The AI Negotiator skill for Claude

You download the zip at the top of this page, put it in your Claude environment, and from that moment Claude prepares conversations following this logic. What a SKILL.md actually is, and why a text file steers an AI's behaviour, is explained in what Claude skills are NL.

You probably recognise the problem the skill solves. A conversation about money is coming up, you roughly know what you want, and beyond that you do not really prepare because you do not know how you would. Then the conversation starts, the other side opens with a number, and from that point it is only ever about that number. You nudge it up, they nudge it down, and whoever holds out longest wins. That is called positional bargaining, and the method this skill rests on is explicitly an answer to it.

It is written for people who negotiate without it being their profession: freelancers and agencies who have to defend their rate, business owners renewing a supplier contract, employees having a salary conversation, anyone with an expiring contract on their desk. It is part of the skill library NL we make available for free through the AI and automation service, with no account and no sales email afterwards.

One thing you should know upfront: the skill does not negotiate for you. It prepares you for a conversation you have yourself. It delivers a strategy, not a pep talk, it does not invent a counteroffer you have never heard, and it does not promise that a particular outcome will be reached. What it does do is make explicit the two things most people leave implicit: what you do if there is no deal, and where the line sits above which a deal is better than no deal.

The position gets mistaken for the interest. The position is what someone asks for. The interest is why they ask for it. Behind fifteen per cent discount, according to the file, there is often a fixed internal budget, a buyer who has to justify their decisions, or a cash flow that only frees up in two months. On the position you can only say yes or no. On the interest you can think up five different answers, and three of those cost you nothing.

The BATNA is never spelled out. Your best alternative if this deal falls through determines your negotiating position, not your arguments. And it is not what you hope, it is what you would actually do. Two outstanding proposals are not two clients. As long as you have not said that out loud, you behave at the table as if your alternative is stronger than it is, and the other side notices that before you do.

Brainstorming and deciding happen at the same time. The method says: first think up options that leave both sides better off, and only then choose. Do those two things at once and you shoot down every idea before it is fully formed, leaving you with exactly the one variable you started with.

There is no objective standard. Without an external standard the outcome is decided by whoever pushes hardest, and nobody walks away from that feeling good. Market prices, benchmarks, third-party quotes, an index or a valuation move the conversation away from willpower and onto facts.

Every concession is free. This is the mistake that runs deepest. You give something away to create movement, you ask for nothing in return, and the only thing you learn is that there is more to be had. In its review mode the skill therefore flags every concession made without a counter-ask, and it never proposes one itself either.

In diff form, with lines from the skill's own example: what comes out and what replaces it.

Then comes the theory section, with the four pillars of the method spelled out. Separate the person from the problem, so tough on substance and soft on the relationship. Focus on interests, not positions. Think up options that leave both sides better off before you choose, and pull brainstorming and deciding apart. And use objective criteria, so the outcome is not decided by whoever pushes hardest. Alongside that, BATNA and ZOPA appear as the two concepts that steer the entire preparation, with the emphasis that a BATNA is what you actually do, not what you hope for.

The core is a list of seven fixed actions. It summarises the situation in a short diagnosis and states whether this is mainly distributive or integrative. It splits the positions from the interests for both parties. It makes your BATNA explicit, including an honest verdict of strong, reasonable or weak, plus at least one way to strengthen it. It estimates the counterparty's BATNA, with the assumption behind that estimate clearly marked.

It determines the ZOPA in concrete numbers or conditions and names the walk-away point. It delivers at least five win-win options that are not about price, such as payment term, contract length, volume, scope, exclusivity, reference rights or a start date. And it writes an opening line that names the counterparty's interest out loud, plus answers to the two or three most likely objections.

Before anything gets worked out, the skill runs through a mandatory input checklist: what is being negotiated and what is currently on the table, who the counterparty is and what role that person has within their own organisation, your goal in a number or a concrete condition, what you actually do if the deal falls through, your floor and whether that is yours or someone else's in your business, what you already know about the pressure the counterparty is under, and whether the relationship continues after this deal or is a one-off.

It asks for everything missing all at once and then stops asking. If you do not know something, it proceeds with a marked assumption instead of continuing to press for an answer.

The output has a fixed structure of eight blocks, always in the same order: situation diagnosis, positions and interests per party, the BATNA analysis, ZOPA and walk away, five or more win-win options each stating what it costs you and which interest of the other side it addresses, objective criteria, an opening line with objections, and finally a source and accountability section stating which model was used and which assumptions the analysis rests on.

Here too, remarkably many rules are about language. The file requires the you-form, forbids dashes in running text, forbids rule-of-three lists, and keeps a list of AI words that may not be used. There is one more rule specific to this skill that makes the difference: conversation lines are written as spoken language, because you need to be able to say them out loud without cringing. You can hear that in the opening line in FIG.01, which sounds like someone talking, not someone reading aloud.

