ChatGPT Stopped Searching the Web. Does It Already Know Your Business?

The rules changed on a Tuesday and nobody sent a memo

On August 8, GPT-5.6 became ChatGPT's default model. On paper that is a version bump. In practice it rewrote how you get recommended, and it did it overnight.

Here is the change in one sentence. ChatGPT stopped searching the open web and citing whatever ranked, and started pre-selecting a shortlist of domains it already trusts, then searching only those. Multiple independent datasets caught the same shift. Reddit's citation share fell from roughly four percent to under half a percent. Help centers and official documentation jumped from two percent of citations to thirty-two, the single biggest winner. Smaller company domains got squeezed.

Read that again, because it is the whole article. The model decides who is in the room before it starts looking for answers. If your business is not on the shortlist, no amount of great content gets retrieved, because the search never touches your page.

Getting retrieved is dead. Getting named is the job.

For two years the GEO playbook was retrievability. Get your content crawlable, get it cited, seed a few Reddit threads, land on a listicle. That playbook stopped working on a single Tuesday. Reddit did not get outranked. It stopped getting invited.

The new question is not whether your content can be found. It is whether the model knows your name for your category before it searches. Ask ChatGPT for the best enterprise software in some category and it will query specific brands by name, because it already decided which vendors belong. That is category eligibility, and it is a reputation problem, not a ranking problem.

There is a nastier wrinkle underneath. A domain can show up heavily in the retrieval layer and still earn zero citations. Reddit did exactly that in one test, present in the results, absent from the answer. So counting citations tells you almost nothing now. Presence and naming are the game.

A customer is asking an assistant about your category right now

Someone in your service area is asking an AI assistant who to call for a plumber, a roofer, a dentist, today, before they ever open a search results page. The assistant is going to name a few businesses. The only question is whether yours is one of them.

If the model has no idea you exist for your category, you are not losing a ranking. You are absent from the shortlist entirely, which means you never enter the conversation. The searcher gets three names, calls one, and the job is booked before your website would have loaded.

This is the part that should get your attention. It is not that you rank lower. It is that the decision now happens on a surface you do not own, in a moment you cannot see, against a shortlist you were never added to.

Local businesses were building this moat before it had a name

My agency, Local Howl, does local SEO for service businesses. HVAC, roofing, the trades that live and die by the map pack. And when I read that AI search now rewards durable brand authority over placement hacks, my first thought was that we have been building durable brand authority for local businesses the entire time. We just called it other things.

A consistent name, address, and phone across every surface. Google Business Profile categories that match the services on the site. Steady reviews on the platforms customers actually trust. Being the business that is clearly, repeatedly associated with a category in a place. That is category eligibility for a local market. We were building the shortlist before the shortlist was the point.

The enterprise SEO world is now panicking about how to get named by a model. Local operators have a head start, because getting named in your town is the work we already do. The map pack is a shortlist. Reviews are a reputation signal. The assistant is just reading the same evidence and deciding who belongs.

You cannot fix a shortlist you cannot see

Here is the problem with all of this. The shortlist is invisible. You cannot open a dashboard and see which domains ChatGPT trusts for your category, whether it names your business, or how that changed after the last model update. And model updates now break strategies from one day to the next, so guessing is not a plan.

That is a measurement gap, and measurement is where I stop hand-waving and reach for a tool. You need to see, per category and per market, whether AI assistants actually surface your business, which competitors get named while you do not, and which sources are feeding those answers.

Semrush's AI Visibility Toolkit is what I would point you at, because it measures exactly this: how often your brand appears in AI answers across ChatGPT, Gemini, Perplexity, and Google AI Overviews, how your share of voice compares to the competitors getting named instead of you, and which sources and topics are driving those mentions. You cannot correct a mischaracterization you never saw, and you cannot earn a spot on a shortlist you are not tracking.

Stop trying to get found. Start getting named.

The businesses that win AI discovery are not the ones with the cleverest content hacks. They are the ones a model already associates with a category and a place. For local businesses that is winnable, because the signals that earn it are the same ones we have always built. If you want the wider context on why the click stopped being the point, I wrote that up in the zero-click field guide, and the deeper mechanics of why AI cross-checks you against the whole web live in the entity corroboration piece.

Retrievability is over. Eligibility is the game. Make the model know your name for your category, in your market, and you are on the shortlist before the search even starts.