USCIS Policy Manual API: 456 Chapters of Immigration Guidance, Queryable by Volume

Short answer: the USCIS Policy Manual is the agency's own binding instruction set for the officers who adjudicate immigration benefits, and it is 12 volumes containing 456 chapters. It replaced the Adjudicator's Field Manual, and unlike 8 C.F.R. it changes by policy update rather than by notice-and-comment rulemaking. That is why it moves faster than the regulation underneath it, and why an immigration case-management product that only watches 8 C.F.R. is watching the slow layer. This post covers the structure, the unusual citation form, what the Policy Manual does and does not carry legal weight for, and how to query it with corpusType=AGENCY_GUIDANCE and source=uscis_policy_manual.

TL;DR

• The addressable unit is a chapter, not a numbered section. A cite looks like "USCIS Policy Manual, Vol. 12, Pt. D, Ch. 3", so there is no single number to key a record on the way there is for 8 U.S.C. 1255.
• 456 chapters across 12 volumes, all reachable behind one search endpoint with corpusType=AGENCY_GUIDANCE and source=uscis_policy_manual. Refresh is weekly.
• It is binding on adjudicators, not on courts. USCIS wrote it for its own officers. It states agency policy, and it superseded the Adjudicator's Field Manual.
• It changes by policy update, not by rulemaking. A Policy Manual update can shift how a benefit is adjudicated without a single word of 8 C.F.R. changing, which is exactly why monitoring it matters.
• The three-layer stack is INA at 8 U.S.C., then 8 C.F.R., then the Policy Manual. Each layer answers a different question, and the double citation form of the INA trips most parsers.
• This corpus is not BIA decisions, not immigration case law, and not Federal Register immigration rules. Those rules live in a separate FEDERAL_REGISTER corpus of 202,526 sections.

This post belongs to our federal agency guidance series.

For related coverage, see Federal Agency Guidance API: 34 Sub-Regulatory Sources in One Endpoint, MPEP and TMEP API: Patent and Trademark Examining Procedure You Can Query for the same manual-shaped problem on the IP side, and DOJ Justice Manual API: 1,548 Sections of Federal Prosecution Policy.

Disclosure: we build Vaquill AI. The counts here are our own, measured against the live index on 2026-08-30. The free GET /us/statutes/coverage endpoint returns the same numbers so you can check them without a key.

What the Policy Manual is

USCIS adjudicates benefits. Adjustment of status, naturalization, employment authorization, waivers, humanitarian relief. Officers making those decisions need something more operational than a regulation, and the Policy Manual is that thing.

It is written by USCIS for USCIS. It states agency policy on how a benefit is adjudicated, what evidence satisfies a requirement, and how discretion gets exercised.

For years the equivalent document was the Adjudicator's Field Manual, a legacy compilation that the Policy Manual has since superseded. If your internal knowledge base still cites the AFM, those cites point at a rescinded document.

Be precise about the legal weight. The Policy Manual is binding on USCIS officers and not binding on a court or on the Board of Immigration Appeals. It is agency policy, not a regulation and not a statute.

That distinction cuts both ways in practice. You cannot win an appeal by citing the Policy Manual alone. You also cannot predict what the service center will do without it.

The structure, and why it is integration-relevant

The manual is organized as volumes, then parts, then chapters. Twelve volumes, each covering one benefit family or cross-cutting topic.

Underneath those twelve volumes sit 456 chapters, and the chapter is the unit that carries text.

The citation form breaks ordinary parsers

Here is the shape of a real cite:

USCIS Policy Manual, Vol. 12, Pt. D, Ch. 3

Compare that to the two layers underneath it:

The Policy Manual has no single section number. The address is a triple, and the middle element is a letter. A regex written for \d+\s+C\.F\.R\.\s+\d+ finds nothing, and a database schema with one section_number column has nowhere to put a part letter.

There is a second trap, and it belongs to immigration specifically. The INA has two citation forms for the same provision. Practitioners cite INA 245; the codified form is 8 U.S.C. 1255. INA 212(a) is 8 U.S.C. 1182(a). INA 101(a)(15) is 8 U.S.C. 1101(a)(15).

Your extractor has to handle both, because attorneys write the INA form and the corpus is keyed on the codified form. If you normalize only one direction, half your citations dead-end, which is the general citation-parsing problem in miniature.

Look at the two inputs on the left. 8 C.F.R. changes through Federal Register rulemaking, which means a proposed rule, a comment period, and a final rule. The Policy Manual changes through a policy update, which USCIS issues on its own.