Finally, the file has a separate mode for reviewing existing work. Hand it a proposal, an email exchange or a planned approach, and you get five things back: a diagnosis of where the weakness lies, a marking per passage placing POSITION or INTEREST after each sentence so you can see how positional your writing is, a list of what is missing from the checklist, a concession analysis flagging every concession made without a counter-ask, and a rewritten version in the standard output structure. If you want to learn how to set up an instruction file like this yourself, writing a SKILL.md NL explains how that structure works.

The method comes from Getting to Yes, 1981. Roger Fisher and William Ury described the work of the Harvard Negotiation Project in that book, in later editions together with Bruce Patton. The method is also known as principled negotiation and is explicitly meant as an answer to positional bargaining: two parties inching towards each other from an opening bid, where the outcome is decided mainly by whoever holds out longest.

BATNA comes from the same source. The term is generally credited to the Harvard Negotiation Project and is by now the most cited piece of negotiation vocabulary there is. The core is simpler than the acronym suggests: your power comes not from your arguments but from what you will do if the conversation fails. Whoever improves that alternative first has to perform less well at the table.

Getting Past No covers the difficult counterparty. Ury wrote that follow-up in 1991, for situations where the other side attacks, blocks or uses tricks. As an addition, it comes in if the other side threatens to walk away or keeps adding new demands after you have given in: it adds a block on naming that behaviour without answering it.

3-D Negotiation widens the field of play. David Lax and James Sebenius described in 2006 that the table is only one of three dimensions. Alongside what you say at the table, the setup of the deal itself and the question of who you invite to the table also matter. If a conversation gets stuck on the table itself, it reaches for this: look at who else should be sitting there, or how you could carve up the deal differently.

Never Split the Difference supplies the tone. Chris Voss described tactical empathy, labelling and calibrated questions in 2016, based on his work as a hostage negotiator. For emotionally charged conversations it draws on this, for example a salary conversation or a conflict with a long-running client, with labels like it sounds like this is mainly about timing.

Distributive versus integrative. Finally, the file uses the distinction between dividing a fixed pie and growing it. That distinction sits in step 1 of the skill for good reason: it determines the whole approach. In a one-off transaction with no follow-up, such as selling a used piece of business equipment, the anchor price matters more than the interest analysis. In a relationship that continues, it is exactly the other way round.

Even without installing the skill, you can use these principles to prepare your next conversation better: write down what you do if there is no deal, find an objective standard to go with it, and make sure there is more than one variable on the table before you sit down.

After that, using it is simple: describe the negotiation in plain sentences, as in FIG.01. What is on the table, who the counterparty is and what role that person has, your goal in a number, what you do if it falls through, your floor, and whether the relationship continues afterwards. If you do not know something, just say so: the skill proceeds with a marked assumption instead of continuing to press for an answer. If you get stuck on the installation itself, the full route is in installing Claude skills NL. Want your team to learn how to set this kind of work up? That is the subject of our AI training NL.

The file itself names, per situation, which addition applies. When a counterparty attacks, threatens to walk away or keeps adding new demands after you have given in, Getting Past No comes into play, with a block on naming that behaviour without answering it. If the deal gets stuck on the table itself, it switches to 3-D Negotiation: look at who else should be sitting at the table, or how you could carve up the deal differently.

If the conversation is emotionally charged, for example a salary conversation or a conflict with a long-running client, the tactical empathy from Never Split the Difference comes into play. And for a one-off transaction with no follow-up, it switches back to purely distributive negotiating, because then the anchor price matters more than the interest analysis.

There are also situations where you are better off leaving it be, and those are stated in the file itself. Legal advice on contract clauses is one: preparing for a conversation is a different thing from writing down what becomes binding, and for that you need a lawyer. Labour disputes with a formal legal procedure are the second. And mediation between two parties where you do not take a side is the third: this skill by definition takes your side, so it is no good as a neutral mediator.

The biggest limitation is a different one, and it is not technical. It prepares the negotiation, it does not conduct it. You are the one sitting at the table, and all sorts of things happen there that no preparation catches: the other side names a number you did not expect, it turns out someone else has a say who you did not know about, or the conversation suddenly turns to something completely different. What the preparation gives you is that in those moments you know where your line sits and which variables you can still use, and that is exactly what disappears first under pressure.

One more honest limit: the quality of the analysis depends entirely on the quality of your description. If you do not mention that your contact does not decide alone, the skill will not account for that. If you name a floor that actually belongs to your business partner, it builds a strategy on a number that is not yours. The checklist asks about this on purpose, and that is not a formality but exactly where most preparations quietly go wrong.

The most honest way to negotiate better is not found in a conversation technique. It is found in your diary. Anyone with enough enquiries can let a deal go, and that is exactly what a strong BATNA is. If too few enquiries come in, that is where the real work sits. The free SEO scan shows in seconds where your site stands. And if you want to talk further about what else AI can do for your business, from loose skills to full automation, we simply do that in a conversation.