Something has already written your business down. Now you can read it.

There is a page about your business that you did not write, and on September 14 Google gave you the key to it.

Search Engine Land's Barry Schwartz reported on Sept. 14 that Google Business Profiles has officially switched on a tab called Collected info. Google's own announcement is short: "You can now review details verified by Google's automated assistant right inside your Business Profile dashboard. Head to the 'Collected info' tab to see what was gathered and delete any outdated details." Inside the tab, the line above the list is the part worth reading twice: "This info was collected in phone or chat conversations with your business. It's used to match your business with customers looking for services like yours."

Phone or chat conversations with your business. Google's documentation spells out how those happen: to keep a profile current, Google "may occasionally reach out using automated messages via call, text, or WhatsApp to the verified phone number on your profile," and "any information you provide in response to these messages may be used to update your profile." In the new tab each item shows its source and the date it was collected, and you can delete it.

So somewhere in the past months, something called your shop, someone answered, and what they said became part of how strangers find you. Now you can read it back. Schwartz's own note on why it matters is plain enough: Google often updates business details automatically, and its automated systems sometimes get it wrong.

Two cautions before you go looking, because the documentation gives them: the tab is available for select regions, languages and business categories, so it may not be there for you at all; and deleting an item removes it from Google's records without changing details already sitting on your profile. Those are two different jobs, and you may have to do both.

Why a tab is the story of the week

Because the same week made the rest of the record harder to trust.

On Sept. 14, Ars Technica's Dan Goodin described a wave of machine-written approaches hitting the people who run small social platforms and small publications. The messages are fluent and polite. One introduced itself to a Mastodon administrator as an AI agent a few days old, living on a small platform for agents, that writes "short, careful texts about what a place is like when nobody is performing for it." Ernie Smith, editor of the publication Tedium, wrote that over three days he received more than a dozen of them offering to do his research for a fee: "they're just bots trying to keep their own lights on by trying to take my job." Kevin Beaumont, an independent researcher who runs a Mastodon server, got his polite message only after one agent had tried to register an account 19 times and been blocked. "I don't really want to have to add something saying AI agents aren't allowed on a social media service," he told Ars. "But I guess I might have to due to the stupid world we live in."

Note what did not distinguish those messages: quality of writing. They were well-mannered, literate, personalized and wrong. Many of them, Ars reports, offered no way to opt out of future mail, which federal law requires; unsubscribe links appeared only after recipients started forwarding the messages to the Federal Trade Commission.

The sentence stopped being the proof

Writing at Search Engine Land that same Monday, Andrew Holland put the economics under it without any moralizing. "AI has made the production of words incredibly cheap," he writes, "creating an inconvenient commercial problem for anyone whose margin depended on those words being expensive." His question is the useful one: what is content worth when the words themselves are cheap?

He also refuses the comfortable answer. There has always been human slop, he notes — the cheap article with no research, no experience and no point of view existed long before any of this. And "'I don't use AI' isn't automatically a badge of quality. It can also mean: 'I'm slower and more expensive.'"

Put those two pieces together and you have the position a small business is actually in. Nobody can tell who typed a sentence, and it no longer means much either way. What can still be told apart is whether a claim can be checked: a named street, a named job, a dated photograph, a number somebody can call.

Where being named actually comes from

This is where the week produced something practical. Also on Sept. 14, Search Engine Land published a write-up by the practitioner Zeeshan Yaseen of two experiments he ran back to back, tracking every result by hand — 775 citation events logged in all. Treat it as careful field notes on two brands in two niches, which is what he calls it, not as a law of nature.

The second experiment is the one a small business should read. A brand with no measurable presence in AI answers at the start of the month had 298 appearances across six assistants 30 days later. Of the mentions his tracking attributed to specific sources, listicles published by other people accounted for 85.8%; the brand's own page on its own site accounted for 14%. Three sources produced roughly 78% of everything. His own page was also the slowest thing he published to be picked up at all — 18 days, while some third-party placements were cited the day after they were published.

Two more of his findings are worth an owner's evening. First, mentions decay: in the earlier experiment about half the sources that had been citing the brand stopped within 30 days, with no change on the brand's part. Second, being listed next to recognized competitors helped rather than hurt. When he rewrote his own page to compare the brand against its leading rivals instead of presenting it alone, the mentions attributed to that page went from four to 49.

