Your PI Firm's AI Search Traffic Is Hiding in "Direct": A 2026 Measurement Playbook

The Case You Cannot See in Your Reports

A woman is rear-ended on a Tuesday. That night she opens ChatGPT on her phone and asks what to do about a soft-tissue injury when the other driver's insurer is already calling. The answer names three firms in her metro. She taps one, reads two pages, closes the tab, and calls the office the next morning from her contacts list.

Your intake sheet records that as a phone call with no source. Your GA4 records the website visit as Direct. Your SEO report shows organic sessions down 6% year over year. Nothing anywhere in your reporting says "AI search produced a signed auto case."

That is the actual state of AI search measurement for law firms in 2026. The channel is not hypothetical and it is not small — it is just badly labeled. Firms are quietly making two opposite mistakes because of it: cutting content budgets because organic clicks are down, or claiming AI-search wins they cannot substantiate. Both come from the same gap.

What the Data Actually Says

Four numbers are worth knowing before you spend anything on this. Each one is publicly locatable and dated, and each one has a real limitation you should hold onto.

Put plainly: fewer clicks per search, more clicks arriving from AI search platforms, and the ones that do arrive tend to be further down the decision path. For a practice area where one signed case can be worth six figures, a channel that is small in sessions can still be material in revenue. You cannot manage it while it is invisible.

Why It Hides: The Four Leaks

1. Mobile apps strip the referrer

A tap inside the ChatGPT iOS or Android app opens an in-app browser that generally passes no Referer header. GA4 has no signal to work with, so the session is Direct. Given how much PI research happens on a phone, this is the single largest leak.

2. Copy-paste and brand recall

ChatGPT or Gemini names your firm; the user searches your name in Google or types the domain. That becomes Branded Organic or Direct. The AI answer caused it, and no analytics tool will ever attribute it.

3. AI Overview clicks are bundled into Organic

A click from a Google AI Overview is still a google.com organic click. It does not break out as its own channel in GA4.

4. The phone call

PI intake is phone-heavy. If your call tracking does not capture the web session that preceded the call, the source dies at the front desk. This is the leak most firms can actually close this quarter.

The Five-Layer Measurement Stack

No single tool answers "is AI search working." Five layers, each proving something different, do.

Layer 1: GA4 — turn the floor on

Since May 13, 2026, GA4 assigns the medium ai-assistant and groups recognized chatbot referrers into a native AI Assistant default channel, with no configuration required. Google names ChatGPT, Gemini, and Claude as examples but has not published the full recognized list.

Keep the native channel and add your own custom channel group anyway, with a referrer regex covering chatgpt.com, chat.openai.com, perplexity.ai, claude.ai, gemini.google.com, and copilot.microsoft.com. Order it above Referral so it evaluates first. Two things to know: custom channel groups do not backfill, so build a Free-form exploration for history, and referrer reliability varies a lot by platform — Perplexity citations pass referrers reliably, Claude frequently strips them, and app traffic almost never passes one.

Report it as a floor. Every AI-session number your firm reports should be labeled "minimum" in the report itself. That habit is what keeps the next conversation honest.

Layer 2: Search Console — the impression side

Open the generative AI reports and filter to your money pages: practice-area pages, city pages, and the explainer content that answers "what is my case worth" style questions. You are looking for pages that gain AI impressions while losing clicks. That pattern is not failure — it is the new shape of the funnel, and it tells you which pages are doing brand work rather than traffic work. Judge those pages on assisted outcomes, not sessions, before you delete them.

Layer 3: Crawler logs — did they even read the page?

Before debating rankings inside an AI answer, confirm the machines can read you. If your site sits behind Cloudflare (the stack we recommend for law firm sites), AI Crawl Control reports request volume, status codes, and most-requested paths per crawler and operator — GPTBot, ClaudeBot, Bytespider and others, grouped by company. What we look for on PI sites:

• Blocked or 4xx/5xx AI crawler requests — a bot rule or firewall setting quietly excluding you from the index that feeds the answers.
• Which paths get crawled — if attorney bios and results pages get hit and your practice-area pages do not, that is an internal-linking and site-structure problem, not a content problem.
• Crawl after publishing — how fast new content is picked up, which tells you what publishing cadence is even worth it.

Check robots.txt while you are in there. Plenty of firms are blocking the exact crawlers they are paying an agency to win visibility in.

Layer 4: Prompt-level visibility

Sessions and impressions still do not tell you whether ChatGPT, Gemini, or Perplexity recommends you. For that you need a fixed set of buying-intent prompts for your market — "best car accident lawyer in [city]", "do I need a lawyer for a rear-end collision in [state]", "how much do personal injury lawyers charge in [city]" — run on a schedule across ChatGPT, Gemini, Perplexity, and Google AI Mode, scored for whether you are mentioned and whether you are cited. Earning those mentions is its own discipline, covered in our PI AI visibility guide; this post is about counting them.

Two rules make this useful instead of theater. Run it logged out or in a clean session, because your own logged-in history contaminates results. And keep the prompt set frozen between runs — changing prompts and celebrating the new numbers is the most common self-deception in this whole discipline.

