AI Mode Ignored 98.7% of What Google's Own Search Ranked
On September 1, an e-commerce tracking firm called Productrise published a study that got picked up quickly, and for the wrong reason. Over 23 days — August 9 to August 31 — it ran identical shopping queries through Google AI Mode and traditional Google search, same query, same day, same moment, across more than 100,000 result sets and two million product listings. When the exact same product appeared on both sides, the price shown in AI Mode was 21.6% higher on average. Futurism ran it on September 2 under the headline that AI Mode is "ripping shoppers off." Google's response was that both surfaces draw from the same Shopping Graph and that shoppers can click through to compare sellers.
The price gap is the story for retailers. It is not the story for a law firm. Legal services do not sit in a Shopping Graph, and nothing in that dataset says anything useful about how AI Mode handles a question like "best asylum attorney in Houston." Read the price number as retail news and leave it there.
Read the second number differently.
The Overlap Was 1.28%
Of the products that ranked in traditional Google search, only 1.28% also showed up in AI Mode for the same query on the same day. Traditional search returned an average of 27.8 products. AI Mode returned 3.9. Where the two did land on the same product, AI Mode named a different lead seller 49.6% of the time — nearly half.
Set the prices aside and look at what that describes. Two surfaces, one company, one underlying index, one query, one day. And they substantially disagree about which options are worth showing. AI Mode is not compressing the ranked list into a shorter ranked list. It is building a different list, from different criteria, and it is doing so in a format where the searcher sees three or four options instead of twenty-eight.
That mechanic is not retail-specific. That mechanic is how an answer engine works on any question, including yours.
Well Known Is Not the Same as Well Described
Here is what a short list rewards, and it is not what most practices assume.
A ranked list can afford to be generous. Twenty-eight slots means the eighth-best match still appears, and a searcher scanning the page supplies the judgment. A four-item answer cannot be generous. Something has to select, and a model selects on what it can parse, corroborate across independent sources, and state without hedging. Reputation that lives in rooms — the firm everyone at the immigration court knows, the practice other attorneys quietly refer their removal defense to, the office that has run half the county's naturalization clinics for a decade — is invisible to that process. It is real. It is not machine-readable.
So the practice that a local market would call the obvious answer can be precisely the one an answer engine skips. Not because a competitor outranked it. Because when the model went looking for something it could verify and say out loud, the better-known firm had a three-line "Immigration Law" page and the less-established one had a specific, sourced, corroborated description of exactly what it does.
Immigration compounds this. The searcher is often in a hurry and often frightened — a client with an RFE deadline in eleven days, a spouse whose priority date just became current, a family that got a notice to appear. They ask one question and they act on the answer. There is no leisurely comparison of twenty-eight options. There is the short list.
What Actually Gets a Firm Onto a Short List
The retail read on the study was that better product feed data — highlights, structured details, third-party reviews, press — is what makes a model confident enough to recommend a seller. Strip the vocabulary and the principle holds for a practice: a model recommends what it can describe precisely and confirm independently.
Precision means the site says which forms, which case postures, which consular posts, which languages the office actually works in — not "we handle all immigration matters." A model cannot distinguish a firm that does forty PERM filings a year from one that did two, if both pages say "employment-based immigration."
Confirmation means the claim appears somewhere the firm does not control. State bar directories, AILA chapter listings, legal aid partner pages, local news coverage of a clinic, a court's pro bono roster, a law school's externship page. One corroborated specific outweighs a page of adjectives.
Do This Week
- Run five real client questions in AI Mode, signed out, phrased the way a client would phrase them — not the way an attorney would. "How do I fight a denied asylum case in Atlanta," not "asylum appeal counsel." Record which firms are named. Three or four will be.
- Run the same five in traditional Google search and write down the top ten organic results. Compare the two lists. The gap between them is your local version of the 1.28% finding, and it tells you whether ranking is currently doing anything for you on those questions.
- Pick your three highest-value practice areas and rewrite those pages to be specific. Case types, forms, filing venues, consulates, languages, typical timelines you actually see. Replace every claim that could be made by any immigration firm in the country.
- Find three off-site places your specifics already appear — bar profile, AILA listing, an organization's referral page — and make sure they match the site. Mismatched or stale third-party entries actively cost you; a model that finds two conflicting descriptions tends to use neither.
- Log it and repeat in 30 days. One reading is an anecdote. Two readings a month apart is a trend, and this surface is moving fast enough that a trend is what you need.
The Wider Point
The instinct after a study like this is to argue with it — to note that the sample is retail, that Google disputes the read, that a price gap in bidet attachments has nothing to do with legal services. All of that is fair, and none of it changes the part that matters.
Two surfaces owned by the same company, drawing on the same index, answering the same question on the same day, produced almost entirely different answers. Whatever selects for the AI answer is not the thing that has been selecting for the ranked list. Firms have spent fifteen years building for the ranked list. The short list is a different competition, and it is being decided on how well a practice can be described and confirmed — not on how well it is known.
The good news buried in the retail data is that the short list is not simply the incumbent list. It rewarded sellers who were not the cheapest or the most obvious, because their data was better. A smaller practice that describes its work precisely and gets that description corroborated has a genuine opening here. That opening will not stay open indefinitely.
Clientory helps immigration law firms become visible to the families and individuals who need them most — turning AI search into consultations, and consultations into clients.
See Your AI Visibility Score → CLIENTORY.ORG