You can start the process with AI. You should. But there are six things AI will not catch, and every one of them can cost you more than the money you were trying to save.

Four of them apply to any apartment you buy in New York. Two more apply if you are buying new construction.

I will go through all six in plain English. First, why anyone is asking the question.

What does a broker actually cost?

When a home sells in New York, the seller will often agree to compensate the buyer's broker. Often, but not always, and never at a rate anyone is required to charge. Commission is negotiable, and it always was.

That said, 2.5 to 3 percent is a useful example of what that compensation can look like in a New York City deal. On a $2 million apartment, that is $50,000 to $60,000.

As a buyer you usually do not write that check. But that money is part of the economics of the deal you are negotiating. So a rational buyer looks at it and asks the obvious question: if I go without an agent, can I negotiate some of that back to me.

It is a fair question. Anyone who tells you not to ask it is protecting something.

Then two stories went around

Earlier this year, Robert Levine sold his home in Cooper City, Florida using ChatGPT. He got five offers within days, sold for $954,800, and estimated he saved roughly 3 percent in commissions.

Then Stuart Thompson, a technology reporter at the New York Times, sold his Hudson Valley home using Google Gemini. He got just over $600,000 on a house he had initially thought might be worth around $550,000, and estimated roughly $36,000 in avoided commissions.

Both stories are true. Both men were happy. The internet drew the obvious conclusion.

But both of them were selling, and neither was in New York

This is the part everyone skipped.

When you sell, you already own the place. You know the roof leaks. You know what you paid and what you fixed. The job is setting a price and getting eyes on it, and AI is genuinely good at that.

Buying is the opposite job. You are trying to find out what is wrong with something you do not own, from people who are paid not to tell you, on a deadline, with your money.

Both of those were also single family houses. One owner, one deed, one tax bill, one lawn.

A New York apartment is not that. Here is what you are actually buying.

In a condominium, you own your apartment outright and you own a share of the hallways, roof, and lobby along with everyone else. You pay a monthly common charge to run the place.

In a cooperative, which is most of the older housing stock in Manhattan, you do not own real estate at all. You own shares in a corporation that owns the building, and those shares come with a lease on your apartment. You pay monthly maintenance.

Either way, you are buying into a shared entity with a budget, a bank account, debt, a board of directors, and a list of repairs coming. That entity is going to decide a lot about your life and your money. Some of it is public. Much of what matters most is not

Part one: buying a condo or co-op 

1. You are not buying an apartment, you are buying into a business 

The listing shows you a kitchen. It does not show you the balance sheet.

Before you buy here, someone needs to read the building's financial statements and answer basic questions. How much money is in reserve for repairs. Whether the building carries a mortgage of its own and when it comes due. Whether owners are behind on payments. What the board has been discussing but has not voted on yet.

That last one is where people get hurt. A special assessment is a one time charge the building bills every owner to pay for something big, like a new roof or a facade repair. They can run into the tens of thousands of dollars per apartment. A building can be six months away from voting one in and nothing about it will appear in a listing, a public record, or anything a chatbot can read.

New York also has a climate law, Local Law 97, that fines buildings for using too much energy. It generally covers buildings over 25,000 square feet, plus certain groups of buildings over 50,000. A building over its emissions limit faces penalties of $268 for every metric ton above it. On a large older building that can run into tens of thousands of dollars a year, and it eventually lands on the owners through the building's finances.

Some of this is public. The parts that matter most usually come from financial statements, board minutes, a management questionnaire, and conversations with people who know the building. You get them by asking, reading, and knowing what a thin reserve looks like for a building of that age and size.

2. The building may be one banks will not lend on

If banks will not write mortgages in a building, buyers cannot buy there. Which means when you go to sell, your pool of buyers shrinks to people paying cash, and your price goes with it.

This happens more than people realize, and it happens for a long list of reasons:

• Active litigation involving the building, especially over construction or money
• One owner or one entity holding too many of the units
• Too many apartments rented out rather than lived in by owners
• Too much of the building's square footage being commercial space
• Thin reserves, or a special assessment that just landed
• Insurance that does not meet a lender's requirements
• Short term rental or hotel style operation in the building
• Land lease buildings, where the building sits on ground it does not own and the lease has a limited number of years left

Fannie Mae and Freddie Mac stand behind a large share of American mortgages. Fannie reviews condo projects and can designate one as Unavailable, which means loans on apartments in that building may not be eligible for Fannie to buy. That determination is not an open consumer database you can go look up.

But that is one list, not the list. Fannie's standards are one layer. Individual banks and portfolio lenders have their own building specific requirements on top of them. For co-ops especially, most lending stays on the bank's own books, so each bank keeps its own roster of buildings it has reviewed and is comfortable in. Your building can work for one lender and not another.

There is no single public database that tells a buyer whether a building will be financeable for them. You find out by calling lenders, or by working with people who have watched deals die at that address before.

