How to Market Your Small Business When Personal AI Assistants Do the Shopping

The next new customer who calls your business may not be a person.

On September 16, 2026, Meta’s assistant, Muse, started placing phone calls to US businesses for the people who use it. Booking a table. Getting on a dentist’s cancellation list. Asking whether something is in stock.

Instinct, a startup assistant you reach by text message, turned on the same ability the same day.

Two weeks later OpenAI shipped Dots, and xAI’s Grok Bot had been signing into websites and finishing errands for its users since August.

I’ve read a lot of “this changes everything” announcements in twenty-plus years of doing this work, and most of them changed very little for a business with twelve employees.

I’m treating this one differently, for a reason that has nothing to do with the technology.

The person on the other end of these assistants is your customer, and a meaningful share of them have already decided to let software make the first contact.

That survey comes from a venture firm with money in AI companies, so read the exact figures loosely.

The direction is hard to argue with.

The companies building these call them personal AI agents.

I’m going to call them assistants, because that’s the word your customer uses, and because “agent” already means three other things in a business like yours.

This post works through the three questions I’m hearing, in order: how you market to a customer who sends software ahead of them, what happens to your website, and how these assistants actually work for your customer right now and where they’re headed.

Then it spends most of its weight on the part I’m surest about, which is the short list of things that win customers no matter what the front door looks like.

Everything above that list is my best read in October 2026. Some of it will be wrong by Spring. The list won’t be.

---

What a Personal AI Assistant Is, and Who Makes the Big Ones Right Now

A chatbot answers. An assistant does the thing, then tells you it’s done.

The useful definition for a business owner is this…

A personal AI assistant is software with its own computer and browser in the cloud, connected to a person’s email, calendar, messages, and often a payment card. The person gives it a goal in plain words, usually by text. It searches, reads, compares, fills out forms, places calls, and pays, checking back only when something needs a decision. It keeps working after the person closes the app.

That last part matters more than it sounds.

The errand you used to be part of, a customer scrolling your site at 9 p.m. deciding whether to call tomorrow, now happens without the customer watching.

The assistant reads your site, decides, and reports back a short verdict in a text thread. Your first impression is now a summary written by software.

Here are the four that matter as I write this.

Assistant	Who makes it	How a person reaches it	How it reaches a business	Who can use it
Muse	Meta	Muse app (iOS, Android, Mac, web) and directly inside WhatsApp	Browses websites, fills forms, pays with a one-time card through Stripe’s Link, and, in a US beta since September 16, places phone calls	Free for most uses, US and Canada; $20 and $100 monthly plans
Dots	OpenAI	ChatGPT (web, desktop, mobile), Slack, Microsoft Teams, voice calls; texting announced as coming	Browses websites and connects to more than 4,000 apps through plugins	Included with ChatGPT Pro, Business Premium, and Enterprise in eligible markets; rolling out since September 29
Grok Bot	xAI	Grok Bot app on desktop and iOS, by message	Signs into websites and apps, including ones with no API, and runs as a team of named “Bots” sharing one cloud computer	Bundled with SuperGrok and paid Cursor plans; in beta since August 11; no free tier beyond a trial
Instinct	Spear Street Technology (founder Noah Shinn)	iMessage, WhatsApp, or email; no app	Browses and signs into websites; Instinct Concierge places phone calls, in early access since September 16	Invite-only; no published pricing

A few things in that table are worth slowing down on.

Three of the four reach the person through text.

Muse has desktop and mobile apps and also lives inside WhatsApp, which Meta says has more than 3 billion monthly users and over 100 million in the United States.

Instinct has no app at all; you text it via Messages or WhatsApp.

Dots is accessed through ChatGPT’s desktop and mobile apps.

So, in most cases, when your customer’s assistant comes back with a recommendation, it arrives as a message, in the same thread as their family group chat.

There is no page for you to design there. There is only what the assistant says about you.

Two of the four now pick up the phone.

That closes the gap that protected a lot of small businesses from the first wave of AI search: if you only took bookings by phone, software couldn’t book you.

As of September 2026 it can call you, ask for Friday at seven, and confirm. One analysis of the early beta reported that when a call goes unanswered, Muse moves to the next business rather than leaving a voicemail. I can’t confirm that from Meta’s own documentation, so treat it as a plausible report rather than a fact. It matches how I’d build it.

And the free one is the one owned by the company that already has your customer’s attention.

Muse displaced ChatGPT as the number one free iPhone app in the United States ten days after launch, and Apptopia estimated 2.8 million US and Canadian downloads in its first twelve days.

