The guide
Why AI outreach all sounds the same
First Person Outbound, explained by SimplyB2B
Buyers are not imagining it. AI outreach genuinely does sound like a single voice wearing different name badges, because most tools generate messages from shared language models trained on the same corpus of "high-converting" templates. The pattern recognition that makes those models useful is exactly what makes the output recognisable. When a buyer has read one, they have read ten thousand.
Key takeaways
- Shared models produce shared patterns: the same sentence rhythms, the same fake-specific openers, the same three-beat closes.
- Buyers have developed a near-instant filter for AI tells: the too-clean sentence structure, the absence of any real opinion, the compliment that could apply to any company.
- Volume is the wrong answer to low reply rates, more of the same message reaches more of the same delete reflex.
- The fix is not a better prompt, it is messages that carry information only the real sender could supply.
- First-person outreach from a real account with a real post history cannot be faked at scale by a rented persona.
Why does AI cold email sound the same?
Every major AI outreach tool generates copy from a shared language model, then steers it with questionnaire inputs like industry, persona, and value proposition. Because thousands of senders feed the same model the same categories, the output converges. The sentence shapes, the opener formulas, and the closing structures are statistically identical across competitors' inboxes.
The problem is structural, not cosmetic. A shared model has no access to how a specific person actually writes under pressure, what words they reach for when they are genuinely excited about a deal, or which details they drop that only an insider would notice. It produces the median of all those senders at once.
Personalization fields like '[First Name]' and '[Company]' are surface-level variables inside an otherwise fixed template. Buyers have learned to see through them the way a reader skips past a mail-merge salutation. The tell is not the variable, it is every sentence surrounding it.
Shared model
Thousands of senders query the same underlying AI
Questionnaire input
Industry, persona, pain point fed as variables
Median output
Model returns the statistical centre of all those inputs
Surface personalisation
Name and company swapped in, structure unchanged
Buyer pattern match
Recipient recognises the template, delete reflex fires
What do buyers actually do when they spot AI outreach?
Most buyers delete without replying, which understates the problem because it leaves the sender with no signal. A growing share mark messages as spam or disconnect the sender on LinkedIn entirely. Some buyers share the worst examples internally or publicly, which attaches reputational cost to the sender's real name rather than the tool's.
The damage compounds over time. A buyer who has been hit by the same AI pattern three times from three different companies does not weigh each message on its merits. The pattern itself triggers the reflex. Senders who keep increasing volume to compensate are accelerating toward a platform-level trust penalty, not away from it.
What buyers respond to is specificity that could not have come from a scrape: a reference to something the sender genuinely observed, a point of view with an edge, a follow-up that reads like a human remembered the last message rather than re-generated it.
What makes AI-written cold outreach so easy to detect?
Three structural tells appear together in AI outreach more reliably than in human writing: an opening that names a surface-level fact about the recipient's company, a value proposition framed as a universal pain the tool has pre-loaded, and a call to action that asks for calendar time before any relationship exists. Human writers break at least one of these patterns naturally.
There are subtler tells too. AI-written messages rarely contain an opinion that could be wrong. They do not trail off, hedge on a small point, or betray any particular knowledge of the vertical. The syntax is too clean, the paragraphs too balanced. A real message from someone who knows the space usually has one sentence that a copywriter would edit out, and that imperfection is exactly what makes it credible.
LinkedIn compounds the problem because the platform surfaces account age, post history, and mutual connections alongside every message. A new account with no posts and no shared connections sending polished outreach has none of the ambient evidence that makes a stranger trustworthy. The message quality becomes irrelevant when the profile context contradicts it.
Can a better prompt fix generic AI outreach?
A better prompt adjusts the output of the same shared model. It can change word choice and tone at the surface, but it cannot give the model access to how this specific sender writes when they are not trying. The convergence is a property of the model, not the instruction. Prompt engineering reduces the generic quality at the margins, it does not solve the structural cause.
