LinkedIn Outreach: What 13 Million Connection Requests Actually Show About Timing and Length

The short answer

Send LinkedIn connection requests in the morning (7-11am), keep the note to roughly 150-200 characters, cap follow-ups at three messages, and don't assume AI personalization is doing you any favors, in a real dataset it measurably underperformed a human-written message on the same accounts. All five findings below come from the same disclosed-methodology report: Expandi's analysis of 13.2 million connection requests across 13,302 active accounts, May 2025 through April 2026. Expandi sells LinkedIn automation, and these are correlations in its own platform data, not controlled tests. For context, the average acceptance rate across all requests was 28.5%.

Timing: mornings win, clearly

The report found day of the week made little difference, time of day made a real one. A connection request competes with everything else arriving in someone's notifications, and a morning send lands while someone's actually triaging their inbox rather than competing with the evening's accumulated noise.

Message length: 151 to 200 characters had the best reliable result

That 15.7% figure for ultra-short messages is real data, but it's worth reading the report's own caveat rather than the headline number: it's drawn from only 9 campaigns, too small a sample to build a strategy around. The range with an actual, reliable sample behind it is 151-200 characters, enough room to name a real reason for reaching out without turning the message into a pitch.

Follow-up count: three messages is the best result in the dataset

This is the finding worth sitting with the longest. More follow-ups isn't more persistence paying off: reply rate was lower at five or more messages, which is consistent with long sequences hurting the sender, though this is correlational data and long sequences may also go to colder lists. The report's own second-message-position data backs this up: the second message in a sequence is the single most effective position per recipient, at 8.8%, meaning most of the value is front-loaded into the first couple of touches, not accumulated by grinding out a longer sequence.

The AI-personalization finding most vendors won't publish

Expandi ran a within-account comparison: the same accounts, some messages written by a human, some generated by AI hyperpersonalization. The AI-generated messages dropped acceptance rate by 12% and left reply rate flat, no improvement on the metric that actually matters, a real cost on the one that's easier to measure.

This is a genuinely inconvenient finding for a lot of the AI-outreach-tool pitch, and it's worth taking seriously precisely because a LinkedIn automation company had no obvious incentive to publish data that complicates the sales pitch for AI personalization. UserGems' survey data on AI adoption points the same way: much of what gets automated with AI still needs a human check before it outperforms the version a person wrote.

There is a more useful finding on the personalization side. In Expandi's analysis of 240 campaigns with at least 50 contacts each, connection notes that referenced a real shared tie, such as a follow, a group or a shared network, were accepted at 1.8 to 2.7 times the 30.2% baseline acceptance rate. Personalize around something real and checkable, not around flattery.

Industry changes the baseline more than the tactics do

Roughly a 2x gap between two industries, both on LinkedIn, both likely running broadly similar sequences. Before benchmarking your own numbers against any of the figures above, it's worth knowing which end of that range your industry tends to sit on, a Computer Software team hitting 9% is closer to normal than underperforming.

What to actually change this week

Shift sends to the 7-11am window if they're currently going out whenever the sequence happens to fire. This is the single cheapest change on this list, a scheduling adjustment, not a new tool.

Trim messages toward 150-200 characters. If a connection note or first message is running long to sound thorough, the data says shorter, specific, and direct outperforms it.

Cap sequences at three messages in total. A real, verifiable reason for the outreach does more work in three messages than a generic one does in six.

Don't assume AI personalization is a free upgrade. Test it against a human-written control on your own accounts before rolling it out everywhere, exactly the comparison Expandi ran, rather than assuming it performs the way the pitch for it implies.

DeliReach schedules sends inside the actual high-acceptance windows automatically and caps sequences at the lengths the real data supports, rather than running on a fixed schedule nobody's checked against outcomes.

For the full outbound workflow, see our outbound lead generation guide.

Sources

• Expandi, State of LinkedIn Outreach H2 2026, 13.2 million connection requests across 13,302 accounts, May 2025-April 2026

FAQ

Related reading

• LinkedIn outreach: what works and what gets you banned
• B2B outbound templates that actually work
• Signal-based outreach: how it actually works