Layer 5: Intake — the only layer that pays

This is where firms stop short, and it is the highest-ROI piece.

• Stamp the session onto the lead. Push GA4 session source/medium, landing page, and first-touch referrer into hidden fields on every form so the CRM record carries them.
• Use dynamic number insertion so phone leads keep their source. Call tracking that maps numbers to marketing sources can pass that source straight into the CRM — CallRail's Lawmatics integration, for example, attributes calls to a marketing source automatically. Configure it once and phone attribution stops being guesswork.
• Add one intake question, asked properly. Not "how did you hear about us" as an open text box. A short list that includes "AI search (ChatGPT, Gemini, Google AI)" as a named option. It is self-reported and imperfect, but self-reported attribution is the only instrument that survives every technical leak above — and in a phone-first practice area it frequently outperforms the analytics.
• Report cost per signed case, not sessions. Once AI sessions and AI-attributed calls both land in the CRM, AI search enters the same table as PPC, LSA, and organic. That is the table that changes budgets.

A Reporting Format That Survives Scrutiny

Here is the shape of the monthly view we use. The value is in the confidence column — a report that admits what it cannot prove is the one a managing partner keeps trusting in month six.

When Not to Do This

An honest limit, because the alternative is selling a dashboard.

• If your intake is leaking, fix intake first. A firm that misses a third of its calls does not have an AI visibility problem; it has an intake problem. Measurement of a new channel is worthless while a known channel is bleeding.
• If AI sessions are in the low double digits per month, do not build a program around them. Turn on the GA4 channel, add the intake question, check crawler access, and revisit in two quarters. Those three steps cost almost nothing and preserve the data you will want later.
• Do not trade paid search budget for this yet. For most PI firms, Google Ads and LSA still produce the volume — and with LSA moving into Google Ads, that channel needs attention of its own this year. AI visibility is a compounding asset that is cheap to start and expensive to start late — it is not a substitute for the channel paying this quarter's payroll.
• Do not report a number you cannot source. If the only evidence is a vendor's proprietary "AI visibility score" with no visible method, leave it out.

The 30-Day Version

1. Week 1: Confirm the GA4 AI Assistant channel is reporting, build the custom channel group above Referral, create the Free-form exploration.
2. Week 2: Pull the Search Console generative AI reports for your top 20 URLs. Check AI crawler access and robots.txt in your CDN logs; unblock anything you are blocking by accident.
3. Week 3: Add hidden session fields to every web form. Verify DNI and that call source reaches the CRM. Rewrite the intake source question as a list with an AI search option.
4. Week 4: Freeze a 25–40 prompt set for your market and practice area, run the first logged-out baseline, and add the five-row table above to your monthly report.

FAQ

Does AI search traffic show up in Google Analytics 4?

Partly. Since May 13, 2026, GA4 assigns an ai-assistant medium and groups recognized chatbot referrers into an AI Assistant channel automatically. It misses traffic from AI mobile apps that pass no referrer, URLs users copy and paste, and clicks from Google AI Overviews, which stay inside Organic. Treat any GA4 number as a floor.

How can I tell if my law firm appears in ChatGPT or Google AI Mode answers?

Two ways, and you want both. Search Console's generative AI performance reports show impressions from AI Overviews and AI Mode for your pages. For platforms outside Google, run a fixed set of buying-intent prompts on a schedule in a logged-out session and score mention and citation rate — your own logged-in results are biased by your history.

Is AI search traffic worth more than organic traffic?

Early evidence suggests it converts better: Semrush found the average LLM-referred visitor was 4.4x as valuable as the average organic visitor by conversion rate (July 2025). That study was in the marketing vertical on small volumes, so use it as a reason to measure your own AI-attributed leads rather than as a number to forecast with.

Should a personal injury firm shift budget from Google Ads to AI visibility?

Not yet, for most firms. Paid search and LSA still drive signed-case volume today. AI visibility is a compounding asset that is inexpensive to begin and expensive to begin late, so the right move is to start measuring and building it alongside paid, not instead of it.

What is the fastest thing I can do this week?

Add "AI search (ChatGPT, Gemini, Google AI)" as a named option to your intake source question, and confirm your CDN or firewall is not blocking AI crawlers. Both take under an hour and both produce data no analytics configuration can recover after the fact.

The Bottom Line

AI search is not going to announce itself in your reporting. It shows up as a slow decline in organic clicks, a rise in Direct sessions, and callers who already know your name and cannot say why. The firms that will handle the next two years well are not the ones with the best AI visibility scores — they are the ones who wired up the plumbing early enough that when the volume arrives, they can see it, price it, and defend the budget for it.

The whole build is a week of work, and it is where our GEO engagements start: a GA4 channel group, a Search Console filter, a crawler-log check, hidden form fields, one better intake question, and a frozen prompt set. Everything after that is just reading your own instruments honestly.

If you want a look at what ChatGPT, Gemini, and Google AI Mode currently say about your firm — and where your measurement is leaking — we will run a baseline on your market and walk you through it. Book a free 30-minute strategy call and we will bring the prompt-level data for your metro to the conversation.