Visa Outcome Tracker — official data + community reports

Read this before using any number here. The US government does not publish visa refusal rates broken down by nationality and visa class together. It publishes refusal rates by nationality for B visitor visas only, and refusal rates by visa class only as a worldwide, all-nationalities figure. There is no official "F-1 refusal rate for India" or "H-1B refusal rate for Nigeria" anywhere. Any site showing you one has invented it. What we can't show you, and why →

Official data points

—

country-years and class-years, each with a source

Countries

—

nationalities in the B-visa panel

Fiscal years

—

FY2006–FY2025 (FY2017/18 unpublished)

Community reports

—

self-reported, self-selected

1 · Visitor visa (B) refusal rate by nationality

The only refusal rate the US publishes per nationality. This is the adjusted rate: (refusals − overcomes) ÷ (issuances + refusals − overcomes), counting each applicant once per fiscal year at final status, and excluding cases that needed a waiver of ineligibility.

Add a country

Or load a preset

Source: US Department of State, Calculation of the Adjusted Visa Refusal Rate for Tourist and Business Travelers Under the Guidelines of the Visa Waiver Program, FY2006–FY2025. One PDF per fiscal year. FY2017 and FY2018 are not published — the line breaks rather than interpolating across them.

2 · Refusal rate by visa class — worldwide

These are global figures covering all nationalities. They are not, and cannot be broken down into, per-country rates. They also use a different construct from the chart above — raw adjudication counts rather than the adjusted methodology — so the two sections are not directly comparable.

Fiscal year

Track one class over time

Refusal rate by class

Over time

Source: US Department of State, NIV Workload by Visa Category, FY2013–FY2024. Columns are Issued, Refused, Total Applications. In FY2024 three rows do not sum exactly (B1/B2 by +2, BBBCC and BBBCV by −1 each); the printed totals are kept as published rather than silently corrected.

3 · Community reports

Individual accounts from public forums and trackers, extracted into a common structure. This layer answers questions the official data cannot — which consulate, what the officer asked, which documents were demanded under 221(g) — but it is not a sample of applicants.

How biased is this sample? Measurably. People post about dramatic outcomes and about problems they need help with; quiet approvals and quiet refusals both go unreported, in different proportions. Treat counts here as texture, never as a rate.

Visa class

Consulate

Outcome

Minimum extraction confidence

Matching reports

—

after filters

Reported approved

—

share of matching reports

Reported 221(g)

—

administrative processing

Reported refused

—

outright refusal

Outcomes reported

Reports by consulate

Individual reports

Interview questions actually reported

Pulled verbatim from the reports above. Frequency here reflects what people chose to write down, not what officers most often ask.

4 · What this tracker will not show you

These are the numbers people most want, and every one of them is unavailable in public US data. We return nothing rather than an estimate.

Requested figure	Status	Why
F-1 refusal rate for India (or any single country)	Not published	Refusal rates by nationality exist for B visas only. By class, only a worldwide figure exists.
H-1B refusal rate by nationality	Not published	Same. USCIS's H-1B Employer Data Hub has no beneficiary nationality field at all.
Approval rate at a specific consulate	Not published	Monthly files exist by post and by nationality, but the two are never crossed, and neither carries refusals.
214(b) or 221(g) rate for a country or class	Not published	Refusal grounds are published globally with no class and no nationality breakdown.
Any refusal figure at monthly frequency	Not published	All monthly State Department files are issuances only.
H-1B lottery odds by nationality	Not published	Selection rate is released as a single global number in a news post.

The statutory reason is 8 U.S.C. § 1202(f), which makes consular records on visa issuance and refusal confidential and operates as a FOIA Exemption 3 withholding statute. Record-level requests are refused under it as a matter of course.

5 · Sources and method

Known limitations of this build

Official tables were extracted from PDFs via a language model, not a deterministic parser. The FY2025 nationality table was verified line-by-line against an independent published copy and matched exactly on all 191 values; the class-level tables were verified by checking every column against the source documents' own printed grand totals, which reconciled for all 12 years. Other years carry set-level checks only.
A printed 0.00% is genuinely ambiguous in the source — it means either every application was issued, or there were no applications at all. Those cells are flagged, not interpreted.
Two negative adjusted rates appear (Luxembourg FY2006, Kuwait FY2009). These are real: overcomes exceeded refusals that year.
Microstates swing between 0% and 100% between years on tiny denominators. The rate is real; it is just close to meaningless.
The community layer is a seed sample gathered from pages that render server-side. It skews heavily toward Indian consulates and H-1B stamping, and toward 2011–2021. Two fabricated records produced during extraction were caught and deleted; others may remain.
Outcome labels are only trusted when the quoted text supports them. An independent audit found roughly a third of the extractor's "approved" labels sat on excerpts that were bare lists of interview questions with no stated result. Every record is now re-checked against its own excerpt, and any label the text does not evidence is downgraded to unclear. The original label is retained in the database as outcome_original so the rule can be audited or reversed.
Records quoting the same source text are collapsed, including cases where the extractor emitted one event twice with contradictory outcomes. Consulate names are normalised (India : Mumbai → Mumbai), and university names are stripped from excerpts.
Community trackers claiming thousands of records serve only the first ~50 rows over plain HTTP — their pagination is client-side. This build contains what was actually readable, not what was advertised.