AI Is Already Recommending Someone In Your Category. Here’s How to Make It You.

I joined Michael Stelzner on the AI Explored podcast (Social Media Examiner) to break down how AI decides which businesses it recommends. Here’s the short version, plus the parts I’d tell a client on day one.

The short answer

Sixty-eight percent of Google searches now end without a click. The person got their answer on the results page, or from ChatGPT, or from Perplexity, and never visited anybody’s website. So the question stopped being “does my site rank” and became “does the machine mention my name.”

Something already gets mentioned when someone asks “best [your category] in [your city].” That answer exists today. If it isn’t you, it’s a competitor, and you are not losing to their marketing budget. You are losing because their content is easier for a machine to quote.

That is a fixable problem. Here is how.

What is a fan-out query?

When someone types one question into an AI tool, the AI doesn’t run one search. It generates and runs dozens of related searches behind the scenes, then assembles an answer from whatever it finds across all of them. That’s a fan-out query.

This is why “we rank for our main keyword” stopped being enough. Somebody asks ChatGPT “should I use a financial advisor or just index funds,” and the AI quietly goes looking for fee structures, tax drag, advisor credentials, market downturn behaviour, and what people complain about on Reddit. You might own the main term and appear in exactly zero of the searches that actually built the answer.

You don’t win a fan-out by writing one big pillar page. You win it by having a clear, specific answer sitting on your site for each of the narrow questions that fan out from the big one.

Why does AI ignore most business content?

Because most business content passes the swap test.

Here’s the test: take one of your articles, swap your company name for your biggest competitor’s name, and read it again. If it still makes perfect sense, the AI has no reason to prefer you. You wrote the category’s content, not your content. There is nothing in there that only you could have said, so there is nothing to cite you for.

What survives the swap test: personal stories, your own data, specific results with specific numbers, and firsthand experience.

A financial advisor writing “diversification reduces risk” is invisible. That same advisor writing about what actually happened when one client restructured their portfolio three weeks into a downturn, what they were afraid of, what it cost, and what he’d do differently, is quotable. A conference organiser writing “networking is valuable” is invisible. The same organiser writing that 240 people showed up to a dinner they’d budgeted 90 for, and what broke, is quotable.

Machines cite the sentence nobody else could write.

How should the content be structured?

In chunks that survive being ripped out of context.

AI doesn’t read your article like a person does. It pulls a chunk: one self-contained piece that directly answers the question it was asked. That chunk gets quoted on its own, next to two other companies’ chunks, in an answer your reader sees instead of your website.

So every section has to make sense with nothing before it and nothing after it. No “as I mentioned above.” No “building on the previous point.” If a section only works in sequence, it can’t be extracted, and if it can’t be extracted, it can’t be cited.

Practically, that means:

• Use the actual question as the heading. Real question, phrased the way a person would say it out loud.
• Answer it inside the first 100 words of that section. Not after the setup. Not after the story. First.
• Then add the story, the caveat, and the specifics underneath.

Yes, it feels repetitive to write. It reads fine, and it’s the difference between being quoted and being scrolled past.

What should I actually write about?

The whole journey, not just the part where someone buys.

Most businesses only write bottom-of-funnel content, because that’s the content that feels like it’s doing sales work. It’s also the most crowded content in every category.

Take a Phoenix real estate agent who works with first-time buyers. Every competitor is writing “how to buy your first home in Phoenix.” Almost nobody is writing about whether specific Phoenix neighbourhoods are safe, how the school systems compare, what HOA rules actually restrict, or how property taxes work there. Those are the questions the buyer asks six months earlier, and they fan out from the buying question anyway. Answer them and you build authority at every stage instead of fighting for the last inch.

You don’t need to guess at these. You already own the list:

• Support tickets. These are the problems your customers genuinely hit, in their own words.
• Sales calls. The same four objections come up every week. Write them down. Those are articles.
• Your inbox. The question you keep answering by email is the question thousands of people are typing into ChatGPT.

The fastest win: your newsletter archive

If you’ve been sending a newsletter, you’re sitting on years of proven content that AI has never seen. Newsletters go to inboxes. Inboxes aren’t crawlable. As far as every AI model is concerned, that work does not exist.

We did this with a consulting firm. Took the newsletter archive, ranked it by what subscribers actually engaged with, and republished the winners two ways: one version on the site for humans, and a second version structured for machines with different headings, question-led formatting, and self-contained chunks. Both live on the same domain, doing different jobs.

Within three weeks that firm owned 72% of its category in AI recommendations. They beat competitors with bigger audiences and a decade more publishing history, because the competitors’ content was generic and theirs wasn’t.

If you have an archive, start there. It’s the cheapest authority you will ever buy.

Is my website blocking AI without me knowing?

Very possibly, and this is the boring part that quietly kills everything above.

Before you write a single new article, check these four things:

1. robots.txt. Legacy settings block AI crawlers on a lot of sites. Somebody added a rule years ago, nobody remembers, and the models simply cannot read you.
2. Cloudflare. There’s an AI blocker setting. It’s supposed to be off by default. I have found it switched on for people who never touched it.
3. JavaScript rendering. If your content only appears after a scroll, a click, or a dynamic load, crawlers struggle with it. Static HTML gets read. Interactive cleverness gets skipped.
4. Sitemaps. Keep both an XML and an HTML sitemap. That gives AI two ways in, and it means a page doesn’t need a spot in your navigation menu to be found.

Then add structured data. FAQ schema and list schema tell the machine what shape your content is in, which makes the right chunk easier to lift.

None of this is glamorous. All of it is the difference between publishing and being read.

Start here: The free AI Visibility scan shows you exactly which buyer questions AI names you in today, and which ones a competitor already owns.

How do I find out whether AI mentions me right now?

Ask it, and then check the answer properly. Not “tell me about my company,” because it will be polite about it. Ask the questions your buyers actually ask, with no company name in them, and see who gets named.

We ran that at scale to find out what normal looks like. We tested 60 real businesses across 41 niches and read 4,291 AI answers from ChatGPT, Claude, Gemini and Perplexity. Two findings worth knowing before you panic or relax:

• AI names a specific business in only 21% to 33% of answers, depending on the engine. Most of the time it names nobody at all, which means the slot in your category is sitting empty rather than lost.
• Half the businesses we tested never appeared in a single answer in their own category. Not ranked low. Absent.

You can run the same thing on yourself for free. The AI Visibility scan runs your business against those same engines and shows you which buyer questions you get named in and which ones a competitor already owns. The full study is on the same page if you want to see the method before you trust the number.

Do that first. It is much easier to write the right twelve articles than another forty of the wrong ones.

Where to start this week

1. Run the free AI Visibility scan so you know which questions you are missing from.
2. Run the swap test on your three most recent articles. Be honest about the result.
3. Pull the last 20 sales calls or support tickets and write down every repeated question.
4. Check robots.txt and your Cloudflare settings.
5. Pick your best-performing newsletter and put it on your website, in question-and-answer form, with the answer in the first 100 words.

That’s a week of work and it will move you further than another quarter of blog posts nobody can quote.

Listen to the full episode: How to Get AI to Recommend Your Business on AI Explored with Michael Stelzner.

I build the Answer Content Engine: a system of agents that finds the questions your buyers are asking, pulls the answers out of your head, structures them so machines can quote them, and keeps running without you.