What a 106,962-article audit reveals about Albania’s media environment

The expansion of Albania’s media audit from 4,536 to 106,962 articles strengthens the empirical challenge to soft-capture assumptions while sharpening the question of what EU evaluation frameworks actually measure.

Tirana Examiner

On 12 May, Buletin Intelligence published the follow-up to the study this publication engaged on 5 May. The original sample of 4,536 political articles across 66 days has been extended to 106,962 articles across 120 days, drawn from 47 outlets. The methodology has been improved on the dimension where the original was most exposed: the earlier audit measured share of voice through article titles alone, and the replication measures it through full article bodies. The corpus is 4.5 times larger and the principal methodological limitation of the original work is resolved.

What this is not is a strict replication. The 5 May piece engaged a sample running from 1 March to 5 May. The new study runs from 1 January to 1 May and overlaps by two months. The expansion is therefore methodological and temporal at once, and the appropriate reading is that the new findings are coherent with the earlier ones across a wider window rather than that the same window has produced the same answer.

On the primary measurement the answer is the same. Opposition share of voice across the full corpus is 68.2 percent against 31.8 percent for the ruling party. The 5 May audit reported a comparable ratio at the smaller scale and treated it as a finding requiring further measurement before it could be load-bearing. The further measurement has now occurred.

On the secondary measurements the picture is more useful than confirmation. The standard deviation across balanced sources, which was the most diagnostically uncomfortable feature of the original sample, drops from 8.2 points to 4.0 under full-text analysis. The Telegraf anomaly, which the original could not explain, resolves to the main cluster at 60.3 percent and turns out to have been an artifact of short titles rather than editorial line. The Herfindahl-Hirschman index for spokesperson concentration reverses, with the ruling party now showing marginally higher concentration than the opposition. The study acknowledges the reversal openly as a correction of a v2 conclusion that did not survive its own methodological improvement. A study that revises its own findings when the methodology improves is operating to a standard the field does not always meet, and the revision is worth naming because it is the kind of honesty the evidence base is entitled to expect from any monitoring service whose work is being read into a candidate-country file.

The tone analysis introduced for the first time runs critical-to-positive coverage of government at roughly 5 to 1 across the period. The study itself flags that quoted opposition criticism is not distinguished from editorial criticism, which bounds what the ratio can carry on its own. It sits consistently with the share-of-voice finding, and that is the weight it takes here.

These findings do not change the structural argument the 5 May piece made. They sharpen one half of it.

The structural conditions documented by Reporters Without Borders are not contested. The Index does not measure published output, does not stratify for share of voice, does not test the soft-capture framework’s primary empirical predictions on the dimensions where soft capture is supposed to be most visible.

The Buletin replication does. The predictions fail at the scale now documented.

The caveats the 5 May audit named carry forward. The study is online only. Television is not included, and the state audiovisual regulator’s March monitoring placed the Democratic Party at 60.12 percent of political television time, which is consistent in direction with the online finding and arrives through a separate methodology in a separate medium. The study does not weight by reach, and a 47-outlet count compresses differences between titles with audiences orders of magnitude apart. The study does not stratify for the topic domains where soft capture is most likely to operate, including major infrastructure procurement, construction-sector political ties, and the networks through which state advertising allocation runs. A country can present an open conversation about the prime minister and a constrained one about the proprietors of its largest mainstream outlets, and the empirical study does not exclude this possibility. The conclusion the evidence sustains is qualified but not vague: incompatible with hard capture, inconsistent with soft capture in the dimensions soft-capture frameworks themselves identify as primary, and silent on a residual set of dimensions that require their own measurement.

The political moment this finding enters is not neutral. On 26 May, the day of the Eighth Intergovernmental Conference, the Prime Minister published an open letter to Reporters Without Borders citing the same study and challenging the framework underlying the 2026 ranking. The letter is a political document and its rhetorical register is the Prime Minister’s own. One sentence in it, however, belongs to the file rather than to the polemic. “An essential part of that journey is, of course, both the progress achieved and the work still to be done regarding the media environment. And we will do it in full cooperation with the EU.” That is the position the evidence actually sustains. The 106,962 articles document a published-output environment incompatible with hard capture. The structural conditions documented by the Index require remedies that have not been delivered. Both propositions are true at once, and both belong to the conversation Albania is having with Brussels this week. The methodology cited in the letter exists independent of who invokes it. The acknowledgment of unfinished work cited above is what the file is owed from the government being evaluated, and what the institutions doing the evaluation are owed on the record.

The question the evidence puts to the evaluation regime is therefore unchanged in substance and sharper in scale. By construction, rank-order country summaries flatten evidence more complex than the ranking can display. The methodology to correct that compression exists. It has now been demonstrated, at 106,962 articles and with the principal limitation of the original audit resolved, that the methodology is feasible for the Albanian case. It is feasible for Serbia, Montenegro, North Macedonia, and Kosovo on terms no less rigorous. The comparative baseline a Chapter 23 evidence base requires is no longer hypothetical. One country has produced one half of it.

The other half is what the structural ranking already captures and what the audit was never going to displace. Ownership opacity, advertising allocation, SLAPP exposure, the precarity of regional journalists, the conditions under which difficult subjects can be covered without cost: these remain. The 5 May piece named them. The 12 May replication does not erase them, and the present piece does not ask anyone to forget them. The Prime Minister, on the record, has not asked anyone to forget them either.

The 83rd-place ranking captures one part of the picture. The empirical reconstruction, at the scale now on the record, captures another. The institutions evaluating Albania during this accession cycle have the evidence to consume both, and the disposition on the government side to expect that both will be consumed. Whether they will is the question the file now contains, and the country whose accession trajectory depends on the answer is entitled to ask it.