Finding virtual assistant clients when you have no portfolio yet
The first assistant client almost never arrives from an advert; it arrives from someone who either knows you or knows someone who knows you. What works when you have no portfolio is this: finish one small, defined job for a real business and come away with one written sentence from them.
- Lesson 5 of 6
- Beginner
- Free, no signup
The route from no portfolio to a steady book
Every station depends on one real person rather than on a number of messages. That is what stops this route from going faster under pressure.
-
1
You still have no portfolio
What works here is not building a profile or a CV, it is picking one business you already know
-
2
One small job, closed
Under a week, an output that can be shown, and a failure that costs the client nothing
-
3
One written sentence from the client
Asked for on delivery day, with one specific question, and permission to publish taken at the same moment
-
4
A referral from that same client
Asked with a concrete picture rather than with if you know anyone
-
5
A steady book with several clients
Several, not one. A single large client is not a steady book, it is employment without the protections
We have no timeline for any of these stations and will not invent one. People clear the second station at wildly different speeds, and the main variable is not skill, it is how many businesses you already deal with.
Last checked: Facts and tool names in this lesson are re-checked against their sources on this date.
Where do I start when I have no portfolio?
From one closed loop. A portfolio in this job means one real piece of work that is finished and one person willing to write a sentence about it. Not a page full of skills, not a CV, not a freshly built profile with seven educational posts on it.
So the order is the reverse of what most people assume. Work first, portfolio second. And for that first job to exist, no stranger has to hire you; it is enough that a real business you already deal with hands you one small task. The dental clinic that never answers its Instagram messages about appointments, the shop whose invoices live on paper, the training centre whose student list is spread across three notebooks. These make the first client, because they already know you and their problem is small and visible.
A short definition of the work itself: a first small job is one that finishes in under a week, whose output can be shown, and whose failure costs the client nothing. Clearing a six month old inbox, entering two hundred rows into a spreadsheet, answering one week of a profile's messages. If a job does not meet those three conditions it is not your first job, even when it pays better.
The question always asked is whether that first job should be free. Our answer is no, with one exception: free work is right when its boundary is written down in advance and you get something specific in return. A three hour task, free, in exchange for one written sentence and permission to use the business name. Free work with no boundary does not turn into a portfolio, it turns into a relationship in which you are the one in debt.
If you are still not sure exactly what this job covers, the first lesson of this path opens up the task list and the boundaries, and we are not repeating them here.
Where are the assistant clients actually?
Four places, and each has its own cost. Which one works for you depends on what you are short of right now: trust, or simply numbers.
People you already know. The warmest route and the least competitive. The trust is already there and all you have to say is what you can take off their hands. The cost is that it is finite and it runs out; if you know fifty people, you can walk through this door fifty times and no more.
Businesses whose messages have got away from them. Shop profiles, clinics, training centres, the places that take orders in direct messages. They display their problem on their own public page, and that is exactly what makes them approachable: a comment three days unanswered, a question about price that never got a reply, a link in the bio that does not open. The cost is that it is cold, and most messages go unanswered.
Freelance platforms. Here the client came of their own accord and said what they want, so the buying intent is certain. In exchange, competition sets the price and a beginner stands next to profiles carrying thirty reviews. One thing said plainly, because we have a stake in it: RGB's own freelance platform has a virtual assistant and admin category, its commission is zero and it takes nothing out of your pay. But it is new and the number of open projects on it is small, so count it as one venue among several rather than the answer. We are the ones selling that venue, and you should hear this sentence from us.
Agencies and other assistants. The least talked about door and in practice the best one from your second client onward. An assistant whose capacity is full turns extra work away, and an agency that landed a large project farms out its admin. Someone who builds a good relationship with an experienced assistant gets work sent to them without going looking for it. The cost is that somebody has to have seen you work first, which means this door opens after the first job rather than before it.
