Clicks Are Down 52%. Customers Aren’t. Welcome to the New Scoreboard.
“What gets measured gets managed.” — Peter Drucker
Dustin
August 5, 2026 · 9 min read
Last week a business owner sent me a screenshot of his analytics dashboard with one line highlighted and a one-word message: “See?”
The line was his referral traffic from ChatGPT. It was down. Way down. His conclusion — and I hear a version of this every week now — was that the whole AI search thing must be cooling off. The clicks were drying up, so the customers must be too.
Here’s the problem. That same month, his phone was ringing more than it had all year. New patients kept saying some version of “I asked ChatGPT and it said to call you.”
His dashboard said AI was dying. His front desk said AI was his best salesperson.
They were both looking at the same funnel. Only one of them was measuring it correctly.
This post is about the scoreboard. Because the old one — the one built on clicks and referral traffic — just stopped telling the truth. And if you’re still using it to decide whether AI visibility is “working,” you’re about to make an expensive mistake in one of two directions.
The Number Everyone Is Misreading
Let’s start with the stat that’s causing all the panic.
ChatGPT is sending roughly 52% less referral traffic to websites than it was earlier this year. That’s not a rumor; it’s showing up across multiple independent datasets, and publishers have been loud about it. If your only lens is “how many people clicked through to my site from an AI,” things look grim.
Now hold that number up against these, all from Q2 of this year:
- AI agent requests across the web grew 45% in a single quarter, to 17.7 billion requests.
- Automated systems now generate 57.4% of all web requests.
- Since GPT-5 launched, OpenAI’s crawling activity has roughly tripled, with its search crawler now generating more activity than its training crawler.
- In some industries the increase is staggering. Healthcare sites saw OpenAI search-crawl activity jump over 700%.
Read those two paragraphs together and the story writes itself.
The engines are reading the web more than they ever have. They’re just not sending anyone back.
AI Didn’t Stop Reading. It Stopped Clicking.
Here’s the mental shift — and it’s the whole post in three sentences.
The old internet worked like a phone book: the customer looked, then visited. Every step left a footprint in your analytics. The new one works like a trusted friend: the customer asks, the AI reads everything about you behind the scenes, and then it answers. The customer gets a name, maybe a phone number, maybe a one-paragraph summary of why you’re the right choice, and they never touch your website at all.
That’s not a failure of AI search. That’s AI search working exactly as designed.
And it means the click was never the product. The recommendation is the product. The click was just the receipt — and the receipts are going away.
The New Scoreboard
The people who measure this for a living have already made the switch. They’ve stopped asking “how much traffic did AI send me?” and started asking three better questions. Steal all three.
Mention Rate
Of all the AI answers where your business should appear, how often does it actually appear? When someone asks “best family law attorney in my area” or “who should I use for X,” are you in the answer or not?
Across industries right now, the median mention rate is about 16%. That’s the bar, and it’s low. Most businesses in most categories are simply absent.
Citation Rate
When the engines do link out, are they linking to you? Median right now is around 6%.
This one matters because a citation is the closest thing the new world has to a ranking. It means the engine didn’t just know about you; it staked its credibility on you as a source.
Share of Voice
When the AI names three businesses in your category, how often are you one of the three — and who are the other two?
This is the one that should keep you up at night, because unlike the old search results page, there is no page two. There are three names in the answer, or two, or one. If it’s not you, it’s your competitor, every single time, invisibly.
Median Mention Rate
16%
of relevant AI answers, across industries
Median Citation Rate
6%
of answers include an outbound link to you
Most businesses
~0%
are simply absent from AI answers
Notice what’s not on the list: sessions, pageviews, referral traffic. Those still exist, and they’re not useless. But they’ve been demoted from verdict to footnote.
The Customers Are Still There. They’re Just Filed Wrong.
So if AI-influenced customers don’t show up as AI referrals, where do they show up?
They show up as direct traffic, because they heard your name from ChatGPT and typed it into a browser. They show up as branded search, because the AI said “look into these three companies” and they Googled you to double-check. They show up as phone calls with no digital trail at all.
Your analytics tool files all of that under “they just knew about us.” It has no idea an AI made the introduction.
Which is why the single cheapest measurement fix available to you costs nothing and takes one sentence of training: make your front desk ask how they heard about you, and make “an AI / ChatGPT” one of the answers you actually track. Intake forms too. We’ve watched businesses discover double-digit percentages of new customers coming from AI recommendations that their dashboards attributed to nothing at all.
