When I see news stories like “AI-generated fake photos lead people to spend money on renovations,” my first thought isn’t how realistic generated images have become. It’s another question closer to businesses’ everyday concerns: if every shop online looks beautiful, professional, and attentive, what exactly gives ChatGPT a reason to pick yours?
I lean toward an answer that’s a bit of a buzzkill: If you want to be recommended, first make it clear who you’re not a good fit for.
This isn’t some mysterious prompt guaranteed to get you selected. My reasoning is pretty basic: recommendations are about matching needs, and “we can do everything” offers almost no clues for making that match.
What Does “Top Local Choice” Actually Choose?
Suppose someone asks: “My mom has bad knees. We want to eat in Banqiao on Sunday, six people, preferably somewhere without stairs where we can hear each other talk.”
Restaurant A’s website says: exceptional cuisine, a welcoming atmosphere, the top choice for group dining.
Restaurant B says: there is a step at the entrance, and a portable ramp is available; the restroom is on the second floor; Sunday lunch is fairly noisy, but it gets quieter after 2 p.m.; tables for six require a reservation.
B looks less perfect. But for this particular need, it provides conditions that can be used both to find a match and to rule it out. A offers only adjectives. The adjectives are trying hard, but the knees aren’t buying it.
I can’t conclude from this that ChatGPT will definitely recommend B. Whether the information is found, whether it’s credible, and which sources the answer uses all affect the result. But B at least gives the recommendation a reason that can be checked.
Being found and being recommended are two different engineering problems.
Marketing Fears Turning People Away, but Matching Requires Ruling Things Out
Traditional copywriting often carries a particular anxiety: list too many limitations, and customers will leave.
That concern is reasonable. A photographer who publicly states “no same-day delivery” may indeed lose an inquiry; an air-conditioning technician who writes “we don’t take on suspended work on high-rise exterior walls” will also narrow their customer base.
But when the need becomes “I need the photos tomorrow” or “the outdoor unit is on the exterior wall of the twelfth floor,” those limitations are crucial information. Hiding them may only buy twenty minutes wasted on both sides, with no deal possible in the end.
There’s a tension here: exposure wants the widest possible entrance, while recommendations want the most precise conditions. Carrying the logic of search traffic straight over to AI recommendations makes it easy to mistake “more people see you” for “more of the right people come to you.”
So I’d start by revising the service page, without rushing to fill it with “best,” “recommended,” and “expert.” Spell out the service area, how to book, turnaround times, what the price includes, and which jobs need to be referred elsewhere. Turn professional expertise into concrete conditions, and there’s finally something to match against.
Don’t Treat Structured Data as a Consecration Ritual
Some people pin their hopes on formatting, markup, or an article supposedly “loved by AI.” Organizing information helps; treating a format as a guarantee of recommendations is a bit like replacing your network cable and expecting sales to double automatically.
The real headache is conflicting information: the official website says the business is open on Sundays, while other pages say it’s closed; the service page promises home visits throughout Taiwan, but only when you book do you learn that they serve Taipei alone. Even if the information is found, these contradictions leave the reasoning behind a recommendation on shaky ground.
And clear writing still doesn’t mean good work. Fabricated case studies, attractive fake photos, and outdated promises can all be packaged into tidy prose. Work that can be verified, clear dates, and specific service records offer more to discuss than yet another “trustworthy.”
At ten at night, a plumbing and electrical technician deletes “Whatever the problem, we’ll get it done” from his homepage and replaces it with: “For leaks in older homes, send photos first for an initial assessment; issues involving shared pipes require cooperation from the building management committee; I can’t come out tonight.”
Those words sound a little less heroic. The person on the other end of the phone reads them, sends three photos, and adds: “Someone from the management committee will be there at nine tomorrow morning. Can you come then?”