In under a year, the share of Google's AI citations coming from top-10 organic results fell from about 76% to about 38%.

Read that again slowly. Ranking on page one used to be roughly the same thing as being the answer. Twelve months later it is a coin flip.

And it gets more awkward. Fewer than half the brands leading in traditional local search also show up when AI makes a recommendation. The position you spent years earning is slowly becoming a different game, played by different rules, and almost nobody is watching the scoreboard. Only 14% of marketers track whether AI cites them at all.

I have put the whole audit below, and you can run a rough version in a couple minutes.

The 20-second version: Your customers are asking AI who to use. 45% of consumers used ChatGPT, Gemini or Perplexity to find a local business in the past year, up from 6% the year before. Whether you get named is checkable, and almost nobody has checked. Three minutes gets you a rough answer. The full audit gets you a baseline you can measure against.

Time: three minutes, or an afternoon for the proper version. You need: ChatGPT, Perplexity, and Google's AI Mode. No accounts beyond what you have. Payoff: you stop guessing about a channel that grew seven-fold in a year.

Three findings changed how I think about this, and they are all in the guide below.

These systems are brutally selective. SOCi's 2026 Local Visibility Index found AI platforms recommend only 1.2% of business locations on ChatGPT and 7.4% on Perplexity, against 35.9% visibility in Google's local 3-pack. Google shows you a short list. AI names one or two.

Your Google position does not carry over. The citation figure at the top of this email is the clearest evidence, and fewer than half of local search leaders appear in AI recommendations.

Each system is its own world. An analysis of 680 million citations found only 11% of domains are cited by both ChatGPT and Perplexity. Brand recommendations differ by 40 to 60% across platforms for the same query. Strong on Gemini and invisible on ChatGPT is a common pattern, and an average across systems hides it completely.

Do this before you read any further.

Open ChatGPT, logged out or in a private window so your own history does not skew it, and ask three questions. Not your work, business or name. The questions a customer with a problem and no idea who to call would type.

Then write down three things for each answer:

Were you named? Yes or no, no partial credit.

Was what it said accurate? Being described wrong is worse than being absent, because a wrong description gets repeated confidently and nobody tells you.

Who got recommended instead? That list is the most useful thing you will produce today.

Asking the question tells you where you stand. This one tells you why, and it is the single most useful prompt in the whole audit. Run it straight after one of the three above, in the same chat.

Point four is where the value is. The engine will tell you what is missing, and it is usually more specific than you expect: no reviews it can find, no service page for the thing you actually do, no third-party source describing you, or information that contradicts itself across your own site.

Mentioned and recommended are not the same thing, and conflating them is the most common mistake people make reading their own results.

AI might name your business in passing while actively recommending three competitors. Mentioned means you exist in the answer. Recommended means it is pointing the customer at you. Track them as separate columns, because the fixes are different. Being mentioned but never recommended usually means the information about you is thin rather than absent, which is a more solvable problem than invisibility.

Answers move, so one result is not evidence. Ask the same question twice and you may get different businesses named. Models update silently, which is why the date matters as much as the result. Run enough queries that patterns emerge, and treat everything you record as a dated snapshot rather than the truth.

Mention rate. Recommendation rate. Average position when named. Competitor frequency. Source frequency, meaning which domains keep getting cited. And whether your own website is used as a source at all.

Source frequency is the one people skip and the most actionable. Sort those domains, and when four of the top five have no listing for you, that is not a vague problem, that is a list.

Do not panic at a low recommendation rate. Given ChatGPT recommends around 1.2% of business locations, compare yourself to your competitors rather than to a hopeful benchmark.

The three-minute test tells you if there is a problem. It does not give you a baseline, and it cannot tell you where those answers came from, which is the part that actually matters.

So I have put the full thing together for all members as a standalone guide in Fifteen parts. Access it in our resource library: The AI Visibility Audit.

What is in it: the master prompt that builds your own 20-query set from your business details in the words a customer would actually use, the tracking table, six metrics with the formulas, the competitor analysis, the source analysis that tells you which websites are actually shaping answers in your category, a website clarity checklist, and a scoring framework so you can compare yourself against yourself in three months.

The source analysis is the part most people skip and it is the most valuable. Every citation in your results tells you something about the information ecosystem around the businesses AI does recommend. When you sort those domains and find four of the top five have no listing for you, that is not a vague problem. That is a list.

One honest note before you run it. Nobody outside these companies knows how they rank businesses, and anyone telling you otherwise is guessing. The audit gives you evidence about what is happening right now and where the gaps are. That is enough to act on, and it is more than almost anyone in your category has.

