Quick answer
Expandi's H2 2026 report found AI-drafted LinkedIn messages losing to human-written ones on acceptance rate by about 12%, within the same accounts, with reply rate coming out roughly flat. That data has no Nordic breakout, so I can't honestly tell you the gap is worse in Sweden, Norway, Denmark or Finland specifically, nobody has measured that. What I can tell you is that Hofstede's own cross-cultural research scores those four countries among the least self-promotional in the world, and Janteloven's discomfort with visible selling is a real, documented norm there. The mechanism behind the 12% gap, confident, hype-forward copy, is exactly what these markets already discount before an AI tool ever touches it.
The claim doing the rounds
I keep seeing a version of this argument in outbound circles: AI-personalized outreach is failing harder in the Nordics than anywhere else. It's a plausible story, Nordic buyers have a reputation for distrusting a hard sell, and it gets repeated with enough confidence that it starts to sound like a finding rather than a hunch. I went looking for the study behind it. It doesn't exist, at least not as a regional breakout. What does exist is a real, global stat about AI-drafted messages underperforming, and a separate, well-documented body of research on why Nordic business culture in particular has less tolerance for the kind of copy that stat describes. Those two things support each other. They are not the same thing, and conflating them is the exact kind of sloppy sourcing I try to avoid on this blog.
I'm Hlib Storchak. I build and run outbound systems for B2B founders and sales teams, including several selling from the Baltics and further south into Sweden, Finland, Norway and Denmark, and I've booked 2000+ meetings for B2B clients doing it. This is my attempt to separate what the data says from what's a reasonable inference, and to give you something usable either way.
What Expandi's H2 2026 report actually measured
Expandi's State of LinkedIn Outreach H2 2026 report ran a within-account comparison across 118 campaigns from 73 accounts, October 2025 through June 2026. The design matters here: a naive comparison of "accounts using AI" against "accounts not using AI" is confounded, because the operators who adopt AI tools first tend to be stronger operators across the board, so AI looks better than it is. Expandi controlled for that by comparing each account's own AI-generated campaigns against that same account's human-written campaigns. Within that same-account comparison, human-written messages beat AI-generated ones on acceptance rate by about 12%. Reply rate came out effectively identical either way.
That's a real, specific, well-designed finding, and I go into the full breakdown, including the cross-account numbers that made AI look better before the confound was controlled for, in the case against AI-personalized LinkedIn outreach. What it is not, is a study of the Nordics.
The gap: there's no Nordic breakout in that data
I want to be direct about this because it's the part most retellings of the stat skip. Expandi's report doesn't segment its acceptance-rate findings by region. It's a global figure drawn from its own user base, however that base happens to be distributed. There is no published number that says "the 12% gap is 20% in Sweden and 5% in the US." Anyone telling you that is either citing a study I haven't been able to find, or inferring it, same as I'm about to.
The honest version of this article isn't "AI personalization is measurably worse in the Nordics." It's "here's a real global stat about AI-drafted copy underperforming, and here's why the specific way it tends to underperform, overconfident, hype-forward tone, should land worse in markets that already discount that tone in every other context." That's an inference built on two separate, real sources, not a single study that already did the work for me.
Tip. When a stat gets a regional or industry gloss added to it in casual retelling ("worse in the Nordics," "kills it in SaaS"), check the source study for that segmentation before you repeat the gloss. Most of the time the segmentation was added afterward by someone else's inference, not the original researchers'.
Why tone carries more weight in these four markets
Geert Hofstede's cross-cultural research scores countries on a Masculinity dimension, despite the dated name, it measures how much a culture rewards visible achievement, competition and self-promotion versus consensus, modesty and quality of life. Low scores mean a "strongly cooperation-oriented culture," in Hofstede's own framing. I checked the current scores directly on Hofstede Insights' own comparison tool: Sweden scores 5, Norway 8, Denmark 16, and Finland 26, out of 100. For context, that puts all four among the lowest, most cooperation-oriented scores measured anywhere, nowhere close to markets like the US or most of Southern Europe.
| Country | Hofstede Masculinity score | What it signals |
|---|---|---|
| Sweden | 5 | Lowest measured score in Hofstede's full dataset; consensus and modesty valued over standing out |
| Norway | 8 | "Strongly cooperation-oriented," per Hofstede Insights' own country page |
| Denmark | 16 | Same cooperation-oriented profile; quality of life over competition |
| Finland | 26 | Still firmly on the cooperation-oriented side, the highest of the four but low overall |
None of that is a persuasion study. It's a cultural-values framework, and I'd treat it the way I'd treat any single research model, as a real, useful signal rather than a law. But it does explain, with actual data behind it rather than a stereotype, why a message built around bold claims and visible confidence starts from a deficit in these four markets specifically, before an AI tool ever drafts a word of it.