One note about something not on this list: remote hiring groups on Telegram and advert channels. There is work there, but the ratio of messages to opportunities is very high and the fraud is concentrated in exactly those places. If you go there, the rule is simple: pay nothing for registration, for training, or for a guarantee. A real employer does not take money from an assistant.
The four places an assistant client comes from, and what each one costs
Choosing between them depends on which one you are short of: trust, or numbers.
-
People you know
The trust is already there. In exchange it is finite and runs out
-
Businesses buried in messages
Their problem is visible on the public page. In exchange it is cold and most messages go unanswered
-
Freelance platforms
The client came on their own and the buying intent is certain. In exchange competition sets the price
-
Agencies and other assistants
Work arrives without searching. In exchange this door only opens after the first job
We are not neutral about the third door, because we run a freelance platform ourselves. The lesson says so, and you should read it that way.
What do I write in the first message?
One specific thing you noticed, one sentence about what you do, and one small offer with a clear boundary. That is all. The first message is not the place for a CV and not the place for a list of skills.
The reason is simple: the person is not reading your message, they are deciding whether reading it is worth their time. What changes that decision is a first sentence proving you looked at their page. "Hello, I am a virtual assistant and I work across various fields" proves to nobody that you even know who they are.
One specimen, and we stress that it is only a specimen; copying it is precisely what this lesson tells you not to do:
Hello. Four people asked for prices under yesterday's post and did not\nget an answer. That is what I do: I answer a page's messages and\ncomments and keep the orders in one tidy list.\n\nIf you want, I will do the first week free. The condition is that we\nagree in advance what gets answered and what gets passed to you.\n\nOne question: do you quote the prices yourself, or is there a fixed\nlist I could answer from?Four things this message did that unanswered messages usually do not: it brought one specific, checkable observation, it defined the work in two lines, it set the boundary of the free offer itself, and it ended on a question that is easier to answer than to refuse.
And three things deliberately absent: a price, a CV, and a long courtesy. A price in the first message is always wrong, because you do not yet know how big the job is; a number said now either sells you cheap or ends the conversation on the spot. The next lesson in this path takes up pricing, and that is where this belongs.
One rule for messaging apps: send the message as one piece. Five messages in a row on WhatsApp or in direct messages is exactly the experience the client is running from, and the one you are meant to take away from them.
The message that gets a reply against the one that is not even read
In the message
- One specific observation from that business's own page
- The work defined in two lines, in the client's own words
- The boundary of the offer, written by you
- One question easier to answer than to refuse
Not in the message
- A CV and a list of skills
- A price, while the size of the job is still unknown
- Text equally true of two hundred other people
- Five messages in a row instead of one
This list is about the shape of the message, not about how many people reply. Even a correct message goes unanswered most of the time, and that is normal.
Does sending messages in bulk work?
No, and the reason is not only an ethical one. When you use one fixed text for two hundred businesses, two things happen: none of those two hundred messages carries a specific observation about its recipient, and you walk into territory that has rules written by the receiving side rather than by you.
For email this can be said precisely, because Google published it. Under Google's sender guidelines, anyone sending email to Gmail accounts must have SPF or DKIM on their sending domain, valid reverse DNS records, a TLS connection, and must keep the spam rate reported in Postmaster Tools below 0.3 percent. Google's own recommendation is stricter: stay below 0.10 percent and never reach 0.30 percent or higher. And if you send more than 5,000 messages a day to Gmail accounts, a heavier set of requirements applies on top, one-click unsubscribe among them.
What that means in practice for a beginner assistant: the spam complaint rate is set by the recipients, not by you, and the permitted share is far smaller than an irrelevant cold message can get past. Which is to say the strategy people call "just a numbers game" does not survive contact with the numbers Google itself published.
For Instagram and Telegram we have no published figure and we are not going to invent one. What we do know is this: identical cold messages to dozens of accounts is behaviour every platform recognises as spam, and the penalty is a restriction on the one account that is your only working tool.
