The First Controlled Test of Google's AI Mode Is In. Here Is What It Means for Your Intake Pages.

For two years, the claim that AI answers mean fewer visits to your website has rested on correlations — traffic charts that bent downward around the time AI Overviews rolled out, with plenty of other explanations sitting in the same window. In late September, trade press picked up the first study that measures the effect directly.

Researchers at the University of Pennsylvania and Northeastern University ran a preregistered, randomized field experiment on Google Search. They tracked roughly 1,100 U.S. Chrome users through a browser extension and, for seven days in March 2026, assigned each person to one of three conditions: regular Google with AI Overviews hidden, Google as it normally behaves, or Google forced into AI Mode. The paper, AI in Search Reduces Publisher Referrals Without Improving User Experience, was posted to arXiv on August 18.

The headline number: people pushed into AI Mode clicked through to outside websites 18.8 percentage points less often. Clicks to Reddit fell 21.2 points. Clicks to Wikipedia fell 9.9. Search sessions dropped by almost one per day.

What the Study Actually Shows

Two findings matter more than the click figure.

First, the experience got worse, not better. Participants in the AI Mode group reported lower satisfaction, lower usefulness, less sense of control, and less trust in what they found on Google. In open-ended feedback, 34% were negative and 29% positive. The recurring complaints were difficulty navigating directly to websites and — in the researchers' framing — limited links and limited source diversity. Some of them simply went elsewhere: use of a competing engine rose 11.2 points.

Second, the reverse held too. Removing AI Overviews increased clicks to outside sites without any detectable drop in satisfaction.

The honest caveats belong up front. The sample skewed younger and more highly educated than the general population. The week was forced — real users choose when to open AI Mode, and before the experiment it accounted for well under 1% of their searches. And the mechanism for hiding AI Overviews only worked about half the time after Google changed its page markup mid-study. This is one careful experiment, not the final word.

But it is the first time the direction of the effect has been measured rather than inferred. When a synthesized answer sits between the question and the web, fewer people click through.

Why "Traffic Is Dying" Is the Wrong Read

It would be easy to take this study as a reason for despair, or as a reason to stop investing in the firm's website. Both would be mistakes.

Think about who still clicks. Someone asks an assistant, my H-1B employer is filing a transfer — can I start working for the new company before approval? The answer explains H-1B portability, the pending-petition rule, and the basic timeline. The person reads it. Most of them stop there. The ones who click through to a firm's site are not looking for the explanation they just received. They are looking for the next thing: can this practice handle my case, how long does it take, what does it cost, and how do I talk to someone?

That is a different visitor from the one your site was built for. For years, practice-area pages were written to catch the person at the very top of the question — "What is adjustment of status?" "What is an RFE?" — and walk them slowly toward a consultation. The AI answer now does that walking. The visitor who arrives has already been walked.

Fewer visits, arriving later in the decision. That is the practical reading of this study.

What Happens When the Page Re-Explains the Basics

Picture that H-1B visitor landing on a typical immigration firm page. The first four paragraphs define the H-1B classification, explain the specialty occupation standard, and summarize the cap lottery. The information about whether the firm handles transfers, how quickly it can file, and how to schedule a consultation is buried near the bottom, or split across three other pages.

To that visitor, the page reads as a slower, less specific version of the answer they already have. They hit back. And because the study's participants specifically complained about how hard it was to get to websites at all, the few who make the effort have less patience for a page that wastes it.

The pages that hold these visitors answer the questions the AI answer cannot:

• Your process. What happens after they contact you, step by step, for this specific matter type.
• Your timing. How quickly you can review a family-based petition, respond to an RFE, or file a transfer — and what USCIS processing looks like on top of that.
• Your fees. A range, a flat fee, or at minimum how consultation fees work. Visitors who arrive with the basics answered are often comparing firms, and silence on cost reads as a reason to keep comparing.
• Your fit. The case types you handle and the ones you refer out. An asylum practice that does not take employment matters should say so plainly.
• The booking path. One clear way to schedule a consultation, visible without scrolling, with languages offered and what to have ready.

Do This Week

1. Pull your five most-visited practice-area pages. For each, count the paragraphs before a visitor learns how to book a consultation. If it is more than two, move the booking path up.
2. Add a "What happens next" block. Three to five numbered steps describing your intake process for that matter type — from first contact to engagement letter.
3. Publish timing you can stand behind. Your internal turnaround for common filings, clearly separated from agency processing times, with a link to USCIS's published processing times page.
4. State how fees work. Even a sentence — "Consultations are $150 and credited toward your flat fee" — answers the question most AI-referred visitors are carrying.
5. Cut the definitional preamble. Keep a short, accurate summary of the law for context, but put it after the practical information, not before it.
6. Test one question end to end. Ask Google AI Mode, ChatGPT, and Perplexity a real client question you handle — a DACA renewal timing question, a consular processing delay, a PERM question — scoped to your city. Read the answer, then click to your own site as if you were that person. Does your page pick up where the answer left off?

The Bigger Picture

This study landed in a September already crowded with evidence that the front door of search is moving. What it adds is causation: when the answer is synthesized, fewer people walk through to the source.

For a small immigration practice, that is not the end of the website. It is a change in the website's job. The page no longer has to teach a stranger what an adjustment of status is. It has to convince someone who already knows — and is deciding between three firms — that yours is the one to call.

Fewer visitors, better informed, closer to a decision. A firm that builds its pages for that visitor will turn a smaller number of clicks into more consultations than it did when the clicks were plentiful.

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