Last night, Search Engine Journal published a set of numbers from Profound, a company that runs its customers’ prompts through ChatGPT every day and logs where each product recommendation came from. On July 10, the share of ChatGPT Shopping recommendations drawn from merchants’ product feeds jumped from about 8 percent to about 62 percent of the runs it tracks, across 1.76 million prompt runs in July, and it has stayed near 65 percent since. The number of distinct merchants ChatGPT referenced fell by more than a fifth; the top ten stores’ share of references nearly doubled. Profound’s data is one vendor’s tracked prompts, not all of ChatGPT, and OpenAI has said nothing about a change on that date. But OpenAI’s own help pages say the same thing in plainer words: ChatGPT “picks products using structured product data” from stores and data providers. Shopify and Etsy catalogs are already connected. Everyone else is on a waitlist.

Read that as a business owner rather than a retailer and the lesson is uncomfortable. The machine did not start recommending better products. It started recommending the products it could read as data, and it read fewer merchants to do it.

On Monday, the same lesson arrived from the map. Search Engine Land published a recovered binary from Geostore, the system Google uses to represent every place in Maps: 72 ranking signals, 793 data providers, 446 kinds of local intent. The detail that matters to an owner is the difference between the listing you edit and the entity Google actually keeps about you. “The listing is the interface,” the authors write. “The entity sits underneath it.” When you change your hours or your category, your edit does not become the truth. It becomes one more piece of evidence, weighed against everything Google has ever collected about you from 792 other sources.

That is the story of this week: the machines that recommend businesses have stopped reading what you say about yourself and started reading everything about you at once, as data. And they are increasingly answering in a single sentence.

Three names, and none of them is yours

The clearest measurement of that sentence is a few months old, but nothing since has contradicted it. In May, the PR firm 5W published an index of how AI answers when a homeowner asks “my AC stopped working, who do I call?” or “best plumber in my city.” The firm ran more than 65 such prompts through ChatGPT, Claude, Perplexity and Google’s AI Overviews. Three national brands, by the index’s estimate, collected about 19 percent of every consumer answer in the category. Roughly 87 percent of independent contractors had, in the index’s words, “effectively zero AI citation share in their own metro and category.” Including, it notes, contractors with more than 800 five-star reviews and thirty years of customers.

The reason is not that the national brands do better work. The index calls the gap “structural”: AI engines weigh what it calls entity strength, the Wikipedia entries, the corporate filings, the trade-press coverage that a franchise network has and a two-truck plumber does not. The asymmetry it measured is about eightfold. Eight hundred glowing reviews do not close it.

This is where both of this week’s numbers land. The shopping data shows fewer names, chosen from data. The Geostore leak shows the same thing from the inside. Google Maps, Search Engine Land notes, is turning “into a conversational product.” Reviews, chain membership, Wikipedia signals, listing opens, direction requests, website clicks are all in the vocabulary of the ranking. The weights are not public. What is public is the direction of travel: the answer is being assembled from every trace you leave, and then delivered as one recommendation.

People did not leave Google. They added a second question

A Search Engine Journal review last Friday put the numbers next to each other. According to Similarweb, 95 percent of ChatGPT’s users still show up in Google’s audience, a figure that held from September 2025 through May 2026. And according to a Bocconi University working paper on U.S. households, those same people run fewer traditional searches once they have ChatGPT Search: about 9 percent fewer at first, 17 percent fewer after twenty weeks. Both are true. The customer has not switched. The customer now asks twice, and the second answer comes back already decided.

Pew’s February survey, quoted in the same piece, puts it plainly: 42 percent of U.S. adults have used a chatbot to find information, and 60 percent read the AI summary at the top of their search results. The person who reads that summary and calls the first name in it is not a futurist. It is your customer on a Tuesday afternoon with a leaking water heater.

