It gave me four names. Mine was not one of them. Two of the four I would not have hired.

Then I asked it directly about me, and it got my work partly wrong in a way that would have quietly cost me a client. Not offensively wrong. Confidently, plausibly wrong, which is worse, because nobody stops to check.

Everyone googled themselves at some point. Nobody has done the AI version, and AI is now what people use before they call you. 94% of B2B buyers say they use it during purchasing.

The audit below takes about thirty minutes and costs nothing.

The 20-second version: When someone asks AI for a recommendation in your category, or types your name into it before a meeting, an answer comes back. You are in it or you are not, and what it says is either right or it is not. Almost nobody has checked. This is how you check, in about thirty minutes, at no cost.

Time: thirty minutes. You need: ChatGPT, Gemini, Claude, and Perplexity. The no-cost tier of each is fine. Payoff: you find out where you actually stand, which is the part everyone skips.

Why bother, given AI referrals are a small slice of traffic: because of who that slice is. Ahrefs reports AI search at 0.5% of their visitors but 12.1% of their signups, a 23x conversion rate. Webflow sees ChatGPT traffic converting at 24%, six times their Google traffic. Surfer says roughly a quarter of their new customers now arrive from AI assistants.

The mechanism is obvious once you see it. By the time someone arrives from an AI answer, they have already done their research inside the AI, and the AI told them you were the one. They turn up pre-sold.

The other number worth knowing: across 6,820 measured AI answers, a given brand appears roughly one time in six. Five times out of six, somebody else got named.

Not your name on its own. That is the vanity search and it tells you very little.

You want the questions a real person types when they have a problem and no idea who to call. Pick ten from whichever set fits you.

If you sell something:

If you are a freelancer or consultant:

If you are employed or job hunting:

That last set matters more than people think. Recruiters and hiring managers research candidates with AI now, the same way they moved to LinkedIn fifteen years ago. There is a machine-written summary of your professional reputation and you have never read it.

ChatGPT, Gemini, Claude, Perplexity. All four, every prompt.

This is not thoroughness for its own sake. These engines run on different retrieval systems and prefer different sources, so a brand cited constantly on one can be completely absent from another. For scale, ChatGPT accounted for 56.7% of AI assistant traffic in March 2026 and Gemini around 18%, with the rest spread across Claude, Perplexity, Copilot and Grok (Similarweb, via AI Labs Audit).

Use this prompt rather than just asking the question cold, because it makes the engine explain itself:

Point four is the one that turns this from a mirror into a diagnosis. The engine will tell you what is missing.

For every answer, write down three things and nothing else:

Are you named? Yes or no. No partial credit.

Is what it says accurate? This is the one people skip, and it is often where the real damage is. Being described wrong is worse than being absent, because a wrong description gets repeated with confidence.

Who got recommended instead? Write down every name. That list is more useful than anything else you will produce today.

Thirty minutes and you have a page that tells you exactly where you stand.

I have run this for myself and for a few people who let me, and it comes out one of three ways.

Invisible. The engine has no idea you exist and recommends four other people. Most common by a distance, and the least alarming, because at least it is fixable.

Described wrong. It knows you, and what it says is out of date, off-target, or subtly incorrect. This is the one that costs you money quietly, because the prospect never mentions it.

Beaten by someone worse. It names three competitors you consider inferior. Infuriating, and genuinely useful, because it tells you what those competitors have that you do not, and it is almost never that they are better at the job.

The audit tells you which one you have. It does not tell you why, or what to do about it, and those are different problems with different fixes.

🔒 Inside the Pro section:

The AI Visibility Pack. The audit tells you where you stand. This is what to do about it:

In your library: the Connector Playbook, Google's AI Tools guide, and all past resources.

AI earbuds that record, transcribe and summarise the conversation you are actually in. Three microphones, a two-metre pickup range, six hours of battery, and 4G so they work without a phone nearby.

The interesting part is the form factor. Meeting transcription tools have solved the video call. Nobody has really solved the conversation that happens in a room: the site visit, the client lunch, the corridor conversation where the actual decision gets made and nobody writes it down.

Worth watching if you spend your working life in rooms rather than on calls.

📊 Google is adding AI visibility reporting to Search Console. Reporting for AI Overviews, AI Mode, and AI-powered Discover is being added directly into Search Console, starting with a subset of UK websites (MarketingProfs). The manual audit in this issue is the version you can run today. This is the version that eventually runs itself, so get your structure right before the data arrives.

✂️ A plugin cut one developer's token use by about 65%. Caveman strips pleasantries, hedging and transitions out of model responses. The model is unchanged, the output just stops containing filler (MarketingProfs). If you pay per token, specify output length and format instead of paying for conversational padding.

🛡️ OpenAI called for collective action on cyber defence. The company is urging governments, security firms and other AI labs to strengthen defences before AI-enabled attacks become widespread, warning that hospitals, water treatment plants and internet infrastructure face growing risk (Future Tools).

Quick version: From September 15, Cloudflare applies new defaults that block AI agent and training crawlers on ad-supported pages while still allowing search, and every new domain gets those restrictions automatically. This lands in the same year AI became a genuine front door for finding businesses. Two reasonable positions are now in direct conflict, and most site owners will end up on one side by default rather than by choice.

Cloudflare now lets site owners control three types of crawler separately: search bots that index you, agent bots that fetch your page on behalf of a user asking a question, and training bots that use your content to build models.

That separation is genuinely good. Those three things were always different and treating them as one was crude. Wanting to appear in search while objecting to being training data is a perfectly coherent position that was previously hard to express.

The friction is in the default. New domains get agent and training bots blocked automatically (Crescendo). Somebody launching a business site next month inherits a decision they did not make, in the direction of being less visible to exactly the channel this week's workflow is about.

Both sides of the argument are reasonable, which is why this will not resolve quickly.

Publishers have a real grievance. AI answers summarise their work and often send nothing back, so the traffic that funded the work goes to the summariser. Blocking training crawlers is a rational response to being used as raw material.

But the same infrastructure that blocks training also affects being fetched when a customer asks about you. For most small businesses, being quoted in an AI answer is not extraction, it is a referral. The Ahrefs and Webflow numbers earlier in this issue are what that referral is worth.

The awkward truth is that these two groups have opposite interests and are using the same switch.

Which leaves a practical question that most site owners have never considered and now need to answer.

🔒 The Pro breakdown: how to check whether your own site is currently blocking AI, which of the three crawler types you actually want, and the honest case on both sides of letting AI train on your work.

Workflow Library Updates:

NEW Resources:

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

The part that stayed with me was not being left off the list. It was the wrong description, because I would never have found out. Nobody emails to say the AI told them something odd about you. They just do not call.

Half an hour tonight and you will know. Whether you like the answer is a separate question.

👇 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.