Ask an engine which legal AI assistant is safe for a law firm to use with client data, and it does not start with a product. When I ran that question in November, the answer opened with a warning about consumer tools, then cited ABA Formal Opinion 512, then sorted the vendors by how they handle data. Zero data retention. SOC 2 Type II. ISO 42001. A FedRAMP certification for one product. Features came later.
That order tells you who is really asking. A lawyer choosing software carries a professional duty that follows them personally. The question arrives with that duty attached, and the engine answers it that way.
We have written about how security answers hesitate because a bad recommendation is embarrassing, and about how fintech answers lead with who pays when something goes wrong. Legal looks like a close relative of both. It behaves differently in several ways, and each one changes what you should do.
Your facts have to map to the duty
A security certification on a trust page is a fact about your company. In legal, the same fact is only useful once it answers the question a lawyer is obliged to ask: can I put client information in this, and can I show I checked?
So state it in those terms. Whether customer data trains any model, written as a plain sentence and not implied by a badge. Your retention policy, with the number. Which certifications you hold and when they were last renewed. Whether access can be restricted by matter, since the answer I read listed audit logs and permission controls as a pillar of its checklist. The same holds outside legal: in the security review questions buyers now put to AI, the vendors with one plain public page per question were the ones the engine quoted.
Then state your stack, because the follow-up comes fast. Three of the four answers I ran ended by asking what the buyer already uses. iManage and NetDocuments came up by name in the law firm answer, and Salesforce and SAP in the corporate ones. An integration the engine cannot find is one the buyer will assume you do not have.
“Legal” is two markets sharing one word
This is the mechanism most vendors miss, and the search results show why. Look up AI search advice for legal and almost everything on the page is written for law firms trying to win clients. I could not find a guide written for the companies selling software to lawyers.
The engines keep the distinction much better than the marketing does. Asked about legal spend for corporate departments, the answer opened by saying the category is distinct from standard law firm billing tools. Asked about eDiscovery for mid-size firms, it described the leading enterprise platform as too heavy for them and built its shortlist around that.
A law firm partner and a general counsel have different budgets, different workflows, and different words for the same work. If your site says “legal teams” throughout, the engine has to guess which you serve, and it may file you with the wrong buyers. It is the same problem as how the engine decides your product category, with an extra trap, since both sides would describe themselves as legal.
Say which one you sell to, on the page and in the sources. If you sell to both, give each its own page in its own vocabulary: matters and billable hours for firms, intake and outside counsel for departments. Then test each buyer’s questions separately, because a strong result on one tells you nothing about the other.
The incumbents are publishers, so go around them
Most software categories do not have this problem. In legal, the largest vendors have spent decades publishing the material lawyers rely on. In other industries the opposite happens: the trade press barely appears, and other software companies’ lists write the answer. On the safety question, the engine called CoCounsel from Thomson Reuters the gold standard and leaned on its link to Westlaw and Practical Law. On legal spend, the opening of the answer drew on a LexisNexis page, Thomson Reuters came first in the table, and Wolters Kluwer was named for the size of its benchmarking data.
You will not out-publish those companies on their own ground. Look instead at where smaller vendors did appear. The smaller names in these answers each had a row with a specific buyer in it. Juro for startups and mid-market. Xakia for small or newly formed in-house teams. Everlaw for mid-size litigation. LinkSquares for lean teams with a large library of old contracts.
None of those companies had to beat an incumbent overall. Each is the plain answer to a narrower question. The engine has room for them because it builds these answers by buyer type, and the shortlist changes when the question names a segment. Pick the segment you can honestly claim, then make sure the sources say it in those words.
Whoever defines the decision writes the opening line
The most interesting result came from the contract management question. The answer did not open with a ranking. It opened by telling the buyer to find their bottleneck, then split the market into workflow systems and drafting tools. That framing came from a guide published by Spellbook, a vendor in the category, and it shaped the rest of the answer, including when to consider Spellbook.
LawVu did something similar. Its guide on choosing contract software for in-house teams was cited, and LawVu appeared in the table described the way that guide positions it: contracts, intake, matters, and spend in one place.
A clear way to choose suits a buyer trained to want the test before the answer, and the engine is happy to repeat one. If you sell in legal, publish the decision framework for your category: what to work out first, what kind of tool fits which problem, and when your kind is the wrong one.
There is a limit. On the eDiscovery question, a newer vendor’s review page was cited twice and helped set the terms of the answer, including the point about enterprise platforms being too heavy. That vendor did not appear in the recommendations. A framework gets you quoted. Getting picked takes other people naming you, which is the difference between being cited and being recommended.
Small communities have long memories
Legal technology has a small number of places where practitioners compare notes, and the engine reads them. The contract answer drew on The L Suite, a community for in-house lawyers, and on the Association of Corporate Counsel’s job board site. The spend answer used Legaltech Hub. The eDiscovery answer built its most practical advice, using one lightweight system for everyday matters and a heavier one for huge cases, from a thread on r/ediscovery.
That thread was from 2023. The L Suite list that fed the contract answer had 2024 in its title. Old sources keep working here because so few people write about these products in public.
That cuts both ways. A single good discussion can keep paying you for years, and a single complaint can follow you just as long. In a large category, one thread gets diluted. Here it may be one of only a handful the engine can find. Know what those threads say about you, and if something is out of date, correct it at the source. Do not try to manufacture your way into them. Lawyers notice, and the reason the engine trusts these threads is that nobody in them is selling.
What I got wrong
I started this piece with a clean theory: legal buyers are risk-averse professionals, so the engine trusts practitioner publications and ignores review sites and vendor pages. I had even written that down as the argument before running a single question.
The answers did not support it. Practitioner sources were there and they mattered, but so were TrustRadius, Gartner, a software directory, and a lot of vendors’ own pages. Vendor pages were among the most influential sources in these answers, including the page that wrote the opening line of the contract answer. Had I gone with my theory, I would have told you to ignore your own site, and that would have been poor advice.
These were also single runs of four questions on one engine, so treat them as a read of the pattern and not a measurement. Run your own questions before you act on mine.
Where this leaves you
Write your security and data facts as answers to a lawyer’s obligation. Decide whether you sell to firms or departments, and keep them apart on the page and in your testing. Do not fight the publishers on their own ground; claim a segment they are too broad to own. Publish the way to choose in your category, and accept that this earns you quotes before it earns you recommendations. Then read the small number of communities where your buyers talk, because the engine is reading them too, including the old threads.
If you want your legal product checked against the questions your buyers actually ask, split by who is asking, that is the engagement, and it runs on a documented method.
1 response