Someone turns down your proposal. You ask AI to strengthen your arguments. It produces a more persuasive email.
But what if the recipient does not need another argument? Perhaps they need a document. Perhaps they do not have the authority to decide. Or perhaps your proposal genuinely fails to solve their problem.
You've made a better case for an answer to a question nobody asked.
As I worked with my AI assistant, Seraphim, polished wording stopped being enough. I wanted it to notice what I might be missing: the other person's constraints, unsupported assumptions, the hidden workload in a simple request, and the gap between what I meant and what a recipient might actually read.
That work led to five distinct modes: Profiler, Profiler Plus, Dopamine Hooks, Reverse, and Rail.
Not five ways to apply more pressure. Five different operations for making communication clearer, more grounded, and more useful to the people involved.
For understanding and good-faith communication. Never for deception, coercion, or exploiting vulnerabilities. Respect the other person's right to disagree.
We didn't begin with a methodology. We began with what wasn't working.
The work was practical: correspondence with service providers and institutions, banking matters, messages to accountants, partners, and contractors, and instructions for team members. Often, the same task crossed languages and professional cultures.
The practices came before their names. I kept asking the assistant to consider the recipient's position, separate facts from guesses, provide evidence, and stop leaving people to reconstruct the meaning of a request themselves.
Eventually, repeating those corrections was not enough. I wanted them to become distinct, reusable ways of working.
We did not invent five tools in five perfectly timed flashes of insight. Their boundaries developed and overlapped through use. But the logic of the system is clear: understand the situation → prepare the communication → make its value visible → test how it may be received → remove unnecessary work from the next step.
Each new question revealed a limitation in the previous approach. Understanding alone does not produce a message someone can act on. A well-written message still needs checking. A checked message can still leave the recipient without a clear next step.
1. You're arguing with someone's personality. The real problem may be their authority.
Profiler: understand the situation before choosing your arguments.
The first requirement went deeper than "make the email more polite." It was: understand who we are communicating with, what this person is responsible for, and what is genuinely at stake.
How might they approach the task? What do they need in order to respond? What are they accountable for? Which objections are legitimate? What information is missing? Could my request look like an attempt to shift responsibility onto them?
This may involve a person's interests, possible concerns, role, usual approach to an argument, and potential next moves. But anything about another person's internal state must remain a hypothesis unless we have direct evidence.
Profiler is not there to confidently invent someone else's motives. It is there to stop us confusing our own interpretation with reality.
Suppose an employee says, "The bank doesn't allow this." You could assume they simply don't want to help and start pushing harder. Or you could first establish what is actually known: Is there a confirmed rule? Does it apply to this situation? Who has the authority to give a definitive answer?
The recipient is not the only subject of scrutiny. My interpretation of the recipient is being tested too.
The output should distinguish observable facts, plausible explanations, alternatives, important unknowns, and the question most likely to clarify them. Not a psychological verdict, but a working map of the situation.
This is not a diagnostic tool or permission to target vulnerabilities. If we suspect someone is concerned about responsibility, the useful response is to clarify the limits of their role, not to amplify their fear.
But then another problem emerged: I could get an excellent analysis and still have no message ready to send.
2. Understanding someone doesn't mean you've explained the task.
Profiler Plus: turn understanding into communication someone can act on.
We needed to separate two things: analysis for ourselves and an actual deliverable for the recipient.
One practical need illustrated that distinction particularly well: giving someone an effective way to communicate with Thai accountants and auditors. Not simply translating Russian words into Thai, but preserving the intention, the question, the relevant responsibility, and the answer required.
We began expressing the distinction this way: Profiler helps us understand the situation. Profiler Plus helps us prepare the message to send.
"Plus" does not mean a more powerful form of mind-reading. It means moving from analysis to action.
The internal assessment informs the language, tone, level of directness, length, order of arguments, and precise request. The assessment itself does not need to be sent. The recipient has no reason to read our guesses about their status or concerns. What they need is clear, honest communication.
For business messages, we began using a consistent set of questions: What is the situation? What is the goal? What is the problem? What action is needed? Who is responsible? What supports the request? What response do we need? What happens next? These are not mandatory headings in every email. They are the things we should not forget.
A vague instruction such as "Please check everything with the accountant" becomes an actionable request: what to confirm, which documents to present, how to report the answer, and what must remain unresolved until the facts are available.
Working with institutions brought an especially important principle into focus: put the evidence where the question arises. A stack of attachments is not the same as a clear connection between a claim and the document supporting it.
Precision is not the same as softness. Profiler Plus can help prepare a firm demand, a refusal, or a reasoned objection. Its purpose is to remove confusion and unjustified pressure, not to dissolve the sender's interests in politeness.
