Sandery Insights / Professional Signals

Research note 03 · Mitch Sandery · August 2026

AI Can Write the Perfect Cold Message. That Is Why It Is Becoming Worthless.

As personalised communication becomes easier to manufacture at scale, it becomes weaker evidence of genuine interest. In professional markets, the scarce asset is shifting from access to credibility.

Reading time7 minutesEvidence standardResearch + observation

Market observation · Practitioner perspective

Executive career coach Jen Morris recently described receiving automated approaches from job seekers that were technically personalised but not meaningfully personal. Many were filtered into LinkedIn's “Other” inbox; those that reached her rarely created a worthwhile conversation.⁷

Personalisation used to prove that somebody had paid attention. Increasingly, it proves only that somebody has access to the right technology.

Personalisation used to prove something.

Before generative AI, a highly relevant cold message required effort. The sender had to research the recipient, understand the organisation, identify a credible point of connection and construct an individual approach.

The message therefore communicated more than its literal content. It suggested that somebody had invested time in understanding why the conversation might matter.

In economic terms, personalisation functioned—at least partially—as a costly signal. Spence's foundational work explains how observable signals can carry information under uncertainty when producing them is meaningfully associated with an otherwise difficult-to-observe quality.¹

Generative AI changes that relationship. Research, synthesis and superficially bespoke writing can now be produced at scale. The observable output remains. Much of the cost that gave it informational value disappears.

Authentic personalisation and observable personalisation are no longer the same thing.

Good automated outreach contaminates good human outreach.

Empirical evidence + Sandery interpretation

This is not principally a problem of badly written AI messages. Those are comparatively easy to ignore. The deeper problem is that good synthetic personalisation can increasingly resemble genuine human effort.

In six experiments involving 4,600 participants, researchers found that people were unable to reliably identify AI-generated self-presentations. The study concerned professional and other forms of self-presentation rather than recruitment outreach specifically. Its narrower implication remains relevant: polished language is becoming unreliable evidence of its own origin.²

A person who spends half an hour writing a considered approach and a system producing hundreds of individually researched messages may create very similar observable outputs. The recipient cannot see the process behind either one.

Our interpretation: rational recipients will increasingly discount the message itself.

Scale makes the problem self-reinforcing.

AI-assisted personalisation can still improve response rates. LinkedIn reports that its AI-assisted messaging increased responses while enabling personalisation in bulk.³ That makes the technology commercially rational for each sender. The collective effect may be different.

Cheaper communication encourages more communication. More inbound approaches compete for finite attention. Recipients filter more aggressively, response rates come under pressure and senders without superior information compensate with still greater volume.

The same pressure is visible elsewhere in hiring. Ashby's analysis of 109 million applications found that applications per hire tripled between 2021 and 2024 and remained above 300 throughout 2025. This does not establish AI as the sole cause. It demonstrates the scale of the attention problem into which AI-enabled outreach is arriving.⁴

01Cheaper communication

02Greater volume

03Less attention

04Weaker signal

05Greater filtering

Better wording cannot solve a credibility problem.

Improving targeting and writing more thoughtful messages remains better practice. It may produce a stronger conversation when somebody reads the approach. But it does not resolve the underlying information problem.

The recipient must still decide whether the sender is relevant, credible and genuinely interested. Increasingly, the sophistication of the message provides limited evidence with which to make that judgment.

The declining distinctionAutomated communication
versus personalised communication

The emerging distinctionUnverified cold communication
versus trusted context

The problem is no longer whether a message looks personal. It is whether the context behind it is credible.

Trust becomes relatively more valuable.

A credible introduction behaves differently. When a trusted professional says, “I know this person. Their relevance may not be obvious from the conventional signals, but I think you should meet,” the information is not primarily contained in the wording. It is contained in the relationship.

The intermediary places judgment and reputational capital behind the recommendation. Repeatedly making irrelevant introductions would weaken their credibility. The signal therefore remains expensive in a way that generated personalisation is not.

Research on employee referrals does not prove that every warm introduction is valuable. It does show that relational hiring channels can carry economically useful information. A study covering nine large firms found that referred workers were less likely to leave and generated higher profits in the industries where worker-level profitability could be calculated.⁵ Separate labour-market research concludes that referral networks can reduce informational deficiencies and improve matching.⁶

Our interpretation: as synthetic communication becomes cheaper, reputationally expensive signals become relatively more valuable.

Recruitment makes the problem more consequential.

Professional capability is already represented imperfectly. CVs compress years of experience into a few pages. Titles vary between organisations. Achievements become short bullets. Search systems naturally favour legible attributes: employers, keywords, qualifications, sectors and years of experience.

The non-obvious candidate requires interpretation. Historically, that person could partly overcome weak machine legibility through networking and thoughtful direct outreach. If cold personalisation itself becomes a weaker signal, one of the mechanisms available to explain unconventional relevance deteriorates as automated filtering becomes more prevalent.

The people who most need somebody to explain why they are relevant may be the least likely to receive enough attention to make that explanation.

The Sandery view

The scarce asset is no longer simply access.

Finding a profile, locating an email address and generating a plausible message are becoming commodities. What remains scarce is knowing who should speak to whom, understanding why the conversation should happen and possessing sufficient credibility that both parties pay attention.

Sandery is not being built to win an arms race in synthetic warmth. The Career Bank is intended to create a richer, longitudinal understanding of an individual—their experience, decisions, responsibilities, achievements, circumstances and demonstrated capabilities. Combined with human interpretation and durable relationships, that evidence can support a different kind of introduction.

Not

“Here is another candidate who says they are relevant.”

But

“We know this person. We understand the opportunity. The match may not be obvious from the conventional signals, but the underlying evidence says you should meet.”

This remains a hypothesis Sandery intends to test: that as AI reduces the signalling value of personalised communication, trusted relationships, credible intermediation and verifiable evidence become more valuable.

The implication

AI will not eliminate professional intermediation.
It will raise the standard required for it to mean something.

In a market where almost anyone can generate the perfect introduction, the valuable intermediary is the one whose introduction means something.

A richer professional record

Build the evidence behind the introduction.

Explore the Career Bank →Explore Sandery Insights →

References

Evidence and observation used in this note.

The proposed decline in personalisation's signalling value is presented as Sandery interpretation. The research supports its constituent mechanisms, not a universal causal law across professional markets.