Those are outside estimates, not Meta’s numbers. They’re big enough to plan around.

I’ll say the obvious thing once so I don’t have to keep saying it.

This table will be wrong within months.

Products will merge, rename, die, and new ones will pass them. The post carries a “Last verified” line at the bottom and I’ll keep the table current.

If you’re reading this and the names look dated, scroll down before you decide I’m behind.

---

How an Assistant Picks a Business

Here’s the mental model I want you to carry out of this post.

Skift, the travel trade publication, tested Muse the day after it launched.

It asked for a New York hotel for two nights in a price range.

Muse opened a browser and went to Expedia and Hotels.com, the way a person might.

It found a first candidate above budget, kept looking, and compared properties on price, location, subway access, review scores, refundability, bed type, and fees.

It settled on a hotel at $410 a night and flagged that the cheap rate came with twin beds instead of a queen. It never opened either hotel’s own website.

Read that again from the hotel’s side.

The hotel wasn’t outbid. It was outread.

The booking sites had every answer on one page, and the assistant went where the answers were.

Three weeks later Skift and a BTIG analyst ran Muse against the big hotel chains directly.

Muse hit “Something went wrong” at the room-and-rate step on Hilton, Marriott, and Hyatt sites.

A rival assistant, Instinct, finished the same searches on all three and came back with member rates that matched a manual search.

Hilton said it welcomes bookings from assistants and hasn’t blocked any.

I believe them. The plumbing just wasn’t there yet. The assistant didn’t argue about it or try harder. It stopped.

That’s the whole pattern, and it generalizes well past hotels:

• It reads what’s public and cross-checks it. Your website, your Google Business Profile, your reviews, the directories and booking platforms it already knows. If your hours differ between two of those, it doesn’t call to ask which is right.
• It prefers the source with every answer in one place. Price including fees, availability, cancellation terms, what’s included. A page that makes it click four times loses to a page that doesn’t.
• It stops where the path breaks. An unlabeled form field, a “call us for pricing” line, a booking widget that errors, a bot wall that treats it like a scraper. It doesn’t work around the gap the way a person would.
• It moves on when nobody answers. A ringing phone, a form that replies in two days, a chat widget that says “we’ll get back to you.”
• It reports back in a sentence. “Booked you at Washington Square for $410, note the twin beds.” Whatever it noticed about you is what your customer hears.

Software is less forgiving than a person.

A person who can’t find your hours will usually work around you.

Software doesn’t.

That’s bad news for the business with a small gap on its site, and good news for the business that has already closed it, because the assistant will reward that reliability every single time, without being charmed by a competitor’s better photos.

---

When the Big Sites Lock the Door: What Amazon Blocking Muse Means for You

There’s a second way an assistant stops, and it has nothing to do with a broken form. The site tells it no.

On the night of September 20, twelve days after Muse launched, people who sent it to shop on Amazon got a pop-up instead of a cart: “Continued access by an unauthorized AI agent violates Amazon’s Conditions of Use, to which our customers have agreed.”

Amazon had asked Meta to take Amazon out of Muse’s shopping experience. Meta declined. Amazon put up a wall that stops Muse before it can load a search page.

Amazon’s stated reasons, given to GeekWire and Adweek: Meta never told it Muse would shop its store, the assistant doesn’t identify itself as software while it browses, and it appears to capture and store customer credentials.

Meta disputes the credential claim.

Amazon also said something quieter that I read as the real reason.

An assistant shopping for a customer bypasses Amazon’s personalization, its recommendations, and the discovery its store is built to do.

The customer who sends Muse never sees Amazon’s store. Amazon has an assistant of its own for that job.

None of this is new behavior.

Amazon sued Perplexity in November 2025 over its Comet browser doing the same thing, won a preliminary injunction in March 2026, and lost it in August when the Ninth Circuit ruled that the person, not the AI company, was the one accessing Amazon under the federal anti-hacking law.

So Amazon fell back on its terms of use and a bot wall.

Its robots.txt, the file that tells automated visitors what they may read, already carries a hundred named blocks, including Google’s shopping agent, OpenAI’s crawlers, and three of Meta’s.

Muse doesn’t announce a name when it browses, so there was nothing to add to the file. Hence the wall.

Expect more of this, and expect it from the sites that can afford it.

A destination site is one your customer would have gone to anyway: Amazon, the big travel sites, the dominant marketplace in your category.

Those sites have a lot to lose when an assistant shops on the customer’s behalf (the relationship, the data, the next sale) and little to lose from saying no, because the customer comes back the old way.