The outcome may be a ready-to-send message, a conversation outline, or the structure of a supporting document package. You do not need to invoke Profiler separately every time. The necessary analysis can happen within the task itself.
Yet another limitation appeared: a message could be entirely correct and still give the reader little reason to engage with it.
3. You've sparked curiosity. Now deliver the answer.
Dopamine Hooks: make the value visible, then complete the thought.
Another requirement emerged in our work: a strong opening, meaningful emphasis, and a useful discovery within each major section. Not merely conveying information, but showing the reader why it deserves attention.
Compare two headlines:
Five Tools for Working with AI.
And:
AI Can Make You More Persuasive. Even When You're Wrong.
The first names a topic. The second opens a contradiction that this article sets out to explain. But that second headline is justified only if the article actually shows how the error happens and what to do about it.
Over time, this mode became about more than openings. What mattered was a sequence of small, meaningful completions: a question is raised and an explanation follows; a difficulty is identified and a practical response appears; an action is requested and its purpose is clear.
In one banking interaction, I watched an employee read through a three-page document we had prepared and ask no additional questions. That is a personal observation, not a controlled experiment or proof that our method caused the outcome. It does, however, illustrate the problem I wanted to solve: help a person make sense of information in order, rather than force them to assemble the answer from scattered fragments.
Capturing attention is not enough. Every section must earn the reader's decision to continue.
"Dopamine Hooks" is our original working name. In this methodology it refers to structuring attention and providing meaningful closure. We do not claim to measure dopamine, control brain chemistry, or offer a scientifically validated neurobiological effect.
The output might be a stronger opening, clearer emphasis, explanations that actually resolve the questions they raise, and an appropriate next step. It should not rely on manufactured urgency, fake scarcity, or endlessly withholding the promised answer to keep someone engaged.
But compelling structure introduced a further question: how do we know we haven't become too impressed with our own presentation?
4. You know what you meant. The recipient only sees what you wrote.
Reverse: test the message from the other side.
We started asking AI to read a finished message through the recipient's eyes. At different points we called this Recipient Mode and Inbox Test. Later, I chose the name Reverse.
The distinction matters. Profiler examines a situation and possible explanations for behavior. Profiler Plus prepares the message. Reverse asks whether the message holds up when viewed from the other side.
What will someone understand from the opening paragraph? Which claim looks unsupported? Where might they ask, "Why am I the one being asked to do this?" Has an important condition been omitted? Is the requested response clear?
Take a hypothetical example:
Everything is ready. We just need your confirmation.
To the writer, this may seem like a convenient closing line. But what is the recipient supposed to confirm: the scope, the price, or the order itself? "Just" implies that agreement is a formality, even when nothing has been agreed.
Reverse should connect that concern to the actual wording. It should not declare, "The recipient will definitely feel pressured." A more defensible observation is: "This could be read as pressure because you are presenting a decision that hasn't been agreed as a mere formality."
The fix might be simple: name what needs to be agreed, distinguish it from placing an order, and leave room for changes.
The result is a review of the first impression, likely questions, possible sources of doubt, ambiguous passages, and a concrete revision. When useful, the same material can be reviewed from several positions: a specialist, a client, a manager, a journalist, or a skeptical reader.
Sometimes Reverse improves the wording. Sometimes it finds that the underlying proposal lacks sufficient justification. I consider the second outcome not a failure of the assistant, but one of its most valuable capabilities.
This kind of review does not replace actual feedback. The same model can miss its own error, and a real person may read the message differently.
Even after that check, another question remained: the message is clear, but is it actually clear what happens next?
5. You're asking for one answer. You've handed someone an entire project.
Rail: stop making other people finish the work you left undone.
At one point I stated the requirement in deliberately blunt terms: make it so that "the person doesn't have to think at all."
For a public methodology, an important distinction is necessary: not to take away someone's judgment, but to stop making them do the sender's organizational thinking.
Find the correct file. Work out which version is current. Reconstruct a sequence of events. Calculate an amount. Figure out whom to ask. Guess what would count as a complete answer.
All of that can be hidden inside a short request such as "Please clarify." On the surface, you asked for one action. In practice, you handed the recipient an unfinished process to organize.
We separated this additional layer under the working name Cognitive Rail, or simply Rail.
In one assignment, someone needed to collect information about a specific equipment consumable. I asked for a table without empty, unnecessary columns, containing named companies, contact details, and the exact information needed. The person's role was to gather facts. The technical decision belonged to another responsible person.
