How To Use AI To Negotiate Against Other People Using AI

At some point in the last 12–18 months, I realized I was no longer just negotiating against people.

I was negotiating against people who had their own AI sitting in the background.

You can feel it. Emails get tighter. Redlines look strangely consistent. Brokers show up with eerily polished talking points that hit every weak spot in your position.

So I stopped thinking of AI as my private edge and started treating it as the other side’s junior partner. Then I asked a simple question:

“If they are using AI to press their advantage, how do I use AI to negotiate against their AI?”

What The Other Side’s AI Is Probably Doing

You do not need to see their screen to make a pretty good guess.

On any serious CRE negotiation, the other side’s AI is likely helping them:

• Summarize documents faster
  OMs, leases, LOIs, PSAs, lender quotes. It is compressing complexity into talking points.
• Normalize positions across deals
  “What is market” on SNDA language, exclusivity, assignment, caps, carve‑outs.
• Generate counters and “reasonable” alternatives
  Tighter deadlines, smaller caps, higher deposits, more seller covenants.
• Surface your weak spots
  Any inconsistency between what you said before and what you are saying now.

If you pretend they are not doing this, you negotiate like it is 2015 and you are outnumbered on the back end.

Instead, I assume their AI is in the room, and I bring mine too.

Step 1: Build A Model Of The Other Side Before You Ever Talk Numbers

Before I send a serious mark‑up or LOI, I feed my own AI:

• The full email thread
• The OM and major exhibits
• Any public information on the counterparty (prior deals, press, litigation, capital sources)
• Redlines or term sheets they have sent on other deals, if I have them

Then I ask very direct questions:

• “Based on this pattern, what does this party care about more: price, speed, certainty, or control?”
• “Where have they held firm in past deals, and where have they traded?”
• “If you were advising them, what three points would you tell them not to give up here?”

AI is very good at this kind of pattern spotting.

The output is not magic. It is a working profile:

• They hate open‑ended indemnities
• They are obsessed with timeline
• They will swallow some economic pain if they can brag about headline price
• Their credit committee seems rigid on leverage or DSCR

I walk into the negotiation already assuming:

“Their AI has told them exactly where I am weakest and where they have room. I am going to do the same from my side.”

Step 2: Let AI “Red‑Team” Your Own Position Before They Do

I do not send a mark‑up or proposal until my AI has tried to tear it apart.

Here is the sequence.

1. Give AI my draft LOI / redlines plus their last version
2. Ask:
   • “If you are the counterparty’s advisor, what are the five easiest ways to attack this draft?”
   • “Where do my positions look extreme relative to the data and to what I have already said?”
   • “Where am I accidentally signaling that I am more flexible than I claim?”
3. Then tighten or reframe:
   • Language that is vulnerable to being split or re‑interpreted
   • Numbers that do not sync with my stated story
   • Soft spots where my walkaway does not match the economics on the page

This is where AI for CRE is particularly helpful because it knows the domain:

• It can compare my carve‑outs and caps to “market” samples I have fed it
• It can see when my debt assumptions do not line up with current lender term sheets
• It will flag when my “I need 30 days” line makes no sense given the clean file I already admitted I have

I want those hits from my side before their model lands the same punches in someone else’s inbox.

Step 3: Use AI To Design Concessions And Packages, Not Just Responses

Most people use AI to rephrase what they already want to say.

You can use it to structure the entire negotiation.

Given the counterparty profile and my real constraints, I ask:

• “Design three concession packages that keep my economics intact but feel meaningful to them, based on their priorities.”
• “If I must give ground on X, what can I reasonably ask for on Y and Z to more than compensate?”
• “What sequence of asks has the highest chance of getting me to [target outcome] in [N] rounds?”

Examples:

• For a seller who loves certainty:
  • I may let AI help structure tighter financing and DD timelines in exchange for:
    • Real price protection
    • Clear closing mechanics
    • Fewer “hidden” obligations post‑close
• For a buyer obsessed with price:
  • I may frame non‑price terms (reps, survival, escrows, access) as chips they can “win”
  • While AI checks that my total risk does not quietly explode in the background

The point is to negotiate like a chess player, not a checkers player.