Our position is simple: ten messages written for ten separate businesses bring in more clients than five hundred identical ones, and they keep your account alive as well. And the part that takes the time is not the writing, it is the looking, which is exactly what the AI section of this lesson is for.
How do I read silence, and when do I follow up?
Do not read silence as a no, because it usually is not one. A small business's messages are read by the same person who stands behind the counter, and yours slides down among thirty others. But do not read it as a yes and sit waiting either.
The rule that works: one follow up, after three or four working days, in the same thread, shorter than the first message and with no reproach in it. One sentence that reminds, one that leaves the door open. After that, stop. A second and third follow up does nothing except spend next month's chance, and if that business genuinely needs an assistant three months from now, they will remember you with a bad feeling attached.
And something engineers learn earlier than salespeople: sent is not arrived. An email that left your hands without an error may never have reached an inbox; it may be in a spam folder, or Instagram's message request filter may be holding it somewhere the account owner looks once a year. So before you draw a conclusion from one channel, try the same business through another: a message lost in the direct inbox arrives by email, or through a simple phone call.
What makes all of this manageable is a list: each business, the date of the first message, the channel, what you offered, the follow up date. A plain spreadsheet is enough. Without that list, two weeks later you know neither what you said to whom nor whether you said anything at all, and that is exactly where people message the same person twice.
From one small job to a steady book
The first job is done. Now there are two things to collect, and most people collect neither, because they are embarrassed to ask.
One written sentence. The moment matters: the same day you delivered and the client is happy, not two months later. And ask for it with a specific question rather than "leave a review if you were satisfied". The specific question is: "what were you worried about before I started, and how is it now?" The answer to that question is what the next client wants to read, because their worry is the same one.
Take permission to publish at the same moment and ask whether the business name can be used. A client who prefers not to be named is within their rights, and "the manager of a dental clinic in Tehran" is usable without a name. What is not allowed is inventing the sentence.
One referral, asked with a specific question. "Refer me if you know anyone" gets no answer, because it hands the thinking to the client. "Do you know anyone whose page messages have piled up the way yours had?" gets an answer, because it puts one definite picture in front of them and they start matching their acquaintances against it.
From here on, what grows is no longer the number of messages you send, it is the number of people who have seen you work. Three satisfied clients make five referrals and those five make two clients. This is not arithmetic and we have no figure for it; what we do know is that past a certain point the direction of flow reverses and instead of looking for work you start turning it down. Turning work down is part of the job too, and the first lesson covers capacity.
One warning is needed: a steady book with one large client is not a steady book. If eighty percent of your income comes from one person, you are that person's employee without any of the protections an employee has. The day they decide to bring the work in house, eighty percent of your income ends within a week.
The fast path, with AI
The model's job is not to write your message, it is to look. The slowest, hardest part of a cold message is finding the one specific thing lying on a business's public page that nobody picked up. Hand that over and write the sentence yourself.
- Pick ten businesses that are within reach and whose work you understand. Ten, not a hundred.
- For each, collect their own public text: the page description, the bio, the last twenty captions, the comments under recent posts, the site's contact page. Only what they published themselves.
- Run the recipe below on a fast, cheap model. This is mechanical work and a frontier model is wasted on it. Which model is the better pick for this today is kept up to date in this site's AI section.
- Open the output and verify each observation yourself on that business's page. Anything you did not see with your own eyes gets deleted, no exceptions.
- Now write by hand. The first sentence comes from the verified observation and the rest follows the shape given in this lesson. The model does not write the message, because that writing is exactly what the client judges you by.
Copy-ready recipe
The text below is a business's own public text, published by them.
{collected text}
Do three things and nothing else:
1) Pull out at most five "open loops". An open loop is something a person
asked for and did not get, or work that started and never finished. For
each one, quote the exact text that shows it. No quote, no entry.
2) Say how long each open loop has been open, but only if the date is in
the text. If there is no date, write "no date".