Some will block. Some will cut deals, which is why Meta’s connector rules already talk about labeling “commercial relationships that influence recommendations.”

Some will build their own assistant and tell your customer’s assistant to go talk to it.

For a business your size, this cuts two ways, and I want you to hold both.

If you sell through a platform that blocks assistants, your own site may become the only door the assistant can use. Go back to the Skift test. Expedia let Muse in, the hotel chains’ own sites threw errors, and the booking went to Expedia. Now run it the other direction. If the marketplace you sell on locks Muse out, and your products are also on your own store with a checkout an assistant can finish, you’re the one with the open door. That’s a reason to keep your own storefront alive even when the marketplace brings most of your volume.

You’re not Amazon, so don’t act like it. Amazon can turn away an assistant because its customers will come anyway. Yours won’t. If your firewall, your bot filter, or the security plugin your developer installed two years ago treats an assistant like a scraper, you’ve locked out a real customer with real money and no reason to try you again. Blocking is a strategy for a business with leverage. For everyone else it’s a closed sign on the door.

The thing to watch over the next year is whether the big platforms end up deciding which assistants reach which businesses.

If this standoff ends in deals, the assistant your customer uses may only be able to shop the businesses whose platforms signed one.

That’s the version of this future where a small business loses without doing anything wrong, and it’s the strongest argument I know for the fifth item in the list at the end of this post: own the relationship, so your customer names you in the instruction and the assistant comes straight to your door instead of through anyone’s gate.

---

Question 1: How Do You Market to a Customer Through Their Assistant?

You don’t market to the assistant. You qualify for it.

Every tactic you’ve been taught for reaching a person, a sharper headline, a better offer, retargeting, urgency, works on a person.

The assistant doesn’t read your headline. It reads your hours.

There’s nothing to entice. There’s a checklist to pass, and the checklist is written by your customer’s request: “Find me a family law attorney near me who takes new clients, offers a free consultation, and can see me this week.”

The assistant turns that sentence into facts to confirm and a task to finish. You either confirm and finish, or you’re not in the reply.

So the marketing question reorganizes itself around three places you can still show up.

The Facts the Assistant Reads

This is where most of the work is, and almost none of it is new.

The assistant builds its shortlist from the same public record your customer would check if they had an afternoon: your site, your Google Business Profile, your reviews across the sites it trusts, and the platforms it already knows how to book through (OpenTable for restaurants, for one; Muse connected to OpenTable at launch ).

What’s new is the standard.

A person forgives a mismatch.

The assistant treats a mismatch as uncertainty and uncertainty as a reason to pick someone else.

Your name, address, phone, hours, services, and service area need to be identical everywhere they appear.

Your pricing, or at least your pricing range, needs to be public. “Contact us for a quote” is a dead end to software.

If you genuinely can’t publish a price, publish the thing that lets the assistant qualify you anyway: “Most kitchen remodels we do run $45,000 to $90,000; we’ll give you a fixed quote after a one-hour visit.”

Reviews carry more weight here than they ever did with a person, because the assistant can’t walk in and get a feel for the place.

BrightLocal sells review software, so it benefits if you believe reviews matter; the numbers still line up with what I see in client accounts.

The assistant reads the same reviews your customer would, then summarizes them in one line.

A six-month gap in your reviews reads to it the way a dark storefront reads to a person walking by.

If this sounds like what we’ve told you about showing up on Google and getting more reviews, that’s because it is.

We wrote the long versions in How to Show Up on Google and How to Get More Reviews, and the AI-specific layer is in AI Search Optimization for Small Businesses.

The assistant didn’t invent a new channel.

It raised the bar on the one you already had.

The Finish the Assistant Needs

The second place you show up is the moment the assistant tries to complete the task.

Book the table. Request the consultation. Buy the thing. Get on the list.

This is the part that decides whether you’re in the reply or the runner-up.

A beautifully written site with a phone number and a contact form that says “we’ll be in touch” is a dead end to an assistant whose instructions were “book it.”

If your front door is a form, label every field in plain words, confirm instantly on screen and by email, and let the person pick a time if a time is what they need.

If your front door is the phone, answer it, because two of the four assistants now dial.

If you sell online, make sure the cart, taxes, shipping, and return terms are all visible before checkout, because the assistant will compare that total against someone else’s.

One more thing I want to be direct about.

Some businesses have added bot-blocking tools over the past two years to stop scrapers.

Check what yours blocks.

If it treats your customer’s assistant like a scraper, you’ve done on accident what Amazon did on purpose, without Amazon’s reasons for doing it.

