As a freelancer you find new clients through five routes: referrals from satisfied clients, your existing network, visibility through your own content, direct outreach to businesses that fit you, and platforms such as directories or marketplaces. Referrals and network produce a result fastest, but you cannot multiply them on demand. Content and visibility take the longest, yet they stay put and keep working. Direct outreach is the one route where you set the number of first contacts yourself, and at the same time the one with the tightest legal limits. Platforms deliver enquiries quickly, but you are building on someone else's ground. The choice does not depend on which channel is currently being recommended, it depends on one question: is your offer actively searched for, or do you have to explain it first?
I run my business as a solo freelancer and I use a dedicated AI employee for the groundwork of finding clients: Olaf screens towns for businesses with an outdated website and hands me a rated list. The contacting is still me, and this post is about the routes first, the tool second. Which other roles run in my daily office work is in AI for Freelancers: What Pays Off.
The five routes at a glance
The ranking below is my working basis, not a statistic. It helps with the question of where to start when you cannot do everything at once.
None of these routes is the one right answer. Most freelancers I talk to are looking for exactly that: the one channel that works. The better question is in the next section.
Is your offer searched for, or does it need explaining?
This is the switch everything else runs through. Some offers are actively searched for, because people know the problem and have a name for it: a new heating system, a tax return, a website relaunch. And some offers address a problem the client has not yet put into words, because they do not know a solution exists.
If your offer is searched for, you work with the search: content, a findable profile, an entry where people look, and for the question of need rather than the job title, as the insurance broker example shows. You do not have to convince anyone that they have a problem, you only have to be found when they search for it. Platforms and directories work for the same reason, they sit between you and demand that already exists.
If your offer needs explaining first, search runs into nothing, because nobody types the term you want to own. What remains are the routes where a person makes the first move: referral, network and direct outreach. That is why so many visibility projects fail, not because the content is poor, but because it was built for an offer that has no demand yet.
A first, qualitative test is quick: type the words your client would use to describe their problem into a search engine. If offers like yours come back, you are probably being searched for. If nothing fits, or only ads for something entirely different, you have to explain. That proves no search volume, though; results show existing pages and ads. Before you build a channel on it, get numbers: the queries in Search Console, keyword data, and above all the words your last clients used for their problem.
Referral and network: fast, but not on demand
In my experience referrals have the highest close rate of any route, because the trust exists before the first conversation. The catch is the control: you cannot decide that three referrals will arrive next week.
What you can control is the asking. Most freelancers never actively ask for a referral, even though the best moment is easy to name: right after a project has been completed well, in the trades at handover or after the final invoice is paid. Two sentences are enough, and they work better when they are specific. Not "do you know anyone", but "do you know anyone who has exactly this coming up right now".
Network is the slower sibling of the same logic. What counts here is not the number of contacts, but how many people could say in one sentence what you do. If your own network cannot, that is not a network problem, it is an offer problem.
Content and visibility: the slowest route, and the one that stays
Content is the only route that keeps working without you. A post that answers a real question brings enquiries while you are busy with something else. The price is time, and more of it than most people plan for.
Google itself puts it carefully in its own search documentation: some changes take effect within a few hours, others can take several months, and you should generally wait a few weeks before assessing whether your work had a beneficial effect (as of September 2026). Anyone who gives up after three weeks did not test this route, they aborted it.
In return, something stacks up here that happens on no other channel: every post stays. The tenth post does not delete the first, it adds to it. And content supports the other routes too, because someone who hears about you through a referral will first check what they can find about you.
Direct outreach: what is allowed and what is not
With direct outreach you decide yourself how many businesses get approached this week, without waiting for anyone. In Germany, however, it is not open ground, it is regulated in section 7 of the Act Against Unfair Competition (Gesetz gegen den unlauteren Wettbewerb, UWG). The statute states that an unreasonable nuisance is always to be assumed, and the original wording reads:
"1. bei Werbung mit einem Telefonanruf gegenüber einem Verbraucher ohne dessen vorherige ausdrückliche Einwilligung oder gegenüber einem sonstigen Marktteilnehmer ohne dessen zumindest mutmaßliche Einwilligung,
- bei Werbung unter Verwendung einer automatischen Anrufmaschine, eines Faxgerätes oder elektronischer Post, ohne dass eine vorherige ausdrückliche Einwilligung des Adressaten vorliegt"
In plain terms: advertising by telephone call to a consumer without their prior express consent, or to any other market participant without at least their presumed consent, counts as an unreasonable nuisance, and so does advertising by electronic mail without the addressee's prior express consent. So cold emails for advertising purposes are not permitted without prior express consent, towards private individuals and towards companies alike. For calls, the bar towards another business is presumed consent, meaning a concrete interest in that particular call that is recognisable from the circumstances. That is a narrow door, not an open one. I am not a lawyer and this is not legal advice, for an individual case read the statute or ask one.
The practical point is not fear of a letter from a law firm: the legal narrowness forces you into the behaviour that is more careful anyway. Volume is not an option, so precision is what is left. A hundred identical emails name no reason; ten businesses with a verifiable trigger each produce ten different opening sentences.
