Perplexity and Claude for Immigration Legal Research

Immigration legal research has a distinct source hierarchy that general AI tools were not built for. At the top: the Immigration and Nationality Act (INA). Below that: 8 CFR, which implements the INA. Below that: BIA precedent decisions, which bind USCIS officers and immigration judges. Then AAO decisions. Then federal court interpretations. The USCIS Policy Manual sits below the CFR: it interprets regulations but cannot contradict them.

A Westlaw query surfaces sources from across this hierarchy with citator verification. Perplexity + Claude replicates about 70% of that capability for free. Understanding what the 30% gap is, and when it matters, determines whether this stack is appropriate for your research task.

The Source Hierarchy That Matters

Before using any AI tool for immigration research, the source hierarchy needs to be clear because AI tools are not equally good at every level.

INA sections and CFR text are crawled, indexed, and easily retrievable. Both Perplexity and Claude handle them well. Perplexity can find the specific eCFR URL; Claude can read CFR sections you paste and explain how they interact.

BIA precedent decisions present a more complex picture. The major published BIA decisions from the last 25 years are freely available on EOIR (justice.gov/eoir) and indexed by Google. They are in Perplexity's index. They are in Claude's training data through August 2025. Anything issued September 2025 onward is not in Claude's knowledge.

AAO non-precedent decisions (the most directly relevant authority for EB-1A and O-1 cases) are individually published PDFs on USCIS.gov. Perplexity's index of specific AAO decisions is inconsistent. The USCIS AAO non-precedent decision search tool (free, no login) is more reliable for finding a specific decision than any AI tool.

Federal court cases: Google Scholar covers them in full and for free. Claude knows major circuit decisions through August 2025.

The Northwestern Law immigration clinic source guide is the most useful single reference for understanding this hierarchy.

Setting Up Perplexity for Immigration Research

The standard Perplexity web search returns results from across the internet. For legal research that matters, you want something more targeted.

The site: operator

Any Perplexity query supports site-specific restriction. The format:

site:uscis.gov [your query]
site:ecfr.gov [your query]
site:justice.gov [your query]

Example queries that work well:

• site:uscis.gov EB-1A original contribution major significance 2025 (pulls USCIS Policy Manual sections and any policy alert PDFs USCIS has published)
• site:ecfr.gov 8 CFR 204.5 extraordinary ability requirements (pulls the current regulatory text)
• site:justice.gov BIA precedent decision extraordinary ability (pulls BIA precedent decisions published by EOIR)

This does not guarantee you'll find everything; it depends on Perplexity's crawl depth of each domain. For AAO decisions specifically, use the USCIS search tool directly rather than relying on Perplexity's indexing.

Perplexity Spaces with domain restrictions

For attorneys who run similar research repeatedly (tracking EB-1A policy updates, monitoring BIA decisions on specific issues), Spaces with custom domains is the most useful Perplexity feature. Available on Pro.

Set up a Space called "USCIS Research" and restrict it to:

• uscis.gov
• ecfr.gov
• justice.gov
• federalregister.gov

Every search inside that Space only returns results from those four domains. This removes the noise of law firm blogs, immigration service site marketing, and forum discussions from research sessions. The Perplexity domain filtering documentation shows how to configure this in the API; the Space setup in the UI follows the same domain restriction concept.

Deep Research for initial orientation

When you encounter an unfamiliar topic (a circuit split you haven't tracked, a new policy memo, an issue type outside your usual practice area), Deep Research mode is the fastest way to get a structured map.

Deep Research runs autonomously for 2-4 minutes, queries across many sources, and produces a report with inline citations. For a research starting point, it is significantly better than asking Claude from memory. The citations need to be verified; the structure is usually sound.

Practical use case: a client is from a country in Central Asia with complex treaty obligations affecting their visa category. Deep Research on site:uscis.gov country-specific treaty visa obligations [specific country] produces a structured report that would take 45 minutes to assemble manually.

Computer for Counsel (@midpage)

On June 24, 2026, Perplexity launched Computer for Counsel for Enterprise and Max subscribers. The key addition is a Midpage connector; invoke it with @midpage in a query. Midpage provides access to U.S. case law, statutes, and regulations in a purpose-built legal research layer.

For immigration research, this matters primarily for federal circuit court opinions, which Midpage indexes in full. For USCIS-specific research (Policy Manual, AAO decisions, eCFR), the site: operator approach remains more reliable because Midpage's coverage of agency-specific documents is less predictable than its federal court coverage.

