TL;DR
JD Miller posted a joke on Instagram about the “restaurant” on his 110-square-foot Chicago balcony. Within a month, AI-driven sales tools had added him to their prospect lists and vendors were shipping him unsolicited samples — a case of vegan cookies, five pounds of kimchi, and 120 pounds of steak rub from a spice vendor in London. In this episode he uses that absurdity to explain how the algorithms actually work, why roughly half of the Cracker Barrel logo boycott came from bots, and gives operators a concrete crisis playbook: name your crisis team before you need it, build an advocate list, and never respond in the first two hours.
In this episode
Guest: JD Miller, PhD — go-to-market operating advisor, author of The CRO’s Guide to Winning in Private Equity and The AI Handbook for Sales Professionals, TEDx speaker. jdmillerphd.com
Host: Jeremy Julian, President & Chief Revenue Officer, CBS NorthStar
Also mentioned: Cracker Barrel, Instagram, LinkedIn, TikTok, Facebook, Yelp, Google Reviews, Restaurant Technologies, Yelp for Restaurants
Every so often a guest tells me a story so ridiculous that it takes a minute to realize it is actually the most important conversation I have had all month. This was one of those.
JD Miller has spent 25 years in tech as a chief marketing officer and chief sales officer, did his doctoral work on social networks back when the internet was brand new, sat on an Obama White House task force on how retail workers get scheduled, and has written two books — one of them on AI in the sales process. He also chairs the board of a Chicago homeless services organization that serves about 150 hot meals, and he keeps his food sanitation license current to cook there.
And he is the head chef of the Chicago Bell Garden, which is a very fancy name for his rooftop deck.
How did AI mistake a balcony for a restaurant?
The Chicago Bell Garden started as about 110 square feet of wooden balcony overlooking power lines and trash cans in an alley. A baby Q grill, four chairs squeezed around a table, some bells JD had collected from old jobs hanging off the roof, and tiny pots of herbs. He would have friends over for small plates.
When the vegan meat trend hit, he bought a package of vegan burgers, grilled one, and — because he makes burgers the way he makes burgers — buried it under grilled onions, cheese and bacon. As a joke he posted it: the Chicago Bell Garden is now proudly serving this brand’s vegan meat.
Three comments. Four likes. He has no large following. Two days later his phone rang, and it was a competing vegan meat producer asking whether they could get their burger onto his menu instead.
The next month my doorstep started to fill up with all of these unsolicited samples of restaurant products. There was a case of vegan cookies, there was five pounds of kimchi, and then 120 pounds of steak rub came from a spice vendor in London.
JD Miller, PhD
He is, by his own account, still working through the kimchi.
What happened is not mysterious once you know the machinery. AI sales-prospecting tools scrape the internet for signals that a new restaurant exists, package them into a lead list, and hand that list to a vendor. JD posted the phrase “now proudly serving vegan burgers” attached to something with the word “Garden” in it, at the exact moment vegan was the category everybody was chasing. The bot caught the signal and wrote him into a database.
The failure was not the bot. The failure was on the other end.
The salesperson kind of left their humanity at the door and just said, let’s let the bots run. You found 10 new restaurants in Chicago? Great, just go send the samples. That salesperson wasted whatever money to ship me 120 pounds of steak rub. I am never going to order any more steak rub.
JD Miller, PhD
Why does the algorithm spread things that are not true?
This is the part every operator should sit with. JD’s framing is the clearest I have heard.
Offline, information moves through gatekeepers who have skin in the game. When JD recommends a restaurant to me, he is spending credibility. If the recommendation is bad, I discount the next one. That self-correction is the whole mechanism that makes word of mouth trustworthy.
Online, the thing deciding what reaches you is not a person. It is an algorithm with no reputation at stake.
They look at every message and they don’t ask, is there truth in this. They just ask, when I put this on the page, does it make you slow the scroll? It’s optimizing for speed and velocity, while people are thinking about truth and veracity.
JD Miller, PhD
Slow the scroll, and there are more ads beside the post for a little longer. That is the entire objective function. Nothing about it is hostile to your restaurant specifically — it is just indifferent to whether the thing spreading about you is real.
What actually happened with the Cracker Barrel logo?
JD walked through it as the case study, and the number is the part that should stop you. Cracker Barrel launched a streamlined logo. Within days the platforms lit up with boycott calls, the news picked it up, and by his account the stock dropped 14% before the company reversed course and went back to the old mark.
Then the social media researchers went in.
The social media researchers discovered that actually about 50% of those boycott calls came from bots. It was just completely fabricated and made up. I can buy a bot to do that for a couple hundred dollars. But the algorithm that disseminated it didn’t really care about that.
JD Miller, PhD
A couple hundred dollars moved a public company’s stock price and reversed a brand decision. Sit with that for a second.