One of those paths takes months and is loudly telegraphed. The other can land on a Tuesday. If you only watch the regulation, you find out about the policy shift from a client.

Corpus mechanics

The Policy Manual is one of 34 named sources inside a federal agency guidance corpus of 21,906 sections. Two filters isolate it, and the corpus types reference lists the tokens.

Base URL is https://api.vaquill.ai/api/v1. Auth is a bearer token. Search costs 4 credits per call.

curl -X POST https://api.vaquill.ai/api/v1/us/statutes/search \
  -H "Authorization: Bearer vq_key_..." \
  -H "Content-Type: application/json" \
  -d '{
    "query": "continuous residence requirement naturalization absences",
    "corpusType": "AGENCY_GUIDANCE",
    "source": "uscis_policy_manual",
    "matchType": "any",
    "limit": 10,
    "excerptChars": 600
  }'

Every hit carries an actId, the citation, an excerpt, and the official uscis.gov URL for the agency's own copy of that chapter. Take the actId from the response. Ids are stable but not guessable, so a hand-built one usually 404s, which the section identifier reference explains.

Then fetch the chapter:

GET /us/statutes/section/{actId}
GET /us/statutes/section/{actId}/body

Use /body when you are feeding a model and want the text without the metadata wrapper. Use the plain form when you are rendering a citation card that needs the source URL.

Scoping a query to a volume

Be straight about this, because it is the first thing an integrator asks. There is no volume= parameter. The search filter set covers query, corpusType, state, code, titleNumber, chapter, part, source, agency, documentType, matchType, yearFrom, yearTo, publishedFrom, publishedTo, fields, limit, offset, and excerptChars.

So volume scoping is done two ways, and you will probably use both. The browse hierarchy recipe covers the same shape elsewhere in the corpus.

First, put the volume in the query text. Semantic plus keyword hybrid search means the volume label in a chapter's own heading is retrievable.

{
  "query": "Volume 8 Admissibility health related grounds communicable disease",
  "corpusType": "AGENCY_GUIDANCE",
  "source": "uscis_policy_manual",
  "matchType": "any",
  "limit": 20
}

Second, filter the returned citation. Every result carries its citation string, so a client-side filter on the volume token gives you a hard boundary that the query text alone does not.

hits = [h for h in results if "Vol. 8" in h["citation"]]

That is less elegant than a server-side filter and worth saying plainly. It is also stable, because the citation string is the agency's own address for the chapter.

A third pattern for narrow work: run the sweep once, keep the actId values for the volume you care about, and re-fetch by id afterwards. For a corpus of 456 chapters, mirroring the whole volume index locally is cheap and removes the scoping problem entirely.

Paging helps here. Results are cut from one ranking, so pages do not repeat or drop items between calls, and a later page costs the same as the first. limit maxes at 50 and offset at 70.

Worked example: an RFE that cites the manual

A request for evidence lands citing the Policy Manual. Here is the sequence that turns it into a checkable answer.

Step 1: resolve the chapter. Search with source=uscis_policy_manual using the language from the RFE itself, not a paraphrase. The agency's phrasing is usually lifted from the chapter, so it retrieves well.

Step 2: pull the chapter body. Fetch /section/{actId}/body, and keep the source URL from the metadata response so the attorney reviewing the output can open the agency's own page.

Step 3: reach the regulation. The chapter implements something in 8 C.F.R., so query the CFR corpus scoped to Title 8.

{
  "query": "adjustment of status eligibility to receive an immigrant visa",
  "corpusType": "CFR",
  "titleNumber": "8",
  "limit": 5
}

The CFR corpus is 219,114 sections and refreshes daily, the fastest cadence in the federal set.

Step 4: reach the statute. Query the USC corpus scoped to Title 8, translating the INA cite to its codified form first.

{
  "query": "adjustment of status of nonimmigrant to that of person admitted for permanent residence",
  "corpusType": "USC",
  "titleNumber": "8",
  "limit": 5
}

The USC corpus is the 2024 edition, current through 2025-01-06, at 60,170 sections, and it is checked weekly for a new edition. That currency date is published rather than implied, which is what lets you tell a supervising attorney exactly what edition an answer came from.

Step 5: hand the model all three, with their source URLs. Manual chapter, regulation, statute. That is a response a reviewer can verify in three clicks instead of a summary they have to trust.