The hotel chains didn’t mean to lock anyone out. It happened anyway.

The Message the Assistant Hands Back

The third place is the one nobody’s designing for yet.

When the assistant finishes, it sends your customer a message.

“Booked you Thursday at 2 with Dr. Patel, $95 new-patient exam, they texted a confirmation.”

That message is now your first impression.

You don’t control the sentence. You control the facts in it, and you control what you send alongside it.

A confirmation text that lands on the customer’s phone within a minute of the booking, with your name, the time, the address, and one useful line (“park in the back lot, it’s free”), does two things.

It proves to the customer that you’re real and responsive, and it puts your voice in their hand right next to the assistant’s.

The business that’s silent after a booking exists only in the assistant’s summary.

The business that confirms has started a relationship.

What About Connectors, Protocols, and Paid Placement?

You’ll see articles telling you to build for the new protocols.

Here’s my read.

In October 2026, Meta and Sierra announced the Personal Agent Protocol, an open standard for how assistants authenticate with a business and what the business lets them do, with Instinct, Shopify, Stripe, Walmart, and about forty other design partners including OpenAI.

The draft spec was published October 9. Google launched its own Universal Commerce Protocol in January, and Shopify turned on agent checkout for eligible US stores in March.

Meta accepts “connector” applications so a business’s own systems can plug into Muse directly.

That’s a lot of plumbing for a business your size to follow, and you mostly shouldn’t.

One group checked 1,065 live Shopify stores in October to see what they declare to assistants. 1,051 of them declared exactly the same thing, because Shopify writes it for every store.

The researchers’ honest reading was that agent payments on Shopify are, today, a Shopify decision and not a merchant decision.

The same will be true of your booking software, your CRM, and your point-of-sale.

The platforms you already pay for will do the protocol work. Your job is to make sure the data they publish on your behalf is complete and correct, and to say yes when a connector is free and takes an afternoon.

On paid placement: it isn’t here yet for small businesses, and it’s coming.

Muse is free, and Meta’s own connector rules require that “advertising, sponsored results, and commercial relationships that influence recommendations” be labeled.

You don’t write a rule like that unless you expect the thing to exist. One analyst put it plainly to The Verge: if you’re not paying for the assistant, the goods and services you’re seeking may be cutting deals to get access to it.

When that market opens, it’ll be a line item like any other ad channel, and you’ll evaluate it the way you evaluate Google Ads.

It won’t replace the checklist above.

A sponsored slot gets you considered. Verifiable facts and a finishable booking get you picked.

---

Question 2: What Happens to Your Website?

Your website gets more important and less visited by people, at the same time.

Less visited by people because the assistant does the browsing.

In the Skift test, two hotels got evaluated and one got booked, and neither saw a human on its site.

If a share of your customers starts sending software ahead of them, your analytics will show fewer sessions from those customers and, if you’re set up well, the same number of bookings.

That’s going to confuse a lot of monthly reports.

It shouldn’t confuse you.

More important because the website is the one place you fully control what the assistant reads and whether it can finish.

Your Google Business Profile is rented.

Your reviews are earned but hosted elsewhere.

Your listing on a booking platform is on their terms.

Your site is yours, and for the assistant it does two jobs.

It’s the fact sheet. Everything the assistant needs to confirm, on one page or one click from it, in plain words, consistent with everywhere else you appear. Services with prices or ranges. Hours. Service area with the actual towns named. Who you work with and who you don’t. The answers to the twenty questions your front desk gets every week. If the assistant can confirm you from your own site, it doesn’t need to trust a third party’s summary of you.

It’s the finish line. The place the task completes: the booking, the quote request with a real appointment time, the checkout with the real total. Walmart learned this the expensive way. It put about 200,000 products into ChatGPT’s in-chat checkout, and purchases completed inside the chat converted at roughly a third of the rate of purchases that clicked out to Walmart’s own site. The company’s product executive called the experience “unsatisfying” and moved checkout back to Walmart’s own system, putting its own assistant inside ChatGPT instead. If the largest retailer in the country found that owning the finish converted three times better, you should assume the same, and make your finish worth owning.

