Amnesty International’s India reports: what fifty years of grading shows

Plain-language summary of the longitudinal CID analysis. CID ID: CID-0045 (provisional).

What This Report Is

Every year, Amnesty International publishes a report on human rights around the world. Each edition has a section on India. We pulled the India section out of every year we could find — 1973-74 through 2024-25 — and graded each one.

We did not grade whether India’s human rights record is good or bad. We graded Amnesty’s method. The question is simple: could someone outside Amnesty check the work?

What We Looked At

We found 49 India sections spanning about fifty years. We scored each against the Citation Integrity Dashboard rubric.

The rubric ignores whether we agree with Amnesty. It asks structural questions. Are the key terms defined? Can a reader verify the individual claims? Is the evidence independent of the people making the case? Does the report admit its own limits?

What We Found

Every single year scores Deficient. The grade barely moves in five decades.

The reason is the same in 1975 as in 2024. The India section has no codebook, so you cannot see the rule that decided which cases got included. It almost never sources its claims inside the text, so a reader has to trust that Amnesty got it right. And the evidence is Amnesty’s own field research, reported by Amnesty, with no independent check built in.

What changes over time is the news, not the method. The early reports focus on prisoners held without trial during the 1975 Emergency. The 1990s reports cover Punjab and Kashmir. The recent reports focus heavily on attacks against Muslims and Dalits. The events get heavier. The way the information is gathered and presented does not improve.

Amnesty does have real strengths, and the score reflects them. It is independent of the Indian government, and it refuses government money for its research. Its finances are open. Those are genuine credibility signals, and they keep the score out of the basement. But none of that fixes the core problem: the report itself does not show its work.

There is one small shift. Between 2000 and 2013, Amnesty listed references to its own fuller reports at the end of each India section. That gave readers a paper trail, even if it only led back to Amnesty. After 2014 those references moved out of the main text, so the section reads as plain assertion again. That single change is the only thing that ever nudges the grade — by one notch, and only briefly.

The Bottom Line

This is not a claim that Amnesty is wrong about India. It is a claim about how the yearly report is built. It reads like advocacy, not like research you can independently check. That was true in 1975, and it is still true in 2024.

The grade describes the document a reader actually holds — one country section in a global annual report — not Amnesty’s entire operation. A worldwide yearly report cannot carry a full codebook and a verification trail for every country it covers. So the thinness is built into the format. But built-in or not, it means the same thing for anyone trying to rely on the numbers: you are being asked to trust the source, because the source has not given you a way to check.