You cannot manage what you refuse to log.
While We’re Here: Stop Doing These Two Things
Two quick corrections on things I keep seeing businesses spend real time and money on.
Stop treating llms.txt as a strategy. Adoption of these files grew almost ninefold over the past year, because it’s easy and it feels like doing something. Here’s the uncomfortable data point: 97% of llms.txt files received zero requests from AI crawlers. Zero. The engines, so far, mostly aren’t reading them. It costs nothing to have one, and ours is live too. But if a vendor is selling you llms.txt as the centerpiece of AI visibility, you’re buying a lucky charm.
Check what you blocked a year ago. In the great bot panic of 2024 and 2025, a lot of businesses (or their web guys) blocked AI crawlers wholesale to “protect their content.” Those crawlers are not one thing. OpenAI alone runs separate bots for training, for search, and for live user requests — and the search bot is now the busiest of the three. If you blocked them all, you didn’t protect your content. You took yourself out of the answer. Five minutes with your robots.txt file is worth more than most marketing meetings this month.
A Quick Note on the Numbers
Same policy as always in this series: the figures above come from third-party research. The referral decline and crawl data come from large-scale log analyses (billions of events, tracked by firms like Botify and reported across the trade press), the agentic traffic and llms.txt numbers from published Q2 2026 datasets, and the mention and citation medians from publisher-side AI visibility tracking reported by Digiday. They’re directional, not gospel. The engines change their plumbing constantly and the exact percentages will move. The pattern is what matters, and every dataset points the same direction: reading up, clicking down, answers in between.
Your Move
Here’s the thirty-minute version of the new scoreboard — no tools required.
Write down the ten questions your best customers would ask an AI to find a business like yours. Real questions, in plain language, the way an actual person would type them. Once a month, run all ten through ChatGPT, Gemini, and Perplexity. For each one, log three things in a spreadsheet:
- Were you mentioned?
- Were you cited or linked?
- Who else was named?
That’s it. Thirty queries, three columns, once a month. In ninety days you’ll have a trend line for the only metric that matters now: are you in the answer? And you’ll know exactly which competitors are eating the recommendations you’re missing.
Then go retrain your front desk with the one-sentence fix, and give your traffic dashboard the demotion it’s earned.
The customers didn’t leave. The footprints did. Time to start measuring what’s actually happening.
Free · $997 Value
Find Out Where You Stand in AI Search
Get your free AI Visibility Assessment. See exactly how often ChatGPT, Gemini, and Perplexity mention your business, who they recommend instead, and how to change that.
Get Your Free Assessment →Frequently Asked Questions
Does falling AI referral traffic mean AI search isn’t worth investing in?
No, it usually means the opposite. AI engines are reading the web more than ever (OpenAI’s crawl activity roughly tripled after GPT-5), but they increasingly answer customers directly instead of sending clicks. The recommendation still happens; the click doesn’t. Measure mentions and citations, not referral sessions.
How do I measure my AI visibility without expensive software?
Run the same ten customer-style questions through ChatGPT, Gemini, and Perplexity once a month and log whether you were mentioned, whether you were cited, and who else was named. Pair that with asking new customers directly how they heard about you, including “an AI” as a tracked option.
What is a good mention rate in AI answers?
Published tracking across industries puts the median around 16% of relevant answers, with links appearing about 6% of the time. If you’re being named in a quarter or more of the answers where you should appear, you’re well ahead of the field. Most businesses are at or near zero.
Should I create an llms.txt file?
It’s fine to have one, and it takes minutes. Just don’t mistake it for a strategy: 97% of llms.txt files received zero crawler requests in the latest research. Your listings, reviews, structured data, and content depth move the needle. llms.txt, so far, mostly doesn’t.
Could I be blocking AI engines by accident?
Yes, and it’s common. If your site blocked “AI bots” broadly in the last couple of years, you may be blocking the search crawlers that feed AI answers, not just the training crawlers. Have someone review your robots.txt and confirm you’re allowing the search and user-request bots.
— Dustin
AIOnramp helps service businesses get found, and chosen, in AI search. We specialize in Generative Engine Optimization (GEO): the strategies and infrastructure that make your business visible to ChatGPT, Gemini, Perplexity, and the AI tools your customers are already using. Figures cited are drawn from third-party research and are directional. aionramp.ai