So here’s what to do now, and none of it requires a redesign:

1. Run the one-page test. Pick your most common request. Count the clicks from your homepage to a complete answer: what it costs (with fees), when you’re available, what’s included, how to book. If it’s more than two, fix the page, not the assistant.
2. Publish the things you’ve been keeping for the phone call. Price ranges, what you don’t do, whether you’re taking new clients, lead times. Every one of those is a question an assistant will ask and move on from if it can’t answer.
3. Label your forms like a person who’s never seen your business will read them. “Preferred appointment date” beats “Date.” Confirm on screen and by text or email within a minute. If the form is really a booking, make it one.
4. Check your blockers. Ask whoever manages your site what your firewall and bot tools block, and test it by sending an assistant to book you (the six errands at the end of this post do exactly this).
5. Keep the human version excellent. The assistant sends the person a summary, and a share of those people will tap through to look. The site still has to sell to a human in ten seconds. Nothing in this post changes that.

We’ve written the technical version of this twice: the AI-Ready Website Playbook covers the four layers in order, and The Agentic Web covers the longer bet about where all of this converges.

For the conversion side, our Website Conversion Leak Analysis will find the places a human gives up, and they’re usually the same places an assistant does.

One thing I don’t want you to do: don’t build a separate “AI version” of your site. The assistant reads the same page the person does. One site, every answer on it, one finish that works. That’s the brief.

---

Question 3: How Do These Assistants Actually Work for Your Customer, and Where Are They Headed?

You’ll make better decisions about this if you’ve felt it from the customer’s side, so I’d rather you spent thirty minutes with one than took my description.

Muse is free in the US and Canada.

If you’d rather not create a Meta account, borrow a friend’s Instinct or ask someone on ChatGPT Pro to lend you their Dot for an hour.

The prompts at the end of this post are written for exactly that session: the first six get the assistant to explain how it decides, and the last six send it to your own business.

Here’s what you’ll notice.

It asks before it acts, and the customer decides how often. Muse splits every action into reading and writing, and the customer chooses whether it asks once, for this task, for this site, or never. Instinct guarantees a confirmation before it spends money. Dots works toward goals in the background and messages the person with progress and questions. The practical effect for you: the customer sees a short summary and taps Allow. They aren’t reading your page. They’re reading the assistant’s description of your page.

It pays with a card you’ve never seen. Muse checks out through Stripe’s Link, which issues a one-time-use card so the customer’s real card never touches your site, and Link’s purchase protections cover the transaction. Shop Pay is coming. If your checkout rejects virtual cards, or your front desk refuses to take a card number over the phone from a caller that sounds a little flat, you’ve just declined a real customer. Decide your policy now, while it’s a small number of calls.

It lives in text. WhatsApp, iMessage, SMS. Meta’s own commissioned research found 73.3% of consumers across 22 markets prefer messaging when communicating with a business, and Meta sells business messaging, so weigh that figure accordingly. Still, the assistant’s native habitat is a message thread, and your confirmations, follow-ups, and reminders should be able to land there too. You don’t need to be on WhatsApp to serve a WhatsApp customer. You need to be reachable and responsive on whatever channel the assistant uses to reach you, and to confirm by text when a person gives you their number.

It’s getting an email address and a voice. Meta announced that each Muse will get its own email address to accomplish tasks, and both Muse and Instinct now call. So your inbox and your phone line will start receiving contacts from software acting for a real person. The right posture is the one a good front desk already has: treat it as a legitimate request, answer the question, confirm in writing, and never take credentials or payment details by voice. Train your team to recognize it and not to hang up.

Where It’s Headed

Here is where I’ll speculate, and I’ll label it as such.

Your business will get an assistant of its own, and the two will talk. This is the part of the Personal Agent Protocol that I think matters most for you. A business can choose to work with personal assistants through its website, through its software, or through an assistant of its own. At the launch, Rocket demonstrated Muse getting a mortgage pre-approval by finding Rocket’s assistant and letting the two work out the numbers while the customer approved each step. Meta told investors in its second-quarter 2026 earnings call that more than a million businesses a week are already using its business assistants on WhatsApp and Messenger. Within a couple of years I expect the first contact for a lot of your new customers to be their assistant asking your assistant whether you can do Thursday. That’s good news for a small business: a front-desk assistant that answers in two seconds, around the clock, is a thing you can rent for less than one part-time hire, and it’s exactly the kind of system we build.

Teams of assistants, not one. OpenAI says it envisions teams of Dots working together; xAI launched Grok Bot as a team from day one; Instinct lets family members’ assistants coordinate. The customer who sends one assistant to find a contractor may soon send one to find three, one to check licenses and reviews, and one to schedule the walkthroughs.

New surfaces. Meta has said Muse is headed for its glasses. Voice is already here. None of that changes the checklist. A customer asking their glasses to “find me a dentist who can see me today” produces the same facts to confirm and the same task to finish.