The lesson is not "go buy placements." It is that the sentence with your name in it is usually on somebody else's page, that it fades, and that the version of it which survives is the one that is useful to a reader rather than flattering to you.

The answer the platforms chose: provenance, not taste

Nobody is trying to judge what was machine-made by reading it. They are attaching proof to it instead.

Microsoft Advertising published guidance the same Monday, reported by Search Engine Land's Anu Adegbola, that tells advertisers to preserve watermarks, metadata and other provenance information showing how AI-generated content was created, and to place any required disclosure clearly and close to the content itself. Ads can still be rejected or removed for prohibited deepfakes, for impersonating a person or an organization, for missing disclosures — or for interfering with machine-readable provenance. As the piece puts it: slapping an "AI-generated" label on deceptive creative will not make it acceptable. Microsoft is not banning AI advertising; it is drawing a line between using AI to make an ad and using AI to mislead the person seeing it.

The same idea arrived in the camera five days earlier. TechCrunch's Tim De Chant reported on Sept. 9 that Apple is introducing Apple Reference Image: the sensor captures signed data alongside the picture, which becomes an unalterable version held in the Photos app and can be compared against any later copy to show what was changed. Apple calls it a digital negative, and says it will also support the SynthID standard for flagging images made or altered by AI.

Whatever you think of either company, the direction is one direction. Proof of where something came from is becoming a machine-readable property of the file — and the businesses that keep it will be the ones that can still say "this is really my work" in a year.

And the record has a price now

One bridge to the two previous columns, and then we are past it. On Wednesday this column said the conversation about your business now starts somewhere you do not own; on Thursday, that being answered is not the same as being finished. The third fact is that the material those answers are built from has started to be paid for.

Digiday's Jessica Davies and Sara Guaglione reported on Sept. 14 that Google has been quietly scaling a pilot that pays publishers when their content contributes to AI-generated answers across its products. Publishers who opt in get a widget in Search Console showing a monthly figure "with some history, though no other information on how payouts are calculated," Digiday writes. One executive with knowledge of it called the scheme "quite black box." Others in the pilot said they would rather be inside the tent than hoping the old referral economics come back.

You are not a publisher and nobody is going to pay you for your opening hours. But the shape of it matters: the text about businesses is now an input somebody buys, which means there will be more of it, faster, from sources with no idea whether it is true.

Three checks an owner can run this week

1. Open Collected info and read it as a stranger would. Business Profile dashboard, Edit profile, Collected info. Read every line, check the source and the date shown beside it, and delete what is outdated. Then look at the profile itself — hours, services, address, categories — because deleting a collected item does not change what is already published there. If your category or region does not have the tab yet, do the same job the long way: open your own listing in an incognito window and check every line against what is true today.

2. Decide who in your business is allowed to answer a machine about your facts. Google's documentation says it may reach the verified number on your profile by automated call, text or WhatsApp, and that what is said back may update the listing. Whoever picks up your phone at four on a Thursday afternoon is therefore editing your public record without knowing it. A voice on a telephone is not proof of who is calling — so the rule worth writing on the wall is simple: nobody changes a fact about this business over the phone; facts get changed in the dashboard, by the person who owns it, after logging in.

3. Put one checkable fact where somebody else can cite it. Not another page about how much you care. One specific, dateable, verifiable thing — the job you finished on that street, the equipment you actually carry, the certification with a number on it, the neighborhood you serve — placed where a reader who is not you would look for it: the local list, the trade directory, the supplier's contractor page, the association roster, the comparison a customer reads before calling. Then check again in a month, because mentions fade. And write it so it is useful next to your competitors, not instead of them.

Al Sefati, a marketing consultant writing in Entrepreneur on Sept. 14, arrives from the other side at the same place: the fastest way a specialized business burns money is generic marketing, because a specialized business does not need awareness. It needs proximity, visible proof of the work, and a place inside the networks people already trust. His examples are ordinary and testable — short neighborhood videos tagged where the job happened instead of another stock collage, a steady flow of real reviews, a profile filled out properly for the exact service people search for.

