The Cialdini Persuasion skill for Claude

You download the zip at the top of this page, put it in your Claude environment, and from that moment Claude scores texts according to this logic. What a SKILL.md exactly is, and why a text file steers the behaviour of an AI, is explained in what are Claude skills NL.

The problem the skill solves is the vagueness of the advice you normally get. Ask an AI to make your email more persuasive and you get: add social proof, make it more personal, create urgency. That sounds like advice and is not, because it does not say which sentence causes the problem, how heavily that principle weighs in your situation, or whether you even have the evidence it rests on. This file turns that around. It scores per principle, it quotes the sentence that causes the score word for word, it delivers a rewritten alternative that uses only information you have supplied, and it ends with a list of what it deliberately did not apply.

It is written for anyone who writes a text with a decision behind it: business owners sending cold emails, marketers building landing pages, consultants giving pitches, service providers sending out quotes. It is part of the skill library we make freely available from the AI and automation service, with no account and no sales email afterwards.

There is one thing you should know beforehand: this skill makes your text more persuasive, not your offer better. It adds no evidence you do not have. If you have no clients, no references, no figures and no real capacity limit, the analysis stays empty on exactly those points, and that is the outcome. The file says so itself: an invented deadline or an invented number of clients is not persuasion but deception.

You ask before you have given anything. The file checks whether reciprocity is absent here; it does not establish its prevalence across all cold messages. A no-obligation call sounds like a gift but is not: it is a request for time, wrapped up as reassurance. A gift is something the reader keeps, even if they do nothing further with you. In the repaired example, the agency offers to prepare the supplied free analysis; no completed work, ranking or page count is assumed.

The text could have been sent to anyone. Liking and unity are two different things that often get lumped together. Liking is about similarity, a genuine compliment and working together towards the same goal. Unity goes a step further: not resembling, but belonging together. Region, profession, generation, shared history. A supplied connection may be relevant, but the file does not prove that no competitor can share it or that it is almost always unused.

The authority is self-declared. Specialised, expert, market leader and leading are words nobody can check. Cialdini describes authority as something that works only as long as it is visible and demonstrable: titles, publications, measurable experience, published work. An adjective is therefore best replaced with a hard number, and if the number is not there, the skill drops the claim rather than dressing it up.

The scarcity is invented. This is the most dangerous of the seven, because it is also the easiest. Typing a deadline takes a second. The instruction is strict about this: scarcity only works if it is real and the reason is stated alongside it. A real capacity limit of three places this quarter is allowed, a date you make up is not. And once a reader spots that your deadline was fake, they will doubt your numbers too, next time.

All seven at once. More principles is not more persuasion. Cramming all seven into a short text is explicitly forbidden in the instruction itself, because two or three well-executed principles beat seven half-done ones. During an assessment the skill therefore also runs an excess check: more than one scarcity element, or social proof in four sentences in a row, lowers credibility instead of raising it.

In diff form, using the elements from the skill's own example: what the skill refuses, and what it puts in its place.

Then comes the theory section, and for a skill it is unusually extensive. All seven principles are given their own definition, plus the explanation from Pre-Suasion that the moment before the message helps determine how the message lands. That is why the skill assesses not only whether a principle is present, but also where it sits. Alongside that, four variants are listed that it may use when the context calls for them: the classic six without unity, Cialdini combined with an objection analysis, Cialdini combined with AIDA or PAS where AIDA sets the order and Cialdini supplies the content per stage, and a weighted scoring where principles that do not belong stay outside the score instead of getting a zero.

The core is a list of seven fixed actions. It first determines the text type, the target audience and the desired action, and on that basis decides which principles carry weight and which do not apply. It gives each applicable principle a provisional 0-to-5 score with a verbatim sentence or explicit absence. Excluded principles receive N/A, not zero; the score is not a conversion forecast. It quotes the weakest sentence for each principle word for word, or notes that the principle is entirely absent. It delivers a rewritten alternative per principle that uses only information you have supplied.

It ranks the improvements by hypothesised impact to test, so you know what to tackle first. It delivers a complete rewritten text. And it runs an honesty check, in which every principle that would rest on an untruth is flagged and not applied.

Before anything gets scored, the skill runs through a mandatory input checklist: the text itself, in full, the text type and the channel, the target audience and how much they already know about you, the desired action, what evidence genuinely exists, whether there is a real constraint on time, place or capacity, and what you have in common with the reader. It only asks about what is missing and then stops asking, so you get no questionnaire but targeted questions about genuinely missing input.

The output has a fixed structure of six blocks, in the normal order, except that the vulnerable-audience variant puts the honesty check first: a context summary that also states which principles are left out of consideration and why, a score table with score and weight per principle, an analysis per principle in the three-part form of present, problem, alternative, a priority list by expected impact, the full rewritten text with notes where a principle has been added, and the honesty check.