The sellers who see the biggest difference in reply rates are not the ones who spent longer on their prompts. They are the ones whose outreach carries first-person context that only they could supply: their actual observations about a prospect's business, their real take on a problem in the space, a follow-up that references what was said rather than re-running the sequence from step one.
That kind of specificity cannot be produced by any tool that works from a questionnaire and a shared model. It can only come from a system that seeds its voice from the sender's own writing history and treats every edit the sender makes as training data for that sender alone.
How does first-person outreach differ from AI-generated outreach?
First-person outreach runs from the sender's own account, in writing derived from their own sent messages and posts, not a questionnaire. The voice model is private to that account. Every message the sender edits updates only their own voiceprint. Buyers who check the profile see a real post history, account age, and mutual connections: evidence a rented persona cannot produce.
SimplyB2B seeds each account's voice from the user's own LinkedIn sent messages and post history. Nothing is pooled across users. When a user edits a draft, that edit trains their voiceprint and no one else's. Autonomy expands only as edits shrink, so the system never runs ahead of the sender's actual trust in it.
We run our own outreach on SimplyB2B, on the founders' own LinkedIn accounts. In our best month, up to 51% of connection invites were accepted (8 to 31 July 2026, 315 people invited) and up to 27% of comments drew a reply (August 2026, 99 comments). One account ran up to 1,900 outreach actions in a month (August 2026) and started up to 129 real conversations in a month (August 2026). Figures as of 7 October 2026. Those are our own accounts, in our own voices. The mechanism is not volume. It is that the messages read the way the sender actually writes, because they are built from exactly that.
Is LinkedIn outreach safer from the AI-sameness problem than email?
LinkedIn has a built-in credibility layer that email lacks. Every message arrives alongside the sender's profile: years of activity, posts, shared connections, job history. A message that reads like a human wrote it gets ambient corroboration from everything the profile says. A message that reads like AI copy gets exposed by the same context, especially on a thin or new account.
That context cuts both ways. It makes authentic first-person outreach more trusted than the same words sent by email, and it makes AI-pattern outreach more obviously suspicious than it would be in an inbox. Buyers on LinkedIn have a one-click way to check whether the sender is a real person who knows the space, and many do check.
Running outreach from your own established account, with a real post history and genuine mutual connections, puts structural evidence behind every message before the buyer reads a word. That is an asset AI-generated outreach from a rented persona cannot replicate, regardless of how well the copy is written.
Common questions
Does sending more messages fix a low reply rate on AI outreach?
Not if the messages share the same structural tells. Higher volume accelerates buyer pattern recognition, increases spam reports, and can trigger platform-level throttling. The reply rate problem is a message-quality and trust problem, not a delivery-volume problem.
What is a voiceprint and why does it matter for outreach?
A voiceprint is a per-account voice model built from that sender's own writing history, not a shared template or a questionnaire. Messages generated from it carry the sender's actual sentence rhythms and word choices rather than the median style of a shared AI corpus. In SimplyB2B, edits a user makes update only their own voiceprint.
Do buyers really check a LinkedIn profile before deciding whether to reply?
Many do, and the platform makes it trivial. Post history, account age, shared connections, and recent activity are all visible alongside the message. A genuine profile with years of posts corroborates the message, a thin or inactive account contradicts it regardless of how the message itself is written.
Why do AI outreach tools claim their messages sound like you, if they all sound the same?
Most tools mean they apply a tone setting drawn from a questionnaire or a short style sample to a shared underlying model. The tone can shift but the structural patterns stay because they come from the model, not the sender. A genuine per-account voice model trained on that sender's own writing history is a different architectural choice.
Related: Why outbound at scale fails · AI SDR vs founder-led outreach · How to do LinkedIn outreach · Voice matching: how SimplyB2B builds your voiceprint · First-person outbound explained
See what your voice actually sounds like
The free Voiceprint read shows you the voice model before you spend a dollar. Start there.
Free to set up · pay only when you connect · 30 day money back guarantee