Their AI is already generating sequences of counters. I want mine designing the whole game tree from my side, given what we know about them.

Step 4: Let AI Watch The Tape Between Rounds

After a hard call or key email exchange, I drop the transcript or thread into my system.

I ask:

• “What did they reveal about their real constraints or walkaway point that they did not say directly?”
• “Where did they hedge or soften language compared to prior messages?”
• “If you had to estimate their actual priorities in order, what would they be now?”

This is where the “against other AI” part gets interesting.

If they are using AI to polish talking points, the rough edges they fail to smooth out are even more informative:

• A single sentence they refused to edit
• A term they keep restating in their own words
• A clause they “forget” to push on, even though their redlines are aggressive elsewhere

AI is better than my memory at:

• Catching those shifts across many drafts
• Quantifying how much weight they put on an issue over time

So before I answer, I have a short, clear summary of how their position is actually moving, not just how it feels in the moment.

Step 5: Rewrite For Humans, Aware Of Their AI

One more layer that matters:

I assume my emails, memos, and comments are being fed to their AI, just like I feed theirs into mine.

So I write for two audiences:

• The human across the table
• The model that will summarize what I just said

What that changes:

• I keep my core story consistent
  AI hates inconsistency. If I know it is scoring me for that, I stop giving it ammunition.
• I am very explicit about non‑negotiables
  Vague language invites counter‑prompts like “find ambiguity here.” I do not want that.
• I introduce facts and constraints that their model has to carry forward
  Lender terms, IC requirements, third‑party deadlines. I seed those into the record so their system cannot pretend they do not exist.

I will often literally ask my own AI:

“Rewrite this email so that if the other side’s AI summarizes it, it will output: ‘X is firm on A and B, may flex on C and D, and has real constraints on E.’”

That way I am negotiating not just with the person, but with the summary that will be sitting in front of them when they huddle with their own tools or team.

Step 6: Use AI To Decide When To Walk Away Earlier

The final, and maybe most important, piece is exit timing.

Because we track prior negotiations and outcomes in the same system, I can ask:

• “Based on similar counterparties and patterns, how likely is this negotiation to land inside my acceptable band?”
• “What does the pattern of their responses tell you about whether we are actually getting closer, or just burning time?”

If my AI says:

• “You have seen this movie 12 times, and 10 of them ended with bad risk‑adjusted terms or no deal,”

I take that seriously.

Other people’s AI is making them more efficient at dragging negotiations out with polished language and endless “one more ask” rounds.

My AI’s job is to tell me, very bluntly, when I am falling into that trap again and when it is time to reallocate attention to a cleaner opportunity.

What This Looks Like In Real CRE Work, Not Theory

All of this sounds abstract until you see it on a real deal.

On an actual transaction, “using AI to negotiate against their AI” has meant:

• Catching that their supposed “drop dead” date was more flexible than the emails implied, because their language softened right after some public news about their capital partner
• Designing a concession package that traded two low‑value items to me for one high‑value item to them, which got the deal across the line without touching my real walkaway number
• Spotting that their redlines on legal terms were copy‑pasted from another deal that had a different risk profile, and using that to push back credibly
• Walking away from a negotiation one full month earlier than we would have in the past, saving internal time and focus, because the pattern matched a set of prior failed negotiations almost point for point

The AI part is not visible in the signed PSA.

What is visible is a cleaner deal, fewer ugly surprises, and less time lost on negotiations that were never going to land inside a rational box.

If You Want To Try This Without Rebuilding Your Stack

You do not have to rebuild your tech stack to start negotiating this way.

Start small:

1. Pick one live negotiation.
2. Feed your AI: –
   • The email thread
   • The term sheets or LOIs
   • Any public info on the counterparty
3. Ask:
   • “If you were advising them, how would you attack my current position?”
   • “What concession packages could get me to my target outcome?”
   • “What patterns in their behavior suggest where they will and will not move?”

Run that experiment once. See how different your next counter looks.

Because if they already have AI sitting in their corner, you can either pretend it is not there, or you can bring your own and start negotiating with both of them in mind. This is where CRE Agents can help with.

Frequently Asked Questions About Using AI To Negotiate Against AI In CRE