Field Notes · 8 min read ·
Second-prompt optimization, part 1: Google Maps
If you have a shop, being visible to AI for “furniture store near me” is now free: keep your Google Maps record clean and that is the whole job, so paying a monthly fee for it buys nothing. The sale closes one question later, when somebody says “two seats, this is my budget”.
Google just took the thing your AI-visibility vendor is selling you, and gave it away for free.
If you run a shop with a door, "will the AI recommend me" has collapsed into one fairly boring question: is your Google Maps record in good shape? That is not a strategy, and it is not a product either. It is a Tuesday afternoon and a checklist.
The money moved one question later. This post is about where it went, and why almost nobody is standing there.
For physical businesses, Google took prompt 1 off the market
Ask an assistant for a furniture shop nearby and it does not go off reading websites. It asks a map, and the map answers.
Google shipped Grounding with Google Maps for Gemini, wired directly to more than 250 million places. What comes back is not your carefully written copy, it is places, reviews, photos, addresses and opening hours. And it behaves the way a map behaves: a "near me" question uses the coordinates, so your customer never has to type your district for you to be in or out of the running.
Google has also published how it sorts that list, and the three factors are relevance, distance and prominence. Look at what you can actually move. Distance is fixed unless you rent a different building. Relevance is a form any competitor fills in over one afternoon. Prominence is reviews and years of them, and there is no invoice you can pay to skip ahead.
So for this whole category, one equation now holds:
My shop should get recommended by AI ≈ my Google Maps presence should be excellent.
Once that is an equation rather than a strategy, ask what is left for anyone to sell you. Distance cannot be moved, prominence cannot be bought, and everything else is housekeeping you could have done yourself.
It is not even purely Google's map, which is worth knowing. ChatGPT draws its place data through Mapbox, whose point-of-interest layer has been Foursquare's since 2018, so the record one assistant reads is not the record another one reads. Keep yours clean in more than one place. But notice that this is plumbing rather than strategy, and plumbing is not a category anyone should be charging you a monthly fee for.
Prompt 1 is not a market any more. It is maintenance.
Being listed is not being chosen
Here is what the entire industry currently measures: whether you showed up. Share-of-voice dashboards, prompt trackers, visibility scores, all of it is a very expensive way of watching prompt 1.
Think of prompt 1 as the bouncer at the door. He decides who gets inside, which matters enormously, and he has never once decided who anybody goes home with.
Think about what you actually do when a friend gives you three restaurant names. You do not thank them and book the first one. You ask the second question, and the second question is always the real one, because it is the one with your own situation inside it.
People behave the same way with assistants. OpenAI's study with the NBER found that 49% of messages are "Asking", which the authors read as people using the thing as an advisor rather than as a lookup tool. An advisor, by definition, gets asked a follow-up.
Nobody has ever bought a sofa at prompt 1.
The second prompt is personal, and your listing is not
Prompt 1 sounds like "good furniture shops near Kadıköy". Prompt 2 sounds like an actual human being:
- "We are getting married in September and have 80,000 lira for the whole apartment."
- "I am a student, moving alone, and I need it delivered this week."
- "Small flat, third floor, no elevator. Which of these three actually helps?"
Notice what changed between the two. Prompt 1 asked a question about the world, and prompt 2 asks a question about them. To answer it at all, the model has to reach for information that no map has ever stored.
And when it cannot find that information, the assistant does not shrug and say it does not know. It answers from whatever it can see, which means the shop with the most readable public detail wins by default, rather than the shop with the better offer.
Your listing does try to help here. Google gives you business attributes to fill in, and says plainly that adding them can make you show up in searches for places with those attributes. So the field genuinely exists, and you should fill it in.
But look at what is on the form. Wheelchair access, wifi, outdoor seating, accepted payment types. It is a page of checkboxes, and someone else decided what the boxes say. Filling it in feels like a customs declaration when what the customer wanted was a conversation, and a customs form has never once talked anybody into anything.
There is no checkbox for "good for a couple furnishing their first apartment on 80,000 lira."
Campaigns answer the second prompt, and campaigns expire
The obvious objection is that this slot already exists, because Google Business Profile lets you publish offers and posts. Put the campaign in there.
You should, since it is free and it is the only slot you have been given. But look at how the slot was built. Google's own documentation says posts are archived after six months unless you set a date range, and that is not a bug. A registry's job is to hold the things that stay true, while an offer's entire job is to stop being true on a particular Tuesday.
And even a perfectly maintained offer cannot do the one thing a campaign was invented to do.
The campaign you built for newlyweds is invisible to the AI. The student's two-for-one never comes up.
That is not because you executed it badly. It is because "two-for-one for students" was never aimed at a query, it was aimed at a person. The assistant has no idea it is talking to a student, and your listing has no way to say "if it is a student, mention this one." A listing broadcasts what you are to everybody. A campaign leans over and whispers to somebody specific.
So you can win prominence, spend a year doing it, get recommended in every prompt 1 in your district, and still lose the sale to the shop that taped a handwritten sign in the window, because the sign was allowed to say something your listing structurally cannot.
Nobody has measured how big this is, including us
There is one number I went looking for and could not honestly find: what share of AI search is a person hunting for a business with a door.
The most careful public work is the NBER study built on 1.5 million conversations, and it tells you that "seeking information" is one of three uses making up most of the volume, and that it substitutes closely for web search. It does not break out how much of that is local, and I could not find anyone who has.
Every percentage I tried to trace ran back to a marketing blog citing another marketing blog. So when somebody quotes you a confident number here, ask them where it came from, and watch what happens.
A made-up denominator is worse than an admitted gap.
Second-prompt optimization is the actual job
Prompt 1 is still worth doing and worth doing properly. Keep your record right in more than one place, treat reviews as the compounding asset they genuinely are, and accept that prominence is slow on purpose, because that slowness is the only moat in this whole picture that a competitor cannot buy their way past.
Then stop, because that is exactly where the vendors stop, and it is one question short of the money. Optimising prompt 1 harder from here is like a shop repainting its sign every month while the till stays quiet, since the sign was never the reason people were walking past.
What is left over is different in kind. Your real reasons for someone to walk in change every week, and they are aimed at a couple, a student, a person on a third floor with no elevator. None of them fit in a checkbox, none of them survive six months in an archive, and none of them currently live anywhere an assistant can reach.
Getting them somewhere an assistant can reach is what we mean by second-prompt optimization. It is what we are building at EOMA, and it is why "let's increase your AI visibility" has never sounded to us like a finished sentence.
Part 2 is about the third prompt, the one that comes after they have chosen you: what still has to be true on the day, and what makes somebody actually get in the car and walk through your door.