The industry noticed. The industry that sells to you noticed first

Look at who published this week. Dentistry Today reported a dental marketing company adding “answer engine optimization” and “generative engine optimization” to its offer, because “patients are turning to AI tools and voice search for recommendations.” Semrush is building its Spotlight conference next month around AI visibility, citing its own study that more than a third of consumers now start a search with AI rather than a search engine. Search Engine Journal ran two sponsored guides in one week, one promising to get a local business recommended by ChatGPT in ninety days. A topic that lived in search-engineering forums a year ago has reached the trade press of plumbers and dentists. When the people who sell to small business all start selling the same thing in the same week, the thing is real.

Half of it is not yours to fix. The other half is entirely yours

Here is the honest reading. You will not build a Wikipedia entry for a bakery. You are not going to out-cite a national franchise on entity strength by next quarter, and anyone who promises you the top spot in ChatGPT’s answer is selling weather.

The glass door of a closed shop with three blank sheets of paper taped over one another, a passer-by reflected walking past.
Too many versions of the truth: one taped over the other, none of them current.

But Search Engine Journal’s other piece this week, on why AI gets brands wrong, points at the half you do control. AI does not fail for lack of information about you, it argues. It fails because “the brand already has too many versions of the truth”: the website says one thing, an old listing another, last year’s post a third, and the machine “constructs a single answer” that “looks settled even when the underlying evidence is not.” The fix is not to say it louder. It is to have one current version of yourself, everywhere, all the time.

“That’s not something marketing can solve. That is completely operational.”

Jessica Bowman, who has run search inside large companies for twenty years, said the same thing to Search Engine Land yesterday from the other direction. She calls the AI “the world’s most sophisticated buyer advisor,” and she has watched it draw on customer complaints, employee reviews and return policies when deciding whether to recommend a company. One retailer she worked with, ChatGPT actively steered her away from, citing complaints about quality and refunds. Her measure of success is no longer the click: “Are we visible when people ask? Are we mentioned? How are we described?”

A woman in a workshop pins a freshly printed photograph of a finished repair onto a cork board already covered with job photos.
A trail that is fresh and says one thing — today’s work on the board today.

And then there is what happens after the machine finally names you. Plumbing & Mechanical published, on Monday, a review of 100 contractor websites in Dallas and Houston by Michael Carpenter, who runs a marketing platform for contractors and so has a stake, but whose numbers are worth reading anyway. Thirty-four percent of those contractors never answered a web inquiry. “Not slowly. Not poorly. Never.” The top tenth answered in an average of 23 minutes; the bottom group in more than six hours. The profiles that ranked had recent reviews, not merely many; photos updated in the last ninety days, not a gallery from 2021; a services list actually filled in. His conclusion is the one that survives the week: the leads were there. The systems to catch them were not.

Put the two halves together and the picture is not gloomy. The machine reads everything about you, all the time. So the businesses it will name are the ones that look alive, all the time: current, consistent, posting, answering. That used to be a full-time job for a person you could not afford. It no longer is.

How we think about it

A woman alone at a shop counter late in the evening, answering on a tablet by the light of a single desk lamp; dark storefronts across the street.
A Tuesday inquiry gets an answer on Tuesday — including the ones that arrive after closing.
See Tampa Pasta House workSee Casa Lista work

This is exactly the work an AI employee at SmashOne is hired to do, and it does it in the owner’s own channels. An AI SMM manager keeps the pages alive every day, from the business’s own catalog, with the business’s own photos, so that the trail a machine reads is fresh and says one thing. An AI receptionist answers the line and books the appointment. An AI customer-support employee answers where people write, so a Tuesday inquiry gets an answer on Tuesday, not never. Every message and every post is visible to the owner. The setup is a conversation: you describe your business to the ChatGPT, Claude or Gemini app you already use, and the employee is ready. And before you hire, you can watch this exact work being done on two real demo storefronts, a pasta house and a home-repair shop, every day.

The order we argue for has not changed. When a customer asks a machine who to call, the machine names whoever looks most alive. Hire the AI employee first, and let it keep you looking that way. Then, if you still need to, a person.

Sources

SMASHONE CORPORATION, Florida, United States