a foreign-share figure that is absent from the record; the two daily series meant to carry it are 3,529 blanks each. |
The last of them matters for this page in particular. The check that was supposed to separate what Indonesia's own press writes from what everyone else's does was never computed on that route — the two daily series meant to carry it are empty from end to end. The split this page does use comes from a different direction, classifying publishers by origin, and it says that the Indonesian-published share of the material rose from 63.3% in 2017 to 76% in 2026. On that same split, foreign publishers wrote about Indonesia more bleakly than Indonesian publishers did in every one of the 10 years, by up to 1.484 points of tone. Which of those two is the distorted one is not something this record can settle.
What a government or an NGO can actually do with this
Three things follow, and the first two are usually conflated.
- Spend attention where attention already exists. Of the 169 countries in this record, Indonesia's own press accounts for 33.9% of coverage about Indonesia, while the median foreign country contributes 0.26%. A communications budget spread evenly across capitals is mostly spent in rooms where nobody was listening.
- Volume and tone are different problems. Daily tone in this record runs from -5.8 on its worst day to 2.6 on its best, and neither number tells you how much was written. A country that covers Indonesia constantly and negatively needs a different response from one that barely writes at all: the first is a narrative to engage, the second an absence to fill. One "attention" figure collapses the two.
- Use it as a baseline for detecting change, not as a scoreboard. Indonesia's own share of its coverage moved from 31.6% early in the record to 37.5% late — a shift of 5.9 points that is only visible because the record was already being kept. The level is far less informative than the movement, and a country whose coverage shifts sharply is worth a phone call.
What you must not conclude from this. Not that the countries at the bottom of the field do not care about Indonesia — most of what a low bar measures is which outlets in that country an archive happened to watch, and the third view exists to show you exactly where that doubt lives. Not that a bleak month is evidence that something bad happened: tone is a property of sentences, and this work has no way to check any of them against events. Not that attention is influence, or trade, or public opinion — nobody's readership is measured here. And not that a rise or fall in any figure on this page reflects a rise or fall in coverage in absolute terms; the archive underneath changed size by a factor of 4.4, and every number here is deliberately a proportion for that reason.
Could this run in daily operations?
UNPROVEN — none yet — and nothing here has failed
Every pipeline here is a batch backfill: it fetches a season or an archive, not a live feed. Anything operational means rebuilding the ingest for near-real-time arrival, whatever the verdict below says.
What stops it. No check failed, but two of four have not run. Unrun is not passed, so it cannot be called deployable either way.
What it would need. The two pending checks run. Its feeds are already near-real-time, so of everything here this is the one whose readiness is a question of evidence rather than engineering.
If you want the detail
The query battery, all 9 pre-registered events with their full signatures, the publisher study behind the origin split, and everything still uncertain are on the technical article, and the dashboard carries the daily curves themselves. Every figure here is read from the same records as those, so they cannot drift apart.
Words on this page, in plain language 4 terms
- GDELT
- A public archive of news coverage worldwide, machine-read for who and what is mentioned. It records what was reported, not what happened.
- baseline
- The dumbest sensible answer, used as the thing to beat — for example "this year looks like last year". A model that cannot beat its baseline has added nothing, however good its score looks alone. Careful: in the radar sections the same word means something else entirely — a reference level a signal is measured against, as in "3 dB below the cell's own baseline".
- median
- The middle value: half are higher, half lower. Unlike an average, one extreme case cannot drag it.
- PDNS
- Pusat Data Nasional Sementara, Indonesia's interim national data centre. It appears here as the subject of a June 2024 ransomware attack, which is one of the news events this case measures coverage of.
Coverage measurements from GDELT's open feeds, 2017-01-01 to 2026-08-30. Data vintage 2026-08-30. Written to be read without a background in statistics; nothing was rounded to make a point.
Found something wrong on this page? Report a correction — it opens a pre-filled issue — or email taufik.adi@openstudy.id. Corrections are credited by name in the errata, and one that changes a finding says so on the page.