# Your AI Copilot Is Personalising the Wrong Layer of the Sale

> AI copilots swap out company names and job titles and call it personalisation. The layer that actually moves buyers is still blank.

By Greg McCallum, founder of PeerLab. Published 2026-09-08 in Content & Tooling. Source: https://www.peerlab