Computer for Counsel is not a Westlaw replacement. It has no KeyCite equivalent. It cannot verify that a case is still good law. For that, you need Westlaw, Lexis, or AILALink (AILA member benefit, which includes citator coverage for BIA and AAO precedent decisions).

Setting Up Claude for Immigration Document Analysis

Claude's strength is the opposite of Perplexity's. It cannot retrieve new information, but it can process very large documents with analytical precision that general web search cannot replicate.

What to put in a Claude conversation for research:

1. The full regulatory text from eCFR (paste it, or use the URL if using Claude with web access enabled)
2. The relevant Policy Manual sections
3. Any AAO decisions you found through Perplexity or the USCIS search tool
4. Your client's specific facts

Then ask analytical questions: How does the definition in the Policy Manual interact with the CFR text? Does this AAO decision appear to apply the October 2024 USCIS guidance? Where does my client's citation count place them relative to the "top few percent" standard described in this AAO decision?

Claude Projects for persistent research context

Claude Projects (available on Pro and Team plans) maintains persistent context across conversations within a project. For immigration attorneys, the practical setup:

Create a project called "EB-1A Research" and upload to it:

• The relevant eCFR part (8 CFR 204.5 text)
• USCIS Policy Manual Vol. 6, Part F, Ch. 2 (the EB-1A chapter)
• Kazarian v. USCIS (the 9th Circuit opinion PDF)
• Any key AAO decisions you regularly reference
• Your own internal RFE response frameworks

Once uploaded (~200K token limit per project), every new conversation in that project starts with Claude already having read those documents. You ask questions without re-pasting the source material every time. The Claude Projects setup guide covers this in detail for EB-1A specific use.

The knowledge cutoff problem

Claude Sonnet 4.6 and Claude Opus 4.6 both have an August 2025 training data cutoff. This has concrete implications for immigration research in 2026:

• PM-602-0199 (May 21, 2026): the USCIS adjustment of status discretionary denial memo is after Claude's cutoff. It does not know this policy exists.
• The H-1B $100,000 fee under H.R. 1 and its subsequent First Circuit stay: after cutoff.
• Any AAO decisions from September 2025 onward: after cutoff.
• The July 2026 Visa Bulletin: after cutoff.

When Claude describes current policy, it is describing the policy as of August 2025 at the latest. For anything that changed after that date, you must retrieve the current version yourself (via Perplexity or direct USCIS site search) and paste it into the conversation.

The Hallucination Problem in Immigration Research

A Stanford study published in the Journal of Empirical Legal Studies (April 2025) tested 200+ legal queries across general-purpose AI tools and purpose-built legal research platforms. General-purpose models (which include both ChatGPT and Claude) achieved 40-60% accuracy on legal queries and hallucinated at a rate of 43-80%. Purpose-built tools like Lexis+ AI did better but still hallucinated 17% of the time.

For immigration attorneys, the documented risk is in citation fabrication. Dec v. Mullin, No. 25-2417 (7th Cir. Mar. 30, 2026), involved an immigration appeal brief that included two non-existent cases and a false quotation. Counsel initially denied using AI; later admitted copying from a brief they could not locate. The Seventh Circuit admonished the attorney but declined sanctions because the errors were unintentional and counsel was contrite. That result reflects judicial leniency on a first instance. It will not hold as a pattern.

The same risk applies in immigration proceedings before USCIS and EOIR. Submitting a non-existent AAO decision as evidence of administrative practice is worse than submitting no citation at all.

Practical citation verification workflow:

For CFR citations: paste the exact section text from eCFR into your document. Do not cite from AI-generated summaries of what the CFR says.

For Policy Manual: use the keyword search at uscis.gov/policy-manual to find the exact passage, confirm it says what you intend to cite, and link to the chapter URL.

For AAO decisions: search the AAO non-precedent decisions tool by the decision ID number. Confirm the case exists and the holding matches what you are citing.

For federal cases: search Google Scholar by the exact case name. Confirm the case number, circuit, and year. Read the relevant passage directly.

No AI tool substitutes for this verification. Perplexity provides citations with URLs: verify those URLs load the actual document. Claude generates citations without URLs: verify every one against a primary source.