And what could make this whole post irrelevant. Several things, and you should know them. Trust could collapse after a wave of bad purchases; the economics could change when the free tier isn’t free; regulators could slow the calling and paying features; the destination sites could win the standoff Amazon started and leave assistants with only the businesses whose platforms signed deals; or people could decide, as they have before, that they’d rather do their own shopping. Right now the actual volume is small.

A Checkout.com survey of merchants in the UK and US found only 3% of their transactions involved an AI assistant, even though 89% said they were preparing for it. Checkout.com sells payments, so it benefits from the preparing number. The 3% is the number to hold onto. Overall US adoption of AI barely moved this year, from 61% to 64% of adults. The depth is growing. The breadth isn’t, yet.

If every assistant named in this post is gone in two years, the six things in the next section will still be the things that bring you customers. That’s why they get the most room.

---

Where We Stand

I’d rather you argued with a clear position than nodded at a vague one, so here’s ours.

1. The assistant is a hired skeptic. Treat it like one. You win it with facts it can verify, a reply it doesn’t wait for, and a task it can finish. Persuasion doesn’t reach it.
2. The website becomes the fact sheet and the finish line. Its traffic will fall and its importance will rise. Judge it by completed bookings, not sessions. And keep the door open: blocking assistants is a move for a site with Amazon’s leverage, and when the marketplaces block, your own site is the door that’s left.
3. Don’t build for protocols. Feed the platforms you already pay for. Say yes to free connectors. Spend your engineering money on data quality and a booking path that works.
4. Keep the human relationship. The assistant runs errands. The person still decides who they trust with anything that matters, and still tells their friends. Your owned list, your confirmations, your follow-up, your name in their phone: those survive every interface change.
5. Assume it changes, and build only what pays off either way. Everything in the next section pays off whether assistants take over or disappear. Nothing in this post asks you to bet on one company.

---

The Six Things That Always Work, and Work Better Now

Every one of these was true when your customers found you in the phone book.

Every one is true when they find you on Google.

Every one is rewarded harder by an assistant, because the assistant removes the forgiveness a person extends.

Each comes with one thing to do this week.

1. Be Easy to Verify

Why it always worked. Trust starts with the small facts lining up. A business whose hours on the door match the hours on the website feels run by someone paying attention.

Why the assistant rewards it harder. It cross-checks you across sources and treats a mismatch as a reason to pick someone else. It can’t call to ask which is right.

This week. Open your website, your Google Business Profile, and the top two directories for your industry side by side. Make name, address, phone, hours, services, and service area identical, word for word. Publish a price or a price range for your three most common services, even if the final number comes after a visit.

2. Answer in Minutes, on Every Channel

Why it always worked. The lead that gets a call back in five minutes closes. The one that waits a day calls someone else. A 2011 study of lead response, still the one everyone cites, found a lead contacted within five minutes was far more likely to qualify than one contacted after thirty. That study is old; name the year when you quote it. The behavior behind it hasn’t aged.

Why the assistant rewards it harder. It doesn’t wait. A ringing phone or a “we’ll get back to you” form is a closed door, and the next business is one tab away.

This week. Time your own response. Submit your form, call your main line at lunch, send a message on Instagram. If any of them takes more than fifteen minutes to get a real reply, fix that one path. Our guide to marketing automation for small businesses covers how to do it without hiring anyone.

3. Make the Task Finishable Without You

Why it always worked. Customers who could book, buy, or get a firm quote without a phone tag booked more often. Friction has always cost sales.

Why the assistant rewards it harder. “Finish the task” is the assistant’s entire instruction. If it can’t, you weren’t a candidate.

This week. Pick your single most common request and make it completable end to end on your site: a real time slot, a real total, a real confirmation. If you can’t publish availability, publish a request form that confirms a callback within a stated window and then keep the window.

4. Earn Fresh Reviews, Continuously

Why it always worked. People trust people like them. A steady stream of recent reviews is the closest thing to word of mouth you can show a stranger.

Why the assistant rewards it harder. It reads reviews as data and summarizes them in a sentence. Volume, recency, and rating all show up in that sentence. A six-month gap reads as a closed sign.

This week. Ask every customer from the last thirty days, by text, with a direct link. Set up the ask to happen automatically after every completed job from now on. Reply to the last ten reviews, including the bad ones, in your own voice.

5. Own the Relationship

Why it always worked. Repeat customers and referrals are cheaper than any new lead, and nobody can take your list away.

Why the assistant rewards it harder. When a customer tells their assistant “book me at Marlowe’s again,” the assistant goes to Marlowe’s. The strongest marketing in this new world is being the name in the instruction. You earn that spot with the experience and the follow-up, and you keep it with permission-based contact the customer welcomes: a confirmation, a reminder, a useful note, an occasional email they open.

