As AI Legal Research Tools Race to Give Answers, This New Product Gives Only Cases — And That’s the Point

Align Bets Litigators Don't Want AI's Answer — Just the Cases

On August 3, 2026, litigation technology company Align released Align Research, a legal research product that returns only court opinions with key passages highlighted — no memo, no synthesized answer. Founder and CEO Sam Davidoff, a former Williams & Connolly litigation partner, told LawNext's Bob Ambrogi the design is deliberate: an agentic pipeline of models breaks down the research question, runs searches, reads and classifies the resulting cases, and highlights relevant passages, a process Davidoff said can take "a few hours" per query. The company's own framing, quoted in its press release, is that the product "retrieves rather than generates." That puts Align squarely against the current direction of the AI legal research market, where most vendors are racing toward single, synthesized answers.

Why build a research tool that refuses to answer?

Because a tool that only points to existing case law, rather than writing new text about it, cannot fabricate a case that doesn't exist.

Davidoff's logic is narrow but real: hallucination happens in the generation step, so removing generation removes that specific failure mode. He's explicit that this doesn't make the tool infallible — in his own words, it "could miss something, it could highlight something that you don't agree with, but it can't make something up." That's a meaningful distinction between two different error types, and buyers evaluating any AI legal tool should know which one they're being sold protection against.

Does removing the summary actually remove the risk?

No — it shifts risk from fabrication to omission, and omission is harder for a buyer to detect or audit.

Nothing in the LawNext account describes how Align validates that its search-and-classify stages surface the right universe of cases, or what happens when the relevance model gets a borderline call wrong. A missed controlling case is arguably a worse outcome for a litigator than a hallucinated one, because a fabricated citation gets caught on Shepardizing while a silent gap doesn't announce itself. Ask any vendor selling "retrieval, not generation" as a safety feature what their recall methodology is — not just their precision.

Is this actually a new category, or old-school research with an AI wrapper?

It's closer to the latter: automating the "get me every case that ruled this way" task lawyers already delegated to associates.

Davidoff says as much himself, calling it a return to "how we used to do it." The pitch to the market isn't a new capability so much as a philosophical stance on division of labor — keep the reading and the argument-building with the lawyer, hand the AI the volume work of scanning thousands of opinions. For firms already comfortable with retrieval-first, cite-linked AI tools in e-discovery, this will feel familiar rather than novel.

What should a buyer ask before adopting a retrieval-only tool?

Ask what the tool is guaranteeing, what "relevant" means to its models, and how gaps get surfaced rather than hidden.

Specifically: what governs the relevance and classification models' decisions, is there any audit trail showing what was searched and excluded, and what does the vendor do when a highlighted passage is later found unrepresentative of the holding. A platform that always links an answer back to a source exhibit or opinion is only as trustworthy as its search coverage — the citation is not itself proof of completeness.

Frequently asked questions

Is Align Research a full research platform or an add-on?

Based on the LawNext account, it's a standalone product launched to general availability by Align, positioned against synthesized-answer competitors rather than as a plug-in.

How long does a research job take?

Davidoff describes jobs running for "a few hours," reflecting a multi-stage agentic process rather than an instant chat response.

Source: LawNext, reported by Bob Ambrogi.