What is a bot farm and how cheap is it really?
I pushed on this because I have listeners who genuinely do not believe this stuff exists. JD broke it down plainly.
Most of us have a few email addresses — work, personal, one for the junk we do not want following us around. About a decade ago, some people realized they could do that at scale: farms of email addresses under one operator’s control, each spun into an Instagram or Facebook account. Sock puppets. Then a little code on top, and now AI on top of that, so a few hundred of them can be pointed at one post on command.
JD’s number: about nine dollars for a thousand YouTube views.
None of those accounts will ever walk into your restaurant or recommend it to anybody. But the vanity metric is the product, and here is the catch that matters — the algorithm partly weighs early velocity. Ten thousand views in the first twenty minutes reads as traction, and the platform pushes it further. Purchased engagement buys real distribution. That is why a manufactured boycott behaves exactly like an organic one.
Pat Venn and I got into the adjacent problem — how guests actually find you now, and why almost none of it is measurable — in The Recommendation Economy Comes to Restaurants — Pat Venn, PRIMA.
Can a 30-table restaurant really get caught in this?
Yes, and JD brought a recent example precisely because it was not a national chain. A small brunch spot in Atlanta — about 30 tables — had a guest come in, enjoy the meal, and ask the co-owner for a photo together to post. Which is exactly what any hospitality person would say yes to.
The guest turned out to be a politically controversial candidate for statewide office. A handful of people asked whether the photo sat at odds with the restaurant’s mission. The platforms registered people slowing down to look — and then, in JD’s words, the algorithms went wild.
That is the whole exposure in one story. She did nothing wrong. She did the hospitable thing. And the machinery that decides what spreads does not care about intent.
What should a restaurant post before anything goes wrong?
JD’s answer is not “go viral.” It is closer to insurance.
The solution to poison is dilution and lots of water — I think I learned that in my sanitation classes. If I have one misstep in a sea of, you know, I posted every day for two years, 600 posts, we’re in a better place already.
JD Miller, PhD
The frame he gave is one I am going to steal: treat your social presence as an extension of the dining room. The banter you would have with a guest at table six is the register. Genuine, authentic, not enormous.
Practically: pick one or two platforms rather than all of them. Make it somebody’s actual job. Batch it — JD sits down for half an hour, writes ten posts against ten photos, and hands them to an inexpensive scheduling tool to drip out over two weeks. And if you have people on staff who are native to this, give it to them as sidework instead of rolling silverware, as long as you trust their voice.
I added the caveat I always add: you can let guests into the living room without letting them into the basement. Talk about where the dish came from. That is almost always safe ground.
JD’s example of doing this well is a Chicago restaurant that built an entire menu around food waste — fritto misto made only of prawn heads, skate cheeks instead of the prime cut. Is the audience for that huge? No. But the people who care about it become a genuine online community that reinforces what the restaurant already is. Kyle Huck’s community-first approach in Why This Franchisee Gave Back $141K — Kyle Huck, Wings Etc. is the offline version of the same bet.
What is the playbook when it does go sideways?
I told JD that in my day job we keep a written playbook for a credit card breach — ten things you do, in order, because you do not want to be inventing the response during the event. He said online outrage deserves exactly the same treatment. Here is his sequence.
- Name the crisis team before there is a crisis. Probably the owner, the GM, and somebody in your network who does marketing, PR or media. Write down their phone numbers. Deciding who convenes is not something to work out at 11 p.m.
- Build an advocate list now. Five or ten regulars, long-tenured staff with good stories, friends in the neighborhood — with email addresses and phone numbers. These are the people you will need to hand a reporter.
- Press pause. Nothing requires a response in the next thirty seconds. JD’s minimum is two hours, ideally 24. A fast angry reply is engagement, and engagement is what the algorithm is rewarding. A lot of it evaporates on its own.
- Convene and plan the message during that cooling-off period, rather than reacting inside it.
- Designate a single point of contact. One person in front of the cameras, one voice online. Not whoever happens to be standing near the door.
- Make your primary response somewhere other than the platform. A trusted local news anchor, an email to your actual customer list, a mailer. Then reference that response online. The opinions you care about belong to people who physically walk into your store.
- Activate the advocates. Tell the reporter it is not just you — here are three customers, go call them.
- Hold on. These things do pass.
I added one we implemented in our own business recently and JD endorsed it hard: put in the employee handbook exactly who is permitted to speak to the media — and define media to include the TikToker and the YouTuber who walks in asking where you source your beef. JD’s point is that these things most often go sideways through somebody who is not the spokesperson, like a server hired yesterday who is simply wrong about the sourcing.
How should operators handle influencer pitches?
My listeners get these calls daily, and I have not seen a decent guide anywhere. JD’s is two questions.