What helps here is not more persuasive wording, but a prepared route: whom to contact, what to ask, where to record the answer, what to do if information is unavailable, and who should receive the result. When appropriate, it should also be explicit about what the person must not decide or confirm on their own.
Rail makes the next step clear. It does not make agreement compulsory.
The output might be an email, an instruction, a table, a set of files, or a verification sequence. If there is no answer, there is a path for clarification. If authority is insufficient, there is a path for escalation. If the proposal does not work, refusal remains an option.
This is not a rail toward the one outcome that benefits us. It is a route through information that does not require the recipient to design the entire process from scratch.
Five modes should not become a mandatory assembly line.
They have a common purpose, but they should not all switch on automatically every time someone asks a question.
If the situation is unclear, use Profiler. If a message must be prepared, use Profiler Plus. If a valuable idea is getting lost in the presentation, use Dopamine Hooks. If the material is already drafted, use Reverse. If the message is understandable but the task remains difficult to carry out, use Rail.
The general rules apply throughout: do not fabricate facts, make uncertainty visible, retain material conditions, and do not replace the recipient's right to decide with your own desire for agreement.
The AI's personality doesn't change. The operation we ask it to perform does.
Anyone can repeat five names. The difference is in how the work is designed.
I do not claim to have invented the idea of considering another person's perspective, analyzing interests, or reviewing a message from a different point of view.
My contribution is a specific architecture for working with personal AI: five distinct operations, expectations for their outputs, limits on their use, and ways to detect mistakes.
The most important part is the ability to change not only the answer, but the task itself. Instead of arguing about someone's personality, examine their authority. Instead of strengthening an argument, identify missing evidence. Instead of writing an elegant request, make the action genuinely achievable.
AI should be able not only to complete my task, but to help me see when I've framed the task incorrectly.
That structure is what I consider distinctive about the approach. It is not a secret phrase or five artificial personalities. It is a coherent system in which understanding, preparation, attention, review, and action remain separate, complementary operations.
For me, the important shift is that a useful practice no longer depends solely on whether I remember to ask for it in the next conversation. Experience can be described, shared, tested, and corrected. It can then be adapted to a person's own tasks, languages, responsibilities, and way of working.
That is why this methodology begins with an explanation here on prudchenko.com. I also envision a public specification on GitHub and a plugin containing the five independent modes. Those are intended parts of the project, not a claim that they are already available.
Alongside the descriptions, I want the examples, limitations, and evaluations to be open, including unsuccessful results. One useful test is to compare a standard request with the relevant protocol on the same task, and assess factual accuracy, clarity, sufficient justification, and the absence of coercive pressure. Winning someone's agreement does not, by itself, prove that the method worked well. Nor can my confidence as its author substitute for independent evaluation.
Seraphim is the name of my personal AI assistant, not a foundation model I have developed. Selection of the material and responsibility for publication remain mine.
This article is itself shaped using the five modes. Its headings should offer genuine insights rather than suggest hidden powers. Its examples should reveal how the reasoning works. Reverse should challenge the definitions, arguments, and the author's own warning. This is an editorial process, not independent scientific validation.
Interested in applying this approach? Start with the problem, not a library of prompts.
For an executive, that might mean preparing for negotiations. For a team, clearer instructions and better correspondence. For a specialist, requests that account for authority, evidence, and the sequence of decisions.
If you would like to discuss adapting these protocols to your work, tell me what kind of problem you face, whom you need to communicate with, and where your current process breaks down. A general description is enough for an initial conversation. Please do not send confidential correspondence or client documents at this stage.
The right starting point is not installing all five modes. It is identifying which operations are useful for your situation and testing them against the actual task.
A person is not an obstacle standing between you and a "yes."
I sincerely ask you, from the bottom of my heart, to use these methods only for good.
In my own experience, abusing such techniques can turn against the person using them. But honest treatment matters for more than the consequences you may face. Another person deserves respect whether or not they can recognize pressure and protect themselves.
Please do not mistake this methodology for NLP, mind-reading, the "programming" of people, or a guarantee of a desired reaction. Our assumptions may be wrong. Persuasiveness does not give anyone the right to conceal material facts. A refusal is not an obstacle to be quietly bypassed.
The responsibility for good-faith communication lies with the person using the method, not with the recipient's ability to resist it.
Warnings alone cannot prevent misuse. That is why limits must be reflected in the instructions, examples, and evaluations, not just in an appeal from the author.
I want readers to leave these protocols thinking not "Now I know how to control people," but "Now I better understand how to communicate, and where I must stop."
Strong communication helps someone make a decision. It should never take that decision away from them.
Eugene Prudchenko