Janteloven and the discomfort with self-promotion
The cultural mechanism behind those scores has a name across the region: Janteloven, or the Law of Jante, a set of informal social norms, most famously codified by the Danish-Norwegian author Aksel Sandemose, that boil down to "don't act like you're better than anyone else." In practice, that shows up as understatement of credentials, discomfort with visible boasting, and a default suspicion toward anyone who seems to be selling themselves too hard. A founder who's built something genuinely impressive will usually describe it in the most modest terms available and credit the team, not the pitch.
That's the exact backdrop a hyped-up, superlative-heavy opening line lands against. It's not that Nordic buyers don't buy, they buy plenty, and I've built pipeline into all four of these markets. It's that the specific tell of an unedited AI draft, confident claims stated as fact, urgency that isn't earned, "I noticed" openers that clearly weren't personalized by a human, reads as exactly the kind of self-promotion Janteloven trains people to distrust on sight.
What an unedited AI draft defaults to
Run the same prompt through most AI drafting tools and you'll see the same tells recur: a superlative in the first sentence ("game-changing," "industry-leading"), an assumed shared urgency ("I know Q4 is a critical time for teams like yours"), a claim stated with more confidence than the evidence supports, and a close that presumes the meeting rather than asks for it. None of that is unique to any one vendor's model. It's what happens when a system is optimized to sound persuasive in the abstract, with no specific reader in mind and no cultural read on how that reader will hear it.
In a market that already discounts visible selling, every one of those tells is doing double damage: it reads as generic and it reads as pushy, in a culture where pushy is the faster way to lose the reply.
The same opener, before and after an edit
Here's a concrete example rather than an abstract description. Both are drafts for the same fictional prospect, a Head of Ops at a Swedish logistics company.
| Dimension | Unedited AI draft | Edited for a Nordic reader |
|---|---|---|
| Opening claim | "We're the industry-leading platform helping ops teams like yours save 40% on freight costs" | "We built a tool that cut freight cost tracking time for two Baltic logistics teams, results vary by setup" |
| Urgency | "With Q4 approaching, now is the critical time to act" | (removed entirely, no invented deadline) |
| Proof | Unverifiable percentage stated as fact | Specific, modest, attributed claim with a caveat |
| Ask | "Let's grab 15 minutes this week to discuss how we can transform your operations" | "Worth a short call if this is relevant, no pressure either way" |
| Overall tone | Confident, urgent, presumes interest | Understated, factual, leaves the decision with the reader |
The edited version isn't weaker copy. It's the same offer with the overselling stripped out, which in a low-masculinity, Janteloven-influenced culture is what actually reads as credible rather than as a red flag.
Does this carry over to cold email?
Largely yes, the mechanism is about tone, not the channel it arrives through. I go deeper into the broader question of when translating and adapting copy for a market is worth the cost, versus running the same English copy everywhere, in native-language cold email: when localization is actually worth it. Tone matters even inside English copy: a Swedish or Danish reader doesn't need the language changed to notice an overselling opener, they need the register changed. This is exactly why, in the sequencing tool I run for clients, Salesforge, I always treat its AI draft as a first pass I edit by hand for markets like these, never as the thing that goes out.
The fix: AI writes the first draft, never the send
The practical rule I use is simple: AI drafts get you from a blank page to a rough first pass fast, and that's genuinely useful. They should never be the version that gets sent to a Nordic prospect without a human pass that specifically checks for three things: strip every superlative that isn't independently verifiable, remove any invented urgency, and rewrite the ask so it leaves room for the reader to say no comfortably. That last one matters more here than most places. An ask that presumes a yes reads as pushy in a culture where a confident no is the polite, expected response to an offer that isn't earning attention yet.
Tip. Read the draft out loud and ask whether a modest, understated person could have written it and still meant it. If the claim sounds like something the reader's own colleague would be uncomfortable saying about themselves, it needs to come down a register before it goes out.