This week. Send a confirmation text within one minute of every booking from now on, with one useful line in it. Then pick one past-customer message to send this month, with something in it worth reading. If you don’t have a list, start it; our email marketing guide for small businesses walks through the first ninety days.

6. Measure Four Numbers and Review Them Monthly

Why it always worked. A plan you don’t check drifts. The businesses that keep winning know which four numbers moved last month.

Why the assistant rewards it harder. Sessions will fall as software does the browsing. If you’re watching traffic, you’ll panic at exactly the wrong moment. Watch completed bookings, response time, new reviews, and revenue from returning customers instead.

This week. Put those four on one page and look at them on the first Monday of each month. If you built the scoreboard from our small business marketing plan guide, it already has three of them. Add response time.

Notice what’s not on this list.

Nothing about protocols. Nothing about a specific assistant. Nothing you’d have to tear out if Meta cancelled Muse next year.

That’s the point.

These are the things I’d tell you to do if I knew nothing about AI, and they’re the things that happen to decide whether an assistant picks you.

---

Prompts to Copy: Learn How Your Customer’s Assistant Thinks

I’ll say this plainly.

If you own a business, you should be using one of these assistants yourself, every week, for your own errands.

Not to seem current.

Because you can’t plan around something you’ve never watched work, and the only reliable way to understand how it reaches customers is to be a customer who sends it.

A free one takes five minutes to set up.

Any assistant that can browse the web and fill out a form will do for everything below, and none of these prompts name one.

Paste these in one at a time, with your own details in the brackets. Each one is built to show you something specific about how the assistant decides, in its own words. Read the answers for what it checked, what it trusted, and where it hesitated.

How it would shop for a business like yours.

I'm thinking about hiring a [type of business] in [city or area]. Before you search, walk me through exactly how you'd find one and pick one for me: which sources you'd check and in what order, which facts you'd confirm, and what would make you drop a business from the list.

What you’ll learn: the assistant’s actual checklist, before it has ever seen your site. Watch how much weight it puts on your facts matching between sources; it’s usually more than you’d guess.

What it sees when it looks at you.

Visit [your website URL] and tell me in three sentences what this business does, who it's for, and what it costs. Then list every question a customer would need answered that you couldn't answer from the site.

What you’ll learn: the summary your customer would receive about you, and the gaps you can’t see from inside the business.

What it trusts.

When you check reviews for a local business, which sites do you look at, how recent do reviews need to be before you rely on them, and how would you describe a business to me if its reviews were excellent but all more than a year old?

What you’ll learn: how your review recency reads to software, in a sentence you can repeat to your team.

How it would reach you.

If I asked you to contact [your business name] in [city] for me, which methods could you use (website form, online booking, phone call, email, messaging), which would you try first and why, and what would make you give up and move to another business?

What you’ll learn: which of your front doors the assistant prefers, and the exact behavior that makes it leave.

What it would tell your customer.

Pretend you just finished booking [your most common service] at [your business name] in [city] for me. Write the confirmation message you'd send me, using only facts about the business you can verify online right now.

What you’ll learn: the one-paragraph verdict your customer reads instead of your website, and whether your facts are even available to fill it in.

How it actually works.

Explain how you decide when to ask my permission before taking an action, how you pay for things on my behalf, what you do when a website blocks automated visitors, and what you can't do yet that you expect to be able to do soon.

What you’ll learn: permissions, payment cards, blocking, and the roadmap, from the assistant itself in plain language. Ask it this one again in three months. The answer will have changed, and the change is the news.

---

Then Send It to Your Own Business: The Six Errands

Understanding how it thinks is half the job.

The other half is finding the exact place it stops at your business, and the fastest way to do that is to send it.

Subscribers to my Monday Marketing Memo will remember I asked them to open an assistant and ask it to book them.

This is that idea turned into a repeatable test with a score, so you can run it today, fix two things, and run it again next quarter.

You’ll need the same assistant and thirty minutes.

Don’t tell it your business name in the first errand. Watch its screen if the assistant lets you, and note the exact place it hesitates or stops.

Run these six errands, pasting your own details into the brackets.

Errand 1: Find you cold.

Find me a [type of business] in [your city or area] that offers [your most common service] and is taking new customers. Give me your top three with a one-line reason for each.

Score 4 if you’re in the three with an accurate reason. 2 if you appear with a wrong or vague reason. 0 if you don’t appear.