A striking number of rules are about language. The file further requires the you-form, forbids dashes in running text, forbids rule-of-three lists in which every set has exactly three items, and keeps a list of AI words that may not be used: in the world of, discover, unlock, seamless, effortless, take to the next level, dive into, in a nutshell. Hollow superlatives are also on the banned list. And there is a strict quoting rule: when the skill points to a sentence, it quotes it word for word, it never paraphrases.

Finally, the file has a separate assessment mode, which it even calls the default mode. Hand it an existing text for assessment, and two extra parts are added: a marked-up version in which the principle each sentence serves is noted after it, and nothing is noted after sentences that do nothing, which are your candidates for cutting, and an excess check that points out where the text overdoes it. If you want to learn to set up an instruction file like this yourself, writing a SKILL.md NL explains how that structure works.

The six principles come from Influence, 1984. Robert Cialdini described six principles there that can influence whether people say yes: reciprocity, commitment and consistency, social proof, authority, liking and scarcity. The book is not a collection of sales tricks but a report of field research, partly carried out by apprenticing himself to salespeople, fundraisers and advertisers to see what actually worked in practice.

Unity was added later. In later work, including Pre-Suasion from 2016 and the expanded edition of Influence from 2021, Cialdini added unity as a seventh principle. The difference from liking is subtle but important: liking is about resembling, unity is about belonging together. That is why it appears weighted high in the scan in FIG.01 for cold contact, while it can be left out of the score entirely for a short transactional text.

The moment before the message matters too. Pre-Suasion is about exactly that: what the reader's attention is focused on before your offer arrives colours their judgement of that offer. In practice this means the skill assesses not only whether a principle is in the text, but also where. An evidence block below the button works differently from the same evidence block above it.

The applications come from Yes!, 2008. Cialdini wrote that book together with Steve Martin and Noah Goldstein, with fifty worked applications of the principles in real situations. That is the material the skill draws on when it formulates an alternative itself instead of only naming a problem.

The underlying psychology comes from Kahneman. Thinking, Fast and Slow from 2011 describes the fast judgement systems these principles work on. That also explains why the principles are so easily abused: they act on the fast system, which does not check whether the deadline really existed. That is precisely why the honesty check is a fixed output block and not an optional warning.

Fogg is included for when persuasion is not the problem. The file cites B.J. Fogg and his combination of motivation, ability and prompt as a useful addition when a text does persuade but the action still costs too much effort. Then more social proof helps nobody: the threshold needs to come down instead. Anyone who wants to take that further can turn to the Hook Model Builder skill, and anyone who wants to apply the same thinking to behaviour change inside an organisation can turn to the EAST Framework Behaviour Coach skill.

Even without installing the skill you can improve your own text with these principles: give something before you ask, replace an adjective with a number, and leave out scarcity if it is not real.

After that, using it is simple: paste your text and name the channel, the target audience and the desired action. Also say what evidence you genuinely have, because that is the material the alternatives are built from. If anything is missing, the skill asks about it once and then carries on. If you get stuck during installation itself, the extended route is in installing Claude skills NL. If you want to learn to set up this kind of work with your team, that is the subject of our AI training.

The instruction also states where the emphasis should sit in each situation. Heavy on reciprocity and unity in cold contact, with social proof as evidence rather than an opener. Heavy on commitment and consistency for warm leads and multi-step journeys, since there the reader has already taken a small step to build on. Heavy on authority and social proof for high amounts, long contracts and risk-averse buyers. And it leaves out scarcity entirely if there is no real constraint, because an empty score is better, according to the file, than an invented deadline.

There are also situations where you are better off leaving it alone. Texts where persuasion is inappropriate are one of them: a medical package leaflet and a legal notice are explicitly excluded in the file. Those should contain information, not influence, and a leaflet that persuades instead of informs is a fault, not an improvement. For vulnerable audiences and for topics with health or financial consequences it does work, but then the honesty check moves upward, so you read what you may not claim before you read how to say it more persuasively.

The most important limit is a different one, and it is not technical. This skill makes a good offer more visible. It does not make a bad offer good. If your offer does not match what the reader needs, or the price does not match what you deliver, a stronger wording changes nothing about that: you have an offer problem disguised as a text problem. And if you write a text that persuades but the desired action costs too much effort, the problem sits in the threshold, not in the persuasion.

Two practical limits to finish. It only assesses the text you supply yourself, it pulls nothing from your site and measures no conversion. And it predicts no result: no expected response rate comes out of it, because that would be a number it cannot know. What it delivers is a judgement on the text, with a priority list by hypothesised impact to test and the reasoning attached, so that you make the call.

Applying seven principles to your text can substantially raise the odds of a yes, but it counts on a reader who is already there. If your traffic does not change, you are raising a percentage of a small number. Want to know how big that number is right now? The free SEO scan answers in a few seconds. In the guide writing models compared you see how this model relates to the others. And if you want to talk through what else AI can do for your business, from stand-alone skills to work that gets prepared every day, we simply do that in a conversation.