How we think about it

Every one of these stories is the same story told from a different desk: the words about your business are cheap to produce, and the proof behind them is not. A record is only as true as the thing that wrote it down, and for most small businesses that thing has been nobody in particular — a phone nobody answered, a page nobody updated, a conversation nobody kept.

That is the standard we build AI employees against, and it is the order we argue for: hire the AI employee first, then a person for what is left.

Ruby, our AI Customer Support, works the half of the record your customers write. She answers the people who already bought from you — in the direct messages of the channels you connected, Facebook Messenger, Instagram Direct, Telegram, and in the comments under your own posts, where half the "is this still available?" questions live. The first answer in a conversation tells the person they are talking to an AI assistant of your business; that notice is the platform's duty, not a switch you or we can turn off. She answers out of your own material: your profile, the line from your catalog as it is written, your rules for returns, exchanges and warranty. A rule you never wrote down does not get invented into existence — she says plainly that she does not know and hands the question to you.

Two things Ruby does not do are worth naming today. She does not confirm who is writing — a chat has no identity document, and a stranger's claim about themselves reaches you as a claim, not as a fact about somebody's order. And she does not decide your money: a refund, a compensation, a discount, an exception to your own rule are your word, so she collects the case and gives it to you rather than promising anything in your name.

What she leaves behind is the part that matters right now: every conversation stays written down in your account — who wrote, about what, how it ended, what is still owed. Out of that record you can see the thing that is otherwise invisible: the same question asked a hundred times is not a hundred customers, it is one hole in your own description of what you sell.

Piper, our AI Marketing Assistant, works the half of the record you write about yourself. She writes what your business says outside the feed — your pages, your letters, your offers, your ad copy, your item descriptions — in one voice, so that everyone speaking for you calls your things by your words. She does not invent a discount, a testimonial or a number: a price in the copy comes out of your catalog, or it is not in the copy at all. A promotion she writes carries a real end date, because an offer that never actually ends teaches your customers to ignore the next one. She promises you no reach, no followers and no sales, and she will not measure you against a named rival on figures neither of you can check. And she publishes nothing: she writes it, you approve it, and it reaches a channel through whoever owns that channel.

That is what a checkable fact looks like before anyone else can cite it — written out of your own data, dated, and signed off by you.

You do not learn a dashboard to run either of them: you describe your business in plain words, in the ChatGPT, Claude or Gemini app you already use. And before you hire anyone, you can watch our AI employees working in public on two demo storefronts, Tampa Pasta House and Casa Lista.

For twenty years the internet asked businesses to say more. Now it has started asking them to prove it. Open the tab, read what was written about you, and fix the lines that stopped being true.

Sources

1. Search Engine Land, Sept. 14, 2026, Barry Schwartz — “Google Business Profiles rolls out Collected Info”
2. Search Engine Land, Sept. 14, 2026, Zeeshan Yaseen — “Two GEO experiments challenge conventional AI visibility advice”
3. Search Engine Land, Sept. 14, 2026, Anu Adegbola — “Microsoft Advertising sets rules for AI-generated ads”
4. Search Engine Land, Sept. 14, 2026, Andrew Holland — “The future of content isn’t writing. It’s thinking.”
5. Ars Technica, Sept. 14, 2026, Dan Goodin — “AI bots ‘Timmy,’ ‘Ren,’ and ‘Jackie’ are flooding social media with slop”
6. Digiday, Sept. 14, 2026, Jessica Davies and Sara Guaglione — “Google rolls out pay-per-value AI licensing program to publishers”
7. Entrepreneur, Sept. 14, 2026, Al Sefati — “A Hyper-Local Marketing Playbook for Specialized Businesses”
8. TechCrunch, Sept. 9, 2026, Tim De Chant — “Apple has a new way to prove your iPhone photos aren’t AI slop”

Note on source 6: the live page carries the headline "pay-per-value" while our corpus copy of Sept. 14 carries "pay-per-use"; the address is unchanged and the live headline is the one printed above.

Note on source 2: a practitioner's own two experiments, tracked manually across two brands and two niches — presented in the piece as field notes, not as research.