AI for cold outreach: better prepared, not more spam

The German reality: why US-style outbound doesn’t transfer

Many AI sales tools come from the US, and so do their playbooks: pull thousands of contacts from a database, personalise with AI, send in sequences, follow up, scale. In the US, cold email to businesses is largely allowed as long as you follow opt-out and identification rules. Germany works on a different principle: consent first, then marketing.

That doesn’t make AI useless for outbound. It shifts what you use it for: less for sending, more for understanding. The strongest uses are research, finding a reason to reach out, preparing conversations and sorting replies, everything that makes a single, well-prepared contact better.

What’s allowed: consent, GDPR and LinkedIn

Sources: § 7 UWG, LinkedIn User Agreement, section 8.2. Austria, Switzerland and other countries have their own rules; check them for your market.

GDPR: data for B2B outreach

If you collect contact data from the web or enrich it with AI tools, you’re processing personal data. For B2B outreach, companies usually rely on legitimate interests under Art. 6(1)(f) GDPR. That isn’t a blank cheque: you have to weigh interests, inform the people concerned (Art. 14) and honour objections (Art. 21). And legitimate interests don’t replace the consent that Germany’s UWG requires for marketing email: both rules apply side by side.

Outside Germany

In the US, the CAN-SPAM Act allows commercial email without prior consent, but requires accurate headers, a physical address and a working opt-out. If you reach people in Germany, the stricter German rules apply.

AI disclosure

When an AI system itself communicates with people, such as a voice or chat agent, Article 50 of the EU AI Act requires since August 2, 2026 that people are told they’re dealing with an AI. An email you write with AI help and send yourself isn’t covered in the same way, but honesty still pays.

Where AI really helps in outbound

Watch the limit: AI can invent facts about a company or present outdated information as current. Check every claim that ends up in a message against its source.

Allowed routes to a first contact

If cold email is off the table, these routes still work in Germany, and AI makes them much more efficient:

1. A targeted call to a business where there’s a concrete link to what they do (presumed consent). AI prepares: who’s responsible, what’s the reason, which question opens the conversation?
2. LinkedIn by hand: a connection request with one sentence on why this person, no pitch. AI helps you research and phrase it; you send it yourself.
3. Warm contacts: former customers, people you met at events, referrals. AI finds who in your CRM hasn’t heard from you in a while.
4. Inbound instead of outbound: lead magnets, newsletters, tools like a free check that people sign up for voluntarily. More in our guide to AI lead generation.
5. Existing customers: follow-ups within § 7(3) UWG, with an opt-out notice in every email.

Personalisation that doesn’t sound like AI

Most AI-personalised messages are easy to spot: a compliment on the latest post, an invented connection, three paragraphs of pitch. People, and increasingly the AI assistants that pre-sort inboxes, filter out exactly that.

• A real reason: why you’re getting in touch now, backed up and checkable.
• A question instead of a pitch: what’s hard for this person right now?
• Short: three to five sentences. If AI writes more, cut.
• No fake familiarity: no “I read your post with great interest” if you didn’t.
• Your own voice: give the AI two or three of your real messages as a model.

My rule of thumb: when personal AI agents pre-sort email and requests, the agent decides what the person sees at all. Generic mass mail gets filtered out first. A clear, sourced reason is readable for both human and agent.

Deliverability: the technical basics for legitimate email

Even legitimate email, to existing customers or after consent, lands in spam quickly if the setup is wrong. Google requires senders to Gmail to use SPF or DKIM, among other things, and bulk senders additionally need DMARC and one-click unsubscribe.

Source: Google Workspace Admin Help: email sender guidelines.

Technical setup doesn’t replace consent. It just makes sure legitimate email arrives.

Tools at a glance

Before you buy, check whether a tool sends automated email to bought lists or automated LinkedIn messages: neither fits the German rules above.

“Try” links are affiliate links (advertising): I earn a commission at no extra cost to you.

Checklist before the first contact

• Channel allowed? Email only with consent or the existing-customer exception; calls to businesses only with a concrete link.
• Data clean? Origin documented, Art. 14 GDPR notice ready, objections honoured.
• Reason sourced? Every claim about the company checked at the source.
• A person sends it? No automated LinkedIn messages, no mass email.
• Short and honest? No invented compliments, no promises.
• Opt-out and imprint? In every email.
• Replies handled? Every reply gets a personal response.