First, are the followers real? Circle back to the bot conversation — half a million followers and half a million likes can all be purchased. If you are evaluating on follower count alone, buyer beware.
Second, and more useful: even if the audience is completely real, does it match yours?
If I’m that Chicago seafood place serving the prawn heads and you’re a TikToker who has 500,000 followers who don’t care about that — even if I pay you whatever the money is to come in and eat the risotto and say it was great, it’s not going to resonate authentically with your audience. It’s not going to translate over.
JD Miller, PhD
His test for the good version: would this person promote you anyway, if the money were not on the table? If yes, a comp or a fee is fine. If no, you are buying distribution to strangers. Brett Linkletter and I spent a whole episode on why so little restaurant marketing spend can be traced to a cover in Restaurant Marketing You Can Track — Brett Linkletter, Dishio.
Where does AI actually belong in this?
JD told one more story that landed for me. A small Chicago accounting firm was auto-posting on LinkedIn — software finding small businesses and writing the posts, untouched by a human. The posts had drifted into insulting. Somebody announced they had finally opened their yoga studio after five years, and the bot replied that nothing says procrastinator like waiting three years, and by the way you need accounting services.
JD picked up the phone and called the guy. The answer was, I bought this really cool technology and I don’t touch it at all.
You need to put a human back in the loop. Let it go find all the people, but every morning have it queue up for you — here’s the yoga studio, here’s what I’m saying — and you be the one to press send. Because the bot will drift. It probably started with something that was really cool and funny, but over time it adapted a personality that really isn’t representing you very well.
JD Miller, PhD
That is the whole thing. AI is genuinely good at automating tasks. It is not good at judgment, and the drift is invisible from the inside. I told JD I have three teenagers and a ten-year-old at home and every one of them can spot AI slop instantly and turns it off the moment they do. Your guests are no different. Jon Harmeling made a complementary case for AI as a leadership tool rather than a content machine in How a Chick-fil-A Operator Uses AI to Lead — Jon Harmeling.
Frequently asked questions
How long should a restaurant wait before responding to online outrage?
JD Miller recommends a minimum two-hour cooling-off period and ideally 24 hours. His reasoning is mechanical rather than emotional: platform algorithms reward engagement, so an immediate combative reply feeds the distribution of the thing you want to shrink. Much of it dissipates without any response at all, and the pause gives you time to convene your crisis team and plan a message.
Who should be on a restaurant’s crisis communications team?
Typically the owner, the general manager, and somebody in your network with marketing, PR or media experience — named and contactable before an incident happens. JD Miller also recommends designating one single spokesperson for both media and online response, and keeping a standing list of five to ten advocates (regulars, long-tenured staff, neighborhood friends) with their contact information.
Are social media boycotts real or driven by bots?
Both, and the distinction is hard to see in the moment. JD Miller cites the Cracker Barrel logo episode, where social media researchers later found roughly 50% of the boycott calls came from bots, while the stock still dropped 14% and the company reversed the decision. Purchased engagement is cheap — he cites about nine dollars for a thousand YouTube views — and early velocity causes real algorithmic amplification.
How do you evaluate an influencer who wants a free meal?
Ask two questions. Are the followers real, given that both follower counts and likes can be purchased from bot farms? And even if real, does that audience overlap with your guests? JD Miller’s test for genuine advocacy is whether the person would promote you anyway if the money were off the table.
Should restaurants automate social media posting with AI?
Scheduling tools are fine — JD Miller batches ten posts in half an hour and drips them over two weeks. Fully autonomous AI posting is not, because the output drifts away from your voice over time without anyone noticing. His rule is to keep a human in the loop who presses send, and to make sure the content would pass the smell test if you saw it in the real world.
Takeaways for operators
- Write the playbook before you need it. Same discipline as a credit card breach response. Names, phone numbers, sequence.
- Post steadily so one bad day is not your whole footprint. Dilution is the defense.
- The first two hours are the dangerous ones. Responding fast feeds the machine that is amplifying you.
- Move the real response off-platform. Your email list and local press reach the people who actually spend money with you.
- Put media policy in the handbook and define media broadly enough to include anyone with a phone and a following.
- Audience fit beats follower count every time, and follower counts can be bought for pocket change.
- Keep a human pressing send. Automation without judgment is how a brand voice drifts into something you would never say.
JD’s TEDx talk is Reclaiming Our Humanity in the Age of the Algorithm, and he is putting a downloadable crisis PR toolkit on his site. Both are at jdmillerphd.com.
Want the whole conversation? Read the full transcript of this episode with JD Miller.
Thanks to our sponsors
This episode is brought to you by Restaurant Technologies, whose closed-loop cooking oil management keeps kitchens safer, cleaner and more consistent, and by Yelp, helping restaurants get discovered by the guests already looking for them.