3. The recent sales are not telling you what you think

Everyone prices off comparable sales, meaning what similar apartments recently sold for. AI is very good at pulling those up. The trouble is what the numbers actually mean.

Public sale records run months behind. In a market that is moving, you are reading last winter.

Square footage in New York is not standardized and not verified by anyone. Two listings in the same building can measure the same apartment differently. So price per square foot, which looks like hard math, is often built on a number somebody estimated.

And within one building, the same layout can be worth very different money. Higher floor, better light, quieter side of the building, a line that gets the park view and a line that faces an air shaft. A model comparing two sales in the same building will treat them as similar. Anyone who has stood in both apartments knows they are not.

 4. The building has to say yes, not just the seller 

In a co-op, the board interviews you and votes on you. They can turn you down and they do not have to tell you why. People lose apartments at this stage after months of work.

Boards also set their own financial rules and most are not published anywhere:

• How much you can borrow. Some buildings cap financing at 50 percent. Some are all cash only.
• How much cash you need left over after closing, often one to two years of maintenance payments sitting in the bank.
• How much of your monthly income can go to housing. Many co-ops cap this at 25 percent, where a bank would let you go to 40.
• Whether they will even count your retirement accounts toward your reserves. Some do, some do not.

Condos are looser but not free. The building typically has the right to match your offer and buy the apartment instead. There are rules about renting it out, renovating, and pets that will affect what you can do and what it is worth later.

Ask AI whether you qualify to buy an apartment and it will answer using bank standards. The board is not a bank and does not use them.

Part two: if you are buying new construction

5. The recorded price is not the price that was paid

When you buy a brand new apartment, you are buying from the developer who built it. The industry calls them the sponsor.

Sponsors care enormously about the number that shows up on paper, because every future buyer and every appraiser will look at it. So instead of dropping the price, they pay you in other ways. Covering your closing taxes. A year of free monthly charges. A storage unit. A check at closing. The industry word for this is a concession.

The sale gets recorded with the city at the full sticker price. The concession is not recorded anywhere.

So when your chatbot pulls up recent sales in the building and reports that units are trading at $1,850 per square foot, the real deal may have been $1,720. It is reading a number that did not happen, and it has no way to know.

6. The closing costs are much bigger here, and only negotiable before you sign

Closing costs are the fees you pay on top of the purchase price on the day you take ownership. What they add up to varies a lot depending on whether it is a condo or a co-op, whether you are financing, the purchase price, mansion tax, mortgage recording tax, title insurance, and a handful of other things.

Buying from a sponsor can be materially more expensive, because the contract often shifts costs onto you that a resale seller would normally carry.

The biggest one is transfer taxes. When property changes hands in New York, the city and state each take a cut. Normally the seller pays it. In new construction contracts, the buyer frequently does. On a $1 million purchase, the combined city and state transfer taxes come to roughly $18,250 that nobody warned you about.

Sponsor attorney fees and a working capital contribution, a one time payment into the building's bank account, may apply on top of that.

Here is the part that matters. Many of those items may be negotiable before you sign. Your leverage after you have agreed to the terms is much worse.

I have never seen AI stop someone and say: hold on, before you sign this, go back and ask them to eat the transfer taxes. It will explain what a transfer tax is very clearly. Usually after you have agreed to pay it.

The pattern underneath all six

AI only answers the question you ask

Every problem above has an answer. Ask the right question, at the right moment, in the right order, and you will get a good one.

But you have to know to ask. And you do not, because you buy an apartment maybe twice in your life.

A good broker tells you the thing you did not ask about. That is most of the job. It is not looking things up. It is knowing what to go dig into before you sign.

AI waits to be asked. And it still gets things wrong while sounding completely sure, which matters a lot more when the wrong answer costs $200,000 than when it costs you a bad dinner.

And the person across the table is a person

Every deal here has two sides and both have someone working them.

Walk into a sales office with AI in your pocket and you are negotiating against an experienced local salesperson whose loyalty is to the developer. Not to you. They are doing their job well when you do worse.

Negotiating is not a document. It is knowing the developer has 11 apartments left and a construction loan coming due. It is calling an agent I closed two deals with last year at 9pm to save something that is falling apart. AI gives you black and white. Deals get done in the gray.

So what is AI actually good for

I use it all the time.

Filtering listings against what you actually need. Running the monthly numbers on three scenarios in a minute. Explaining a page of building documents in plain English once I have handed it the document. Writing the list of questions to send the building manager. Double checking what a broker just told you, including me.

It is a very good second opinion and a very good preparation tool. It makes an informed buyer faster. It does not make an uninformed buyer safe.

Would you bet a million dollars on it?

That is really the whole question.

You are about to spend somewhere between $1 million, $10 million or more. Would you hand that decision to a piece of software and trust that it found you the right apartment at the right price, and that nothing it forgot to mention shows up down the road and makes your home unappealing to the next buyer.

I would not. Not yet.

Ask me again in two years and the answer might be different. Right now the honest one is that AI has made me better at my job, and it has not come close to doing it fully.