Hello Lunch Break. Who’s made it to the new Target? I used to love a good Target run. I might be the outlier here but I stopped going to Target years ago. Maybe it’s just the season of life I’m in but I’d much rather just have Amazon drop it at my door. Now, what will I go out for?…A brunch. Heading to Casa Blanca today. Toasted egg wrap. And the goat cheese for the win.
Funny thing. When I asked ChatGPT this morning where to go, two different prompt styles gave me two completely different answers. Which, it turns out, is the whole story today.
Alright. Clock's ticking. Back to work.
THE GAME CHANGED
Thinking mode reads a different web.
I ran a small experiment this morning. Two tabs.
Tab one: "Best brunch in Wilmington, NC."
Tab two: "I need a brunch spot in Wilmington, NC for a group of six that includes a vegan, a toddler, and grandparents who hate loud places, with a wait time under 20 minutes on a Saturday. Compare the top three."
Tab one came back fast with a six-spot starter pack. Tab two paused, and came back with a six-column comparison table plus an "avoid this place" mention.
The detail that matters: of the six brunch spots ChatGPT recommended in the casual answer (Tab one), only one showed up in the considered answer (Tab two).
A Growth Memo study published this week quantifies the pattern I’m seeing. Kevin Indig ran 100 prompts twice through ChatGPT-5.2. First a casual prompt like Tab one, then considered, requiring high reasoning, like Tab two. Here’s what he saw.
Citation rate climbed from 50% to 68%.
Sources per response nearly doubled.
4.6x more sub-searches than casual mode.
Only 25.6% of cited domains overlapped between modes.
When the model thinks, it researches three times harder and picks a nearly different set of winners. If your website ranks for "vegan brunch Wilmington" but not for "kid-friendly brunch Wilmington," you might only show up in some of the sub-searches, which weakens your odds of being in the final synthesized answer.
My brunch screenshots are not an outlier. They are the rule.
Here’s what you need to know: AI commoditizes basic information. It cannot replicate earned expertise. When the model thinks, it goes looking for the latter.
THE WINDOW IS NOW
Generic gets you generic
If your "AI search audit" happened more than two months ago, you may have optimized for the generic, autopilot version. That version still recommends businesses. It just recommends the same handful to everyone. The chains. The listicle darlings. The obvious names.
Thinking mode is where differentiation wins. If you serve a specific customer better than anyone else (because of your specialty, your experience, or your approach), you have to name it. Out loud. In writing. In customer language. In enough places that AI can connect the dots.
Generic gets you generic. Specific gets you cited. Right now that's an open window. Soon it's the only way in.
YOUR COMPETITORS ARE ASLEEP
Twenty minutes on Yelp.
BrightLocal confirmed this week: Yelp is still cited in roughly one-third of AI-generated local search results. You read that right. One in three.
Your twenty-minute action:
Log into Yelp business.
Fill in "Specialties" with three to five sentences naming services, neighborhoods, and customer situations. The more multi-criteria, the better.
Fill in "History." Two paragraphs. Who, when, what you fix that others don't.
Fill in “Meet the Business Owner.”
Reply to your three most recent reviews.
Totally Free. And 70% of your competitors have not done this.
WHAT’S THE PROMPT
Two questions, two different answers.
Okay, your turn. Open ChatGPT in two tabs.
Tab one: "Best [insert your service category] in [your city, state]" Run the minimal prompt →
Tab two: "I'm in [your city, state]. I'm looking for the best [your service category]. I need someone who handles [specific situation], is available [your typical availability], and has a strong track record with [target customer type]. Compare the top three in detail. Use web search." Run the high-reasoning prompt →
If you're not in the top three, scroll to the citations. Every source it used is listed. Anywhere a competitor is cited and you aren't is a fix.
CITED
All 3 fetched this dog spa.
Ask any AI to find a "pet groomer in Wilmington, NC who can handle anxious or senior dogs, uses low-stress handling techniques, doesn't use cage dryers, and books out further than two weeks." The answer is the same across ChatGPT, Gemini, and Perplexity: Von Barkee's Dog Spa & Bakery
Source: ChatGPT
Source: Gemini
Source: Perplexity
So, how? Take a look at their homepage and put it next to the query.
“AKC S.A.F.E.-Certified groomers."
"Cage-free options."
"Gentle grooming for senior pets."
They wrote the answer the customer was asking for, before the customer ever typed it. Word for word.
Other signals working for them: an AKC marketplace listing, multiple legitimate local "best of" awards, 13+ years in business, two separately indexable location pages, the exact-match domain. Nothing crazy here. They just named what they actually do, very specifically, in language a real customer would use, in enough places that AI could find it from multiple angles.
Source: AI Mode
Now go fetch,
- Anna
THE POLL