What Each Tool Is Actually For

Use Perplexity to:

• Find which chapter of the Policy Manual covers your issue (initial orientation)
• Locate a specific USCIS policy alert or memo by date or topic
• Survey recent circuit court holdings on a specific immigration question
• Track pending litigation or legislative changes affecting a practice area
• Find AAO decision IDs to then verify on the USCIS search tool

Use Claude to:

• Read and synthesize documents you have already retrieved and verified
• Compare two Policy Manual sections or AAO decisions for consistency
• Extract structured information from dense regulatory text
• Draft a research memo from source material you supply
• Identify inconsistencies between a CFR section and the Policy Manual's interpretation
• Explain how a recent AAO decision you paste applies to your client's specific facts

Use neither for:

• Verifying a case is still good law (requires KeyCite or Shepard's)
• Full litigation analytics
• Citator-level checking of whether a BIA decision has been overruled
• Any policy development after August 2025 from Claude's memory

What the free stack covers:

For EB-1A and O-1 research specifically, the combination of Perplexity (free or $20/month Pro), Claude (free or $20/month Pro), the USCIS AAO search tool (free), eCFR (free), and Google Scholar (free) handles the majority of research tasks. The gap versus Westlaw is primarily: citator coverage (knowing whether your authority is still good law) and curated treatise commentary (INA commentary, practitioner guides). For well-settled areas of law, the gap is small. For cutting-edge issues with active circuit splits, use Westlaw.

ABA Ethics Requirements for AI Research Tools

ABA Formal Opinion 512 (July 29, 2024) is the first formal ABA ethics guidance on generative AI in legal practice. Five requirements from it apply directly to using Perplexity and Claude for legal research:

Competence (Rule 1.1): You must understand what these tools can and cannot do, including the hallucination risk and knowledge cutoff issues described above. This understanding must be periodically updated as the tools evolve.

Confidentiality (Rule 1.6): Consumer plans (Claude Pro, Perplexity Pro) may use your prompts to train their models unless you disable this in privacy settings. For client-specific research involving identifying information, use enterprise accounts with data processing agreements, strip identifying information from prompts, or consult your bar's specific guidance on AI tool use.

Candor (Rules 3.1 and 3.3): Every AI-generated legal citation must be independently verified before it goes into any filing. The verification workflow above is not optional.

Supervision (Rules 5.1 and 5.3): You are responsible for how your staff uses these tools. A paralegal who submits an AI-drafted brief with unverified citations is your responsibility.

Fees (Rule 1.5): Using AI to produce work faster than manual methods has billing implications your bar may have guidance on.

The full text of Opinion 512 is behind the ABA paywall. Law school library summaries of it (try UNC Law or RIPS Law Librarian blog) are publicly accessible and accurate. Read them before deploying AI tools in your practice.

A Practical Research Session

Here is what a complete session looks like for an attorney researching original contributions (C5) for a client in computational biology:

Step 1 (Perplexity, 5 minutes): Query site:uscis.gov EB-1A original contributions major significance computational biology 2025. Find the relevant Policy Manual chapter URL and any recent policy alerts. Note the AAO decision IDs that appear.

Step 2 (USCIS AAO search tool, 10 minutes): Search for the noted AAO decision IDs. Download the 2-3 most recent and relevant C5 decisions. Read them.

Step 3 (Perplexity Deep Research, 4 minutes): Query EB-1A original contributions denial patterns 2024 2025 site:uscis.gov OR site:aao.uscis.gov. Collect the structured report.

Step 4 (Claude, 20 minutes): Paste into a conversation: (a) the relevant Policy Manual section you found in Step 1, (b) the AAO decisions from Step 2, (c) your client's specific evidence list. Ask: "Based on these Policy Manual standards and these AAO decisions, which aspects of my client's record are strong, which are weak, and what additional evidence would most directly address the current AAO standard for 'widely implemented throughout the field'?"

Step 5 (Verification, 10 minutes): Verify every CFR citation Claude used against eCFR. Verify every AAO citation against the USCIS search tool. Replace any citation you cannot verify.

Total: under an hour for a research foundation that would take 2-3 hours of traditional Westlaw research for an attorney not already expert in this criterion.

For deeper guidance on how Claude handles specific immigration drafting tasks, the prompts for immigration attorneys article covers tested prompt patterns for research memos, criterion analysis, and RFE response arguments. The AI tools overview for attorneys covers the full set of tools available beyond these two.

Immigration Copilot applies this research-to-draft workflow in the context of complete EB-1A petition packages: evidence classification, criterion mapping, draft generation, and fact verification in a structured pipeline. See how it works.