Monitoring the layer that moves

Weekly refresh means changes get captured on a weekly cadence and become change events you can subscribe to.

Call GET /boards, which costs nothing, and every watchable source comes back stamped with its cadence, its lastRetrievedAt, and its retrievalStatus. That third field matters more than it sounds: it separates a source actually pulled from one merely scheduled.

From there you can create a watch, list the changes it captured, and pull a per-section diff to see what text moved. For an immigration practice the high-value watches are narrow rather than broad, since watching all 456 chapters produces noise.

What this corpus is not

The honest boundary list, because building on a wrong assumption here is expensive.

• Not BIA decisions. Board of Immigration Appeals precedent decisions are adjudicative case law, and they are outside this corpus.
• Not immigration case law generally. Circuit court immigration opinions are not here.
• Not Federal Register immigration rules. Proposed and final DHS rules live in a separate corpus, corpusType=FEDERAL_REGISTER, covering 202,526 sections from 1994 to the present with a weekly refresh. Query that one for rulemaking, this one for policy.
• Not the Foreign Affairs Manual. Consular processing guidance belongs to the State Department, which is a different agency with a different manual.
• Not case status or receipt data. Nothing about an individual filing is in a corpus of published guidance.

If your product needs BIA precedent as a first-class dependency, this is not the piece that provides it. If your product needs to know what an officer is instructed to do before issuing an RFE, it is.

FAQ

Is the USCIS Policy Manual legally binding?

It binds USCIS officers, who are required to follow it when adjudicating benefits. It does not bind a court or the Board of Immigration Appeals, because it is agency policy rather than a statute or a regulation. In practice it is the most reliable predictor of what a service center will do with a filing.

Did the Policy Manual replace the Adjudicator's Field Manual?

Yes. The Policy Manual superseded the Adjudicator's Field Manual as USCIS's operating guidance. Internal knowledge bases that still cite AFM sections are pointing at a rescinded document, and those cites should be remapped.

How do you cite a section of the USCIS Policy Manual?

By volume, part, and chapter, in that order: "USCIS Policy Manual, Vol. 12, Pt. D, Ch. 3". The addressable unit is a chapter, so there is no single section number the way there is for 8 U.S.C. 1255. The part element is a letter, which is what breaks schemas built for numeric section keys.

Why does the Policy Manual change faster than 8 C.F.R.?

Because it changes by policy update rather than by rulemaking. Amending 8 C.F.R. requires the notice-and-comment process, which runs through a proposed rule, a comment period, and a final rule. USCIS can issue a Policy Manual update on its own, so guidance can shift while the regulation stays fixed.

How many volumes and chapters are there?

Twelve volumes and 456 chapters. Volumes cover benefit families and cross-cutting topics, parts subdivide a volume, and chapters carry the text.

Can I filter a search to one volume?

Not with a dedicated parameter, because there is no volume= filter. Scope by putting the volume label into the query text, then filter the returned citation string client-side on the volume token. For narrow work, mirror the chapter index locally once and re-fetch by actId.

Are BIA decisions included?

No. The Board of Immigration Appeals issues adjudicative decisions, which are a different kind of authority from published agency guidance and are not in this corpus. The same is true of circuit court immigration opinions.

How quickly does a new policy update reach the API?

A Policy Manual update is picked up on the weekly refresh of the federal agency guidance corpus, which is the cadence for every source in it. The 8 C.F.R. layer underneath it refreshes daily. Federal Register rules refresh weekly in their own separate corpus.

Can I query the Policy Manual as it read on a past date?

Versioning here runs off the amendment record rather than a date parameter. A chapter is stored at its live text and nothing earlier, and no point-in-time parameter is offered. You still get amendment history, a currency filter, and diffs on watched sources, but a prior version has to come from the agency's own archive of policy updates.

What do I do about INA versus 8 U.S.C. citation forms?

Normalize in both directions before you query. Attorneys write INA 245 and INA 212(a); the corpus is keyed on the codified forms, 8 U.S.C. 1255 and 8 U.S.C. 1182(a). An extractor that handles only one form silently loses every citation written in the other.

Where to start

Call the free coverage endpoint, then run one scoped search against source=uscis_policy_manual using text lifted verbatim from an RFE you already have on file. The agency's own phrasing retrieves well, so you will know quickly whether the results are good enough to build on.

Priyansh Khodiyar

Co-Founder & CTO

Priyansh leads engineering and AI at Vaquill, from the matter workbench to drafting, document comparison, document matrix, and citation-verified research.