Errand 2: Confirm the facts.

For [your business name] in [city], tell me their hours, address, phone number, whether they're open on Saturdays, and what [your most common service] costs or roughly costs.

Score 4 if every answer is right. 2 if one is wrong or “not listed.” 0 if two or more are wrong or missing.

Errand 3: Check your reputation.

Summarize what customers say about [your business name] in [city] in two sentences, and tell me how recent the reviews are.

Score 4 if the summary is accurate and mentions reviews from the last three months. 2 if accurate but dated. 0 if thin, wrong, or “not enough reviews.”

Errand 4: Book or request.

Book me [your most common service] at [your business name] for [a day next week], any time after 3 p.m. If they don't take online bookings, request an appointment through their website and tell me what happens.

Score 4 if it completes a booking or a request with a confirmed time. 2 if it submits a request but gets no time or no confirmation. 0 if it stops (form error, blocked, “call us”).

Errand 5: Call you (only if your assistant places calls, and warn your front desk first).

Call [your business name] at [phone] and ask whether they can do [service] on [day]. Book it if they can.

Score 4 if answered within a few rings and booked or clearly answered. 2 if answered but it couldn’t finish. 0 if unanswered or hung up on.

Errand 6: Compare you.

Compare [your business name] with two other [type of business] in [city] on price, availability this week, and reviews. Which would you pick for me and why?

Score 4 if you’re picked for accurate reasons. 2 if you’re picked for wrong reasons or lose on something you can fix this month. 0 if you lose on a fact that’s wrong or missing.

Your score, out of 24:

• 20 to 24. You’re already the business the assistant picks. Protect it: run the test again every quarter.
• 12 to 19. You’re findable but something stops the finish. The errand you scored lowest on is this month’s project.
• 0 to 11. The assistant can’t find or can’t finish with you. Start with Errands 2 and 4; the fixes are usually a page and a form, not a rebuild.

Write down the exact sentence the assistant said when it stopped.

That sentence is your to-do list.

Put the date and the score in a note somewhere you’ll find it, and run the same six errands next quarter.

The score should go up. If it doesn’t, the assistant changed, and the sentence will tell you how.

---

Who Does the Work: You, Your Team, or a Partner

Everything in this post can be done by you or someone on your team in a few afternoons, and I’d rather you ran the prompts and the six errands yourself than paid anyone to run them, because the stopping point will teach you more than a report would.

Where it makes sense to bring us in is the system behind the six things: a website that answers everything on one page and finishes the booking, a response setup that replies in minutes on every channel, the review and follow-up automation, and, increasingly, a front-desk assistant of your own that can answer your customers’ assistants around the clock.

That’s the work our AI marketing systems are built for.

We can set it up and hand it over, run it with you, or give you an hour of guidance and let you keep the controls.

If you’re weighing that choice, we wrote an honest version of it in DIY marketing with AI vs. hiring an agency.

If you’d like a second set of eyes on your results, book thirty minutes with me and bring the sentence where the assistant stopped.

---

Keeping This Post Current

This is a timely post about a market that’s a few months old, so I’m treating it as a living document.

The assistant table, the stats, and the protocol notes are checked monthly for the first six months and quarterly after that.

Changes get logged in the “What Changed” section at the bottom with a date.

If a new assistant passes the four named here, it goes in the table and the old one gets a line in the changelog. If something in this post is wrong, I’d rather hear it than have you find out on a booking: reply here.

Last verified: October 14, 2026.

---

Frequently Asked Questions About Marketing to Personal AI Assistants

What Changed in This Guide

• 2026-10-10: First version. Assistant table current as of this date (Muse, Dots, Grok Bot, Instinct). Amazon’s block of Muse (September 20) and the Perplexity ruling noted. Personal Agent Protocol draft 0.1 noted. Prompt set v1: six learning prompts, six errands.

---

Your Customer Hired a Skeptic

I’ll end where I started.

The next new customer who contacts your business may be software acting for a person, and the software doesn’t care how good your headline is.

It cares whether your hours match, whether someone answers, whether it can finish, and what your last twenty customers said.

Those were always the things that mattered.

The assistant just stopped forgiving the gaps.

Open one this week and use it for your own errands until you understand how it decides.

Then send it to your own business, write down the sentence where it stops, and fix that.

Then do the six things above the way you’d do them if none of this had ever happened, because that’s the version of this post that will still be true in five years.

Ready to go further?

We build the systems behind all six, including a front-desk assistant that can answer your customers’ assistants at two in the morning. Book thirty minutes with me and bring your score.

---