Metrics that keep you honest

An outreach block with AI, step by step

Here’s what a fixed two-hour weekly outreach block can look like when AI handles the preparation and you have the conversations. It suits small teams and solo founders who sell alongside day-to-day work.

The block is small on purpose. Five well-prepared contacts a week add up to around 60 conversations or attempts over a quarter, each prepared well enough not to embarrass you. If you want more, extend the block, not the mailing list.

Handling replies properly

• Interested: reply personally within one working day, with a concrete time suggestion.
• Question: answer the question, don’t repeat the pitch.
• Later: note the date in your CRM and come back with a new reason, not “just following up”.
• Not interested or objection: say thanks, don’t contact again, record the objection.
• Complaint: take it seriously, disclose where you got the data, delete it if asked.

AI genuinely helps with sorting because it quickly separates out-of-office notes, refusals and real replies. You write the reply yourself: anyone who bothers to answer can tell whether a person or a template is writing back.

AI agents that send on their own: be careful

New sales agents promise to run the whole outreach process alone: find contacts, write messages, send, follow up, book meetings. Technically it works. In Germany it hits the same limits as any mass email, just faster and at greater scale.

• The rules apply to agents too: marketing email without consent stays unlawful whether a person or an agent sends it. You’re responsible.
• Voice agents: when an AI itself talks to people, Article 50 of the EU AI Act requires since August 2, 2026 that they’re told. Cold calls to consumers stay off-limits without consent.
• Loss of control: an agent that writes without sign-off can invent facts, contact the wrong people or miss objections.
• Sensible use: agents that research, suggest and put drafts up for approval. The decision to send stays with you.

My rule of thumb: if you use a vendor’s AI features, you give up control and keep the liability. Read the terms first, then give the agent a clearly scoped job that ends with your sign-off.

12 prompts for outreach with AI

These prompts help with preparation and wording. You decide what goes out, and you send it yourself.

Here are my five best customers with sector, size and why they bought: [list]. Derive a target account profile: traits, buying triggers, typical objections. Flag assumptions I need to check.

Summarise what this website and these news items say about [company]: [texts/links]. Business model, current topics, possible challenges. Cite the source for every claim; invent nothing.

Here’s current information about [company]: [job ads, press releases]. Which would be an honest reason to talk about [my offer]? Rate each: strong, weak, unsuitable, with reasons.

I’m calling [role] at [company], reason: [reason]. Write a two-sentence opener, three open questions, and the three most likely objections with an honest answer. No pressure.

Write a LinkedIn connection request to [person, role], max 250 characters. One sentence on why this person, no pitch, no compliment I can’t back up. I’ll send it myself.

Cut this message to five sentences max: [text]. Keep the reason and one question; remove filler, superlatives and anything that sounds like a mail merge.

Write an email to an existing customer who bought [product], about [similar product of ours]. Max 120 words, one link, ending with a note that they can opt out of such emails at any time.

Sort these replies into: interested, question, later, not interested, unsubscribe, out of office: [replies]. Suggest a next step for each interested reply; I’ll answer myself.

List the ten most common objections to [offer] among [audience] and write a short, honest answer to each, with no pressure and no promises we can’t keep.

Check this outreach plan for risks under Germany’s § 7 UWG and the GDPR: [channel, audience, data source, copy]. Where is consent or a notice missing? A first pass, not legal advice.

Here are my notes from the call with [company]: [notes]. Summarise need, decision-makers, timeline, next step, open questions. Format it for my CRM.

My audience is [audience], my offer [offer]. Suggest three lead magnets or small tools that prospects would sign up for voluntarily with their email, with effort and expected benefit.

The most common mistakes

1. Copying US playbooks one to one. What’s allowed in the US can bring a cease-and-desist letter in Germany.
2. Treating B2B as a free pass. In Germany, marketing email to businesses also needs prior express consent.
3. Automating LinkedIn. Bots and automated messages breach the User Agreement; your account is at stake.
4. Using AI facts unchecked. One invented claim about the company ends the conversation before it starts.
5. Measuring volume instead of preparation. A hundred messages without a reply are worse than five conversations.
6. Auto-answering replies. Anyone who replies deserves a personal response.