A trigger is something you can see from the outside and put into words: the website runs without HTTPS. The site is unusable on a phone, even though a large share of visits tends to come from phones. There is no reachable way to get in touch. The business has excellent reviews, but no page where anyone could read them. That is the difference between a researched list and a bought one.
A trigger does not make the outreach lawful, though. It is a quality criterion for the research; the contact route is a separate check. A visible website fault does not replace consent, and an address published in a legal notice is not marketing consent. On top of that, the list itself triggers data protection law as soon as a sole trader or a named contact appears in it: a legal basis under Art. 6 GDPR, the duty to inform under Art. 14, the right to object under Art. 21. How to keep the three checks apart in the workflow is in Automating Cold Outreach.
Platforms and directories: visible fast, someone else's house
Directories, marketplaces and matching platforms hand you enquiries from demand that already exists. That makes them attractive for the start, especially when your offer is searched for. The bill has three line items: fees or commission, price pressure through direct comparability, and the fact that the client relationship is not yours. If the platform changes its rules, your access to clients changes without you having done anything wrong.
My take: platforms are a good accelerator and a poor foundation. Use them to get early projects and reviews, and build something that belongs to you in parallel.
How Olaf does the groundwork for direct outreach
The part of direct outreach that really eats time is not the writing. It is the research beforehand: who do I approach at all, and why that business in particular? That is exactly the part I handed to an AI employee.
I give Olaf a town and an industry. He comes back with a rated list of local businesses whose website is outdated: every entry with a reason, evidence and an A, B or C rating, plus a ready briefing for the website build. He has a fixed tool chain for that: location research via the Google Places API, a technical quick check per website (mobile-friendly, HTTPS, age of the site builder, a reachable contact address) and screenshots of the site, plus mobile ones in the deep check, because he only judges a design once he has looked at the image, never from a metric alone. The AI opt-out in robots.txt he adds by hand afterwards; the tool does not check it, and in his scoring matrix it is a hard exclusion criterion.
The hard rule that separates him from a bought list: without a solid contact address from an imprint or contact page there is no lead, only a C rating. Addresses are never guessed. Businesses with no website at all are not leads, because there is nothing to compare. And he contacts no one, no emails, no calls. He researches, I reach out.
How strict that is shows in the numbers from my own lead runs in July 2026: 198 candidates screened across eight assignments. In the seven local assignments with 20 candidates each, that typically turned into 3 to 5 real leads; the eighth was a differently built community list. The rest fails the criteria. On top of that come the false alarms Olaf discards himself: one site with outdated tech underneath but a modern design, another with an old copyright year in the footer but otherwise built to current standards. Both were marked as false alarms instead of counted as leads. In the first cross-check by my website colleague 4 of 10 briefings had errors, and since then every lead is checked a second time.
That is the limit of automated customer acquisition: the research can be automated, the cold approach cannot. What comes out is a short list where you can say, for every name on it, why that business has a reason right now. More examples of roles like this are in AI Agents: 10 Real-World Examples, the frame for rebuilding it on your own criteria in AI Employees for Consultants and Agencies.
Frequently asked questions
How do I find new clients fastest as a freelancer?
Through referrals from completed projects and your existing network, because the trust is already there and nothing has to be built up first. The fastest concrete step: ask your last satisfied clients specifically whether they know someone who has exactly this coming up.
What does "finding leads" actually mean?
A lead is a potential client where two things are settled: they fit your offer, and you have a solid way to reach them. An address without a fit is not a lead, and a fitting business with no reachable contact is not one either. Finding leads means checking both at once.
Is cold emailing allowed in Germany?
Advertising by electronic mail without the addressee's prior express consent counts as an unreasonable nuisance under section 7(2) no. 2 UWG, towards consumers and companies alike. For calls to companies, the narrower exception of presumed consent applies. This is not legal advice, when in doubt have the individual case checked.
Can AI take over client acquisition?
The research yes, the outreach no. AI can screen publicly visible businesses against fixed criteria, rate them and evidence the trigger. The contact and the conversation stay with you, not least because the legal conditions for advertising hinge on a consent no software can manufacture.
How long does it take for content to bring in new clients?
Longer than any other route. Google itself writes that some changes take hours and others several months to have an effect, and that you should wait a few weeks before judging whether something worked. So plan content alongside a route that delivers sooner, not as your only channel.
Should I focus on one channel or run several in parallel?
Two is sensible, and they should differ in speed: one that delivers short term (referral, network or direct outreach) and one that carries long term (content). More than two at once usually means that none of them gets taken far enough in a one-person business.
How to continue
Answer the switch question first: is your offer searched for, or do you have to explain it? Then pick exactly two routes, a fast one and a slow one, and leave the other three alone until those two are running. If direct outreach is one of them, start with the research, not with the message: a short list with an evidenced trigger per business beats any long list without one. Setting the criteria for your own industry is where most people stall. If you would rather not decide them alone, put the question into a call or a post; there are people there who have already built the same list for their trade.
How I handed that research to an AI employee is on the page about Olaf, my lead scout. The ready-made template for it, with tool chain and scoring matrix to rebuild, is in my community Claude Practitioners.