3) List everything that cannot be worked out from this text under the
heading "I do not know". For example how many people work there, or
where they record an order.
Rule: do not guess. If something is not in the text, its place is the
"I do not know" list. Write no message and no offer. I write those.Before you trust the output: No observation reaches the message without you seeing it yourself. The model can mix two businesses together or attribute a comment to a post it does not belong to, and naming a problem the business does not have ends the conversation in one line. There is a second rule: put nothing but their own public text into the input. A previous client's private message, a contact list, or anything they did not publish themselves does not belong here.
AI in this kind of work
In finding clients, AI is very useful on one side and harmful on the other. Useful in the mechanical work: reading ten businesses' public text, pulling out quotable details, keeping the follow up list tidy. Harmful exactly where the client is meant to get to know you through your writing. If a model wrote the first message, what the client becomes familiar with is not you.
Tools that actually help
- Claude Good for the looking stage, because when you explicitly say quote it and do not guess, it does that. Iran is on neither of Anthropic's two supported country lists, which we read on their own page rather than measuring.
- Gemini Suitable for Persian caption and comment text, and Persian is on its web app language list. Google writes that the app runs in over 230 countries and territories, and Iran is not on that list.
- ChatGPT Does the same job, and for this stage there is no meaningful difference between them. OpenAI publishes the list of countries that have access to its services, Iran is not on it, and that same page states that access from outside the list may result in an account being blocked or suspended.
Where it backfires
The risk here is not that somebody detects AI text; it is that you put something into the input that is not yours. A page's public text is fine, but private messages, phone lists and a previous client's information are a different matter. Anthropic's own privacy page says that for consumer products, conversations are used to improve models when the user has allowed it, when a conversation is flagged for safety review, or on explicit opt in. Which means the honest answer is "it depends on a setting you have never looked at". Until you have seen that setting, assume anything you paste may stay. The security basics lesson opens the same boundary from the access side.
Sources: Anthropic supported countries Anthropic privacy: is my data used for model training Google: Gemini app countries and languages OpenAI supported countries and territories
Where this advice stops
This lesson is about the Iranian market and assistance in Persian. For a foreign client, two separate problems are added that this page solves neither of: writing English well enough that it is itself the gate, and getting money into Iran. None of the five stations carries a timeline either; someone with fifty businesses in their circle and someone who just moved to a new city are on completely different routes, and we have no figure for either. And to say the conflict of interest again: we run a freelance platform, so we are not neutral in recommending that venue.
From our own work
On our own freelance platform, the proposals table has a unique key on the project and user pair (UNIQUE KEY project_user), which makes submitting a bid an update: when a freelancer changes their amount, the previous amount and text are erased for good. That is exactly why we built a separate, append only table for proposal events that keeps every version as it was. And the comment at the top of that same file carries another line that bears directly on this lesson: handing a message to the mailer is not the same as it arriving in an inbox, because wp_mail only tells you no error occurred. Both points are one lesson for someone looking for clients: keep your own record, and never conclude from an error free send that anything was read.
Real follow-up questions
How long does it take for the first client to arrive?
We have no figure for it and will not invent one, because the main variable is not skill: it is how many businesses you already have access to. Someone who grew up around a local market may land the first job in two weeks, and someone who knows nobody may take months. What shortens the time in both cases is the same: instead of preparing, finish one small real job.
To start, should I build a CV or a work profile?
Neither, until you have finished one real job. A CV without work is a list of claims and a profile without work is a list of educational posts. After the first job, one page you control is enough: what you do, who you did it for, and that one sentence the client wrote. That does more than any CV.
The client says do a free sample first. Should I accept?
If you write the boundary in advance, yes; otherwise no. One small defined task with defined hours, once, in exchange for something worth having: a written sentence and permission to use the name. What gets refused is this: free work with no ceiling, or work whose output is itself the final product, after which the client no longer needs you.