What I actually run for clients selling into these markets
For clients selling into Sweden, Norway, Denmark or Finland, this is the edit pass I make on every AI-drafted opener before a campaign goes live there, and it's the single most common change I make to a draft written for any of these four markets specifically: cutting the confidence level of the first two sentences roughly in half. Not the substance, the register. The offer stays exactly as strong. It just stops sounding like it's trying to convince the reader of something, and starts sounding like information they can act on or ignore. If you're scoping outbound into this region specifically, my B2B lead generation service for the Nordics page has the market-by-market detail this article doesn't have room for.
Questions to ask before you buy an AI drafting tool
If you're evaluating an AI SDR or an AI drafting layer for outbound into these markets, ask the vendor directly whether their model's default output can be tuned for register, not just language, separate the two explicitly when you ask. Ask whether their own benchmark data is segmented by region or market at all, and if not, treat any claim about performance in a specific country as unverified until you test it on your own list. And ask what the human review step actually looks like in their recommended workflow, because "human in the loop" means very different things depending on whether that human reviews every message or spot-checks one in twenty. I go through the broader brand-risk side of this question, beyond just tone, in how to deploy an AI SDR without wrecking your brand, which is worth reading before you sign anything.
The metrics that will actually tell you if this is a problem
Don't wait for a qualitative sense that something feels off. Watch acceptance rate and reply rate separately by market, the same split Expandi's own methodology uses, since a tone problem tends to show up as a connection or acceptance dip before it shows up in replies. If your Nordic-market acceptance rate is meaningfully behind your other markets on materially similar targeting, that's your signal to pull a sample of messages and check them against the before-and-after table above, not to assume the list quality or the offer is the problem first.
- Acceptance or connection rate, split by country, not lumped into one "EMEA" bucket.
- Reply rate on accepted connections specifically, since a low-tone opener can still get accepted and then get ignored.
- Time-to-reply, a faster negative response in these markets is often a polite decline to copy that oversold, not disinterest in the offer itself.
Key takeaways
- Expandi's H2 2026 report found AI-drafted LinkedIn messages losing to human-written ones on acceptance by about 12%, within the same accounts, across 118 campaigns and 73 accounts. That data has no Nordic-specific breakout.
- Hofstede's own comparison tool scores Sweden 5, Norway 8, Denmark 16, and Finland 26 on the Masculinity dimension, among the most cooperation-oriented, least self-promotional scores measured anywhere.
- Janteloven's cultural discomfort with visible self-promotion is a documented norm, not a stereotype, and it's the mechanism connecting the two data points above.
- Unedited AI drafts tend to default to superlatives, invented urgency, and a presumptive ask, exactly the tells a low-masculinity, Janteloven-influenced culture already discounts.
- The fix is a human edit pass focused on register, not substance: keep the offer, cut the confidence level of the opening lines and the ask.
FAQ
Does AI-personalized outreach actually perform worse in the Nordics specifically, with data to prove it?
Not as a published, region-specific finding. Expandi's H2 2026 report documents a real 12% acceptance-rate gap between AI-drafted and human-written LinkedIn messages, but that gap is a global figure with no Nordic breakout. The Nordic-specific case rests on separate, real research: Hofstede's cultural-values scores and the documented Janteloven norm, which explain why the mechanism behind that gap would plausibly matter more here, not a study that already measured it here directly.
What is Janteloven and why does it matter for cold outreach?
Janteloven, or the Law of Jante, is a set of informal Nordic social norms, most associated with the author Aksel Sandemose, discouraging visible self-promotion and boasting. For outreach, it means copy that reads as confident self-promotion, superlatives, unearned urgency, presumptive asks, starts from a deficit of trust before the reader even considers the offer.
Should I stop using AI to draft LinkedIn messages and cold emails for these markets?
No. Use AI for what it's genuinely fast at, a rough first draft, then run a human edit pass focused specifically on register: cut superlatives, remove invented urgency, and soften the ask. The offer stays the same; only the confidence level of the language changes.
Is this backlash unique to the Nordics, or does it apply elsewhere?
The underlying 12% acceptance gap between AI and human-written messages is global, per Expandi's data, so the core lesson (edit AI drafts before sending) applies everywhere. The Nordics are a sharper case because the cultural discount on visible selling is unusually well documented there, not because the underlying mechanism is exclusive to the region.
How do I edit an AI draft so it sounds less like a pitch?
Three checks: remove any claim you can't independently verify, cut any urgency you didn't actually create, and rewrite the closing ask so a reader can decline comfortably rather than feeling pushed toward a yes. Reading the message aloud and asking whether a modest colleague of the reader's would say it about themselves is a fast, practical test.