LegalZoom's new assistant answers from the customer's own record. Read the launch line by line for what to copy

Why read a press release this closely

Most launch announcements for AI assistants say the same things: personalized, instant, always on. LegalZoom's, published October 29, says some of that too. But tucked inside it is a clear description of how a customer-facing assistant should work, and it's worth reading line by line if you run any business where customers have accounts: a membership, a subscription, a portal, a service with paperwork.

This isn't a review of LegalZoom, and we're not judging its legal products. We're reading the design. Here are six lines, and what each suggests for anyone building something similar.

"Existing customers get answers grounded in their own business."

This is the whole idea in one sentence. Most help assistants start from the help center: a pile of articles, searched for the closest match, summarized. That works for "how do I change my address?" It fails for "is my address up to date?", because the help center doesn't know.

LegalZoom's assistant starts from the account. An assistant that starts from the account can answer the question the customer actually asked, which is almost always about them. Not "how do annual reports work," but "is mine due?"

For your own business, the question is which record your customers' questions are really about. For a utility, it's the meter and the bill. For a property manager, it's the lease and the maintenance tickets. For a membership, it's the renewal date and what's included. If you can't name the record, you're building a help-center search with a friendlier face.

"Am I in good standing?"

LegalZoom's example is a business owner in Texas asking whether their company is in good standing. Because the assistant already knows the business type, formation state, registered agent and address, it pulls from the entity record, confirms the Texas LLC is active and flags a Public Information Report due May 15.

Notice what made that possible. Not a clever model, but four fields the company already held, kept accurate. The assistant's best answer is a deadline it can see because the data was there all along.

That's the unglamorous part of building one of these, and usually the longest. Before the assistant, there's an audit of the record: which fields exist, which are reliable, which are years out of date. We wrote about assistants giving stale answers in your assistant is still quoting last year's policy. An assistant grounded in a bad record gives confident wrong answers about the customer personally, which is worse.

"...connects them to a licensed attorney or specialist when judgment is required."

The handoff is named and specific: an attorney or a business specialist, depending on the question. That's different from "contact support," which is where most assistants send anything hard.

Two design choices hide in that line. First, someone decided in advance which questions need judgment. Second, there's a real person on the other side with the right qualification. Both have to be true for the handoff to mean anything.

Then there's the moment of handoff itself. A good one passes along what the assistant already knows, so the customer doesn't repeat themselves, and tells the customer plainly what happens next and roughly when. For your assistant, write the list of questions it must never answer on its own, and say who gets each one. The list will be short. Getting it right matters more than anything else in the build. We covered a related question, what an assistant should remember about a customer between conversations, in what should your assistant remember.

"...most often to check their business and account status, find the right product, and stay ahead of compliance and filing deadlines."

LegalZoom says customers have used the assistant "hundreds of thousands of times" since August, and lists the top three reasons. That's the company's own figure, and we'd treat it as such. The list is more interesting than the number.

Status, the right product, deadlines. None of those are deep questions. They're the questions people phone about because they can't find the answer in their own account. If you're deciding where an assistant would help your customers, start with your support inbox, not a brainstorm. The most common questions are probably status questions, and status is exactly what a record-based assistant answers well.

The next step, and where the assistant lives

"...linking straight to the required LegalZoom filing product." Here's the line to think hardest about. In the Texas example, the next step is a link to a LegalZoom product. That's reasonable: the report is due, and LegalZoom sells the filing. But it shows the pull every company building an assistant will feel. The assistant knows the customer, knows what's due, and sits next to a checkout.

Decide early how your assistant handles that. Does it mention that the customer can do this themselves, if they can? Does the next step depend on what's best for them, or on what you sell? Customers notice when every answer ends in a cart, and the trust a record-based assistant earns is easy to spend. A simple test: pick twenty real customer questions and write down where the ideal answer would send each one. If the honest answer is sometimes "nowhere, you're fine," the assistant should be allowed to say that.

"LegalZoom apps are also live today in ChatGPT, Claude, Copilot, Gemini, and Perplexity." The last line is about reach. Customers can get to the assistant inside their account or from general-purpose AI apps. That's convenient, and it raises a question every business doing this has to answer: when a customer asks about their own account from someone else's chat app, what data leaves your system, and on whose terms?

That's not a reason to stay out of them. Customers already use those apps, and meeting them there can be useful. But the answer should be written down: which fields an outside app can read, whether the customer has to sign in each time, and what the outside app keeps afterward. Start inside your own account pages, where you control the login and the record. Add outside chat apps once you've decided what they're allowed to see.

If you'd like help building an assistant that starts from your customers' records and knows when to hand off, that's the work our AI applications and agents team does. Talk to us.

Related work: Sutton Place.