🔒 Once you have the diagnosis, the question is what to do about it. Four workflows from the companion guide are below, including the one that finds the exact words your customers use to describe you.

AI Visibility Improvement and Management. The audit is the diagnosis. This is the treatment, and it runs on one principle: do not try to trick AI into recommending you, build an information footprint that gives it legitimate reasons to.

That is not only an ethics argument. Ahrefs analysed 75,000 brands and found branded web mentions correlate with AI visibility roughly three times more strongly than backlinks, 0.664 against 0.218. And when Google swapped Gemini 3 into AI Overviews in January with no announcement, thousands of businesses saw visibility shift overnight, which is what happens to anything built on a specific model's quirks.

Inside the guide:

Also in your library: the Connector Playbook, Google's AI Tools guide, the Skills Bank, and every workflow pack from past issues.

A browser where the address bar answers instead of searching. Rather than returning a list of links, Comet reads the live web for you and synthesises what it finds into a cited report as you go.

The reason it belongs in this issue is that it is the same shift the whole workflow is about, just from the other side of the glass. Everything above is about being found when AI answers a question. This is what that answering looks like when it replaces your own browsing.

Where it earns its place in a working week: competitive research, preparing for a meeting with an account you do not know well, and market analysis. Anything where you currently open eleven tabs, skim them, and assemble a view in your head. It replaces search plus manual synthesis rather than replacing a chatbot.

📉 Small websites lost three times more search traffic than large ones. Chartbeat's 2026 network data breaks the decline down by site size: small sites lost 60% of search referral traffic, medium sites 47%, and large sites 22%.

🤖 AI referrals are growing fast and are still tiny. Referrals from tools like ChatGPT grew more than 200% year over year but still account for under 1% of total publisher pageviews (Newor Media). Why this matters for you: anyone telling you to abandon search for AI is early. The case for AI visibility is about who arrives, not how many.

🚪 Some publishers are seriously considering leaving Google entirely. Traffic fell more than 40% for some publications between June 2025 and June 2026, and the blocker is that turning off Google's crawlers stops AI training and search indexing at the same time (Nieman Lab).

🔇 Google changed the model behind AI Overviews without telling anyone. On 27 January 2026 it swapped Gemini 3 into AI Overviews and AI Mode with no announcement, and thousands of businesses saw their visibility shift overnight.

Quick version: Search referral traffic collapsed for small publishers and barely moved for large ones. Small sites down 60%, medium 47%, large 22%. But total pageviews across the tracked network fell only about 6%, which means this is redistribution rather than collapse. The traffic went somewhere. It mostly did not go to AI.

The headline everyone has run for eighteen months is that AI killed search traffic. The data is more specific than that, and the specifics are what matter if you are small.

Chartbeat tracks billions of pageviews monthly across thousands of sites. Their 2026 breakdown by size: small sites lost 60% of search referral traffic, medium sites 47%, large sites 22%.

If you run something small and your traffic fell off a cliff while the big names in your category seem fine, that is not your imagination and it probably is not something you did.

Now the part that reframes it. Total pageviews across that same network fell only about 6% year on year. Google Search pageviews fell 34% and Google Discover 16% between December 2024 and December 2025 (Newor Media).

So the audience did not leave. The route changed.

And here is the finding that should stop anyone from overreacting in the other direction. Referrals from AI chatbots grew more than 200% year over year and still account for less than 1% of total publisher pageviews.

Both things are true at once, which is why the advice in this space is such a mess. AI referral volume is nowhere near replacing search. And AI visibility still matters, because the visitors who do arrive that way are worth several times a standard organic visitor, having already been qualified by the conversation before they clicked.

Volume moved somewhere else entirely. Internal traffic and dark social absorbed most of it, which is a polite way of saying people are arriving through channels you cannot measure and mostly cannot buy.

Which raises the question worth sitting with this week.

🔒 The Pro breakdown: where the traffic actually went, why small sites absorbed three times the damage, and what to build that does not depend on a channel somebody else controls.

Workflow Library Updates:

NEW Resources:

Each resource lives permanently in your Pro account. Use them whenever you need them.

If you have joined in the last week or two, welcome. This is roughly what every Thursday looks like: one thing you can actually run, the story behind why it matters, and no pretending anybody has this fully worked out.

Start with the three-minute version. Three questions, logged out, and write down who gets named instead of you. It takes less time than reading this issue did, and most people find the answer more interesting than they expected.

👇 What’s behind the paywall:

Pro members deploy AI in their work an average of 5-8x more often than free readers (based on reply data from past issues). The difference is having the exact setup, not the concept.