Does AI referral traffic convert better than Google search for a law firm?

The numbers being quoted are not invented. They are just not usable for deciding anything about your firm. That does not make AI search unimportant, and there is something real underneath the claim. It is further down this page than the headline number suggests.

What the studies actually say

Five separate figures get quoted, usually without saying which one is being used.

A range that runs from 23x down to no measurable difference is not a finding. It is a menu. Anyone quoting a single number from that spread has chosen the one that best supports the pitch.

Notice which row is the odd one out. Amsive is the only entry that ran a controlled comparison: six months of GA4 data across 54 sites with validated conversions, organic at 4.6% against LLM referrals at 4.87%, tested with a paired t-test. The p-value came back at 0.794, meaning the gap is what you would expect from chance. Every other row is a vendor or platform reporting on its own traffic without a control.

One nuance worth keeping, because it cuts toward law firms rather than away: Amsive’s B2C sites came out level, but their B2B sites did show a real gap, 2.17% against 1.16%. A firm taking case enquiries looks more like the second group than the first. That is a reason to keep measuring, not a reason to accept 5x.

Then there is the retail number, which has a history. Adobe reported AI traffic converting 38% worse than non-AI traffic in March 2025, and 42% better in March 2026. Same publisher, same method, opposite conclusion twelve months apart. Whatever that measures, it is not a stable property of the channel, and a figure that can swing eighty points in a year should not be the basis of anyone’s budget.

Nobody measured a law firm

Every study above measured software signups or retail checkouts. Those are self-serve conversions that happen in one session, on the same device, with the whole path visible in analytics.

A personal injury case is none of those things. The conversion is a phone call. It is followed by an intake screen, a sign-up decision, and often a wait of days. The event that matters, a signed case, happens inside a case management system, not in a browser. Applying a SaaS signup ratio to that is not a small stretch. It is a different measurement problem entirely.

The math flatters itself

Here is the part that gets skipped. In the Ahrefs figure, AI referrals were 0.5% of all sessions. The comparison is a very small, heavily pre-filtered slice against all organic traffic.

Think about what has already happened to somebody arriving from an AI answer. They asked a question. An engine gave them an answer. That answer named a firm, in a context that already matched their situation, and they chose to click through anyway. Several rounds of qualification happened before they ever reached the site.

Now compare that to organic search, which includes every casual searcher, every person comparing four firms, every student, every competitor, and every accidental visit. Of course the tiny filtered segment converts better. That gap is mostly telling you the traffic was pre-qualified, not that the channel is five times more valuable.

This is a selection effect, and it is the same reason “people who walk into your office convert better than people who visit your website” is true and useless.

What is worth taking from it

Strip out the multiple and something real remains. Traffic arriving from an AI answer tends to be further along, because the answer did the early filtering. That is directionally consistent across every study, in every sector, which is more convincing than any individual number in the table.

The useful conclusion is not “chase AI traffic because it converts 5x.” It is that AI answers are now doing qualification work that used to happen on your website, and a firm that is absent from those answers is absent before the comparison starts.

What a firm should measure instead

Three things, in this order.

1. Signed cases by source. Not sessions, not conversion rate, not form fills. If your intake and call tracking cannot attribute a signed case back to where it came from, no channel comparison you run is real, and this is the first thing to fix.
2. Whether you appear at all, per engine. Citation pools barely overlap between engines. Roughly 91% of citations appear in only one engine, so a single blended “AI visibility” score hides more than it reveals. Presence is a yes or no question per engine before it is a rate question.
3. Cost per signed case, by channel, over a period long enough to matter. This is the number that decides budget. It is also the one that survives whatever the engines do next, because it is measured in your business rather than in someone else’s dashboard.

If a vendor leads with a conversion multiple and cannot produce the first item on that list for your firm, the multiple is decoration.