Interactive explainer

How AI picks a business to recommend

Someone asks an assistant who to call. Seconds later it names three businesses. Scroll to watch one answer get built, step by step, then flip the switches to see what moves you in or out of it.

Pick your industry

Step 1 · The question

It starts with a plain question

Nobody types keywords at an assistant. They ask the way they would ask a friend:

The answer will usually name a handful of businesses. Everything below is about how those few get chosen, and why yours is or is not one of them.

Step 2 · The searches

One question becomes several searches

The assistant does not look up your question word for word. It breaks it into related searches and runs them together. Google calls this “query fan-out” and describes it for AI Overviews and AI Mode.

So the pages that matter are not only the ones matching the question. They are the ones matching each smaller search behind it. The three shown are examples of the kind of searches involved.

Source: Google Search Central, AI features.

Step 3 · Discovery

It gathers every source it can read

Your website is one source among several. The assistant can also draw on your Google Business Profile, review platforms, directories and registers, and local coverage. Each one describes your business in someone else’s words.

A source it cannot reach does not count. If your site turns AI crawlers away, the picture gets built without your own account of who you are. You can test that in seconds with the AI crawler checker.

Step 4 · Verification

It checks whether the sources agree

A claim that appears in one place is weak. The same name, address, hours and services across your site and your listings is strong, because the sources confirm each other.

When your hours differ between your website and a listing, a system comparing them has less reason to trust either. The fix is dull and effective: one set of facts, published everywhere you control, and corrected wherever someone else publishes it.

Step 5 · Recommendation

It writes the answer and names a few

Now the assistant composes a reply that fits the question, and presents the businesses it can best support with evidence. Being used as a source is not the same as being named, and being named is not the same as being recommended.

In a June 2026 study, Semrush found that 61.7% of citations were “ghost citations”: the page was linked as a source, but the brand was never named in the answer.

Source: Semrush with Kevin Indig and Growth Memo.

Step 6 · Action

Then a person has to be able to act

A recommendation is only worth something if the next step is easy: a phone number that works, a booking link, clear hours. Most local journeys still end with a person clicking a link or making a call.

If the assistant recommends you and the visitor then cannot find how to book, the recommendation went to waste.

Your turn · Change the inputs

Flip a switch. Watch the answer change.

These are the things a business controls. Turn them off one at a time and see where ends up.

A simplified model for planning, not a prediction. The business names are placeholders.

A simplified model for planning, not a view inside any assistant. No assistant publishes how it ranks businesses, and each one finds and weighs sources its own way. The business names are placeholders.

Cited, named, recommended: three different results

The picture above ends with a business in the answer. In practice there are three ways to show up, and they are worth very different amounts.

Cited

The answer used one of your pages as a source. Your business may not be named at all.

Named

Your business name appeared in the answer. It may not have been presented as a good choice for this need.

Recommended

The answer presented your business as a suitable choice. This is the one that sends customers.

Counting citations alone overstates progress, because a citation is the weakest of the three. We go through the evidence in Getting Cited Isn’t Enough and the full framework in AI Agents and GEO.

About this model. The steps on this page are a planning framework, the same one we use with clients. They are not a description of any platform’s internal ranking system. ChatGPT, Claude, Gemini and Perplexity each retrieve and weigh sources in their own way, and none of them publishes the recipe. The framework is useful because it separates problems that need different fixes. Nothing here guarantees a recommendation.

What you control at each step

DiscoveryLet AI search crawlers in, and keep the important facts in the page itself, not hidden behind scripts. Check access with the AI crawler checker.
VerificationPublish one set of facts everywhere. Mark them up so machines can read them: our schema templates are ready to copy.
RecommendationEarn mentions on sites you do not own, and collect detailed reviews on more than one platform. Why volume alone is not enough: what reviews tell an AI.
ActionPut a working phone number, booking link and current hours where both a person and an assistant can find them.

Want the version for your field? See how this plays out for .

Frequently asked questions

Is this how ChatGPT actually works inside?

No. The steps on this page are a planning framework, not a description of any assistant's internal system. ChatGPT, Claude, Gemini and Perplexity each find and weigh sources in their own way, and none of them publishes how it ranks businesses. The framework is useful because it separates problems that need different fixes.

Why does AI recommend my competitor and not me?

Usually one of four gaps: the assistant cannot read your site, your details disagree between sources, few other sites describe you, or your pages do not answer the question being asked. The switches on this page show how each gap changes the outcome in this model. The free AI scan shows which ones apply to your business.

Do my Google reviews decide whether AI recommends me?

They are one input among several. A high star count on a single platform gives an assistant little to work with. Detailed reviews spread across more than one platform, alongside consistent business details, give it more.

Can I guarantee that AI will recommend my business?

No, and anyone who promises it is guessing. You can make your business easy to find, easy to verify and easy to choose. The assistant still decides what to say.

How do I find out what AI says about my business today?

Run the free AI scan. It asks ChatGPT, Claude and Gemini the kind of questions your customers ask, shows whether your business is mentioned, and lists what to fix first.

Sources

See the real answer for your business

The picture above is a model. The free AI scan is the real thing: it asks ChatGPT, Claude and Gemini the questions your customers ask and shows whether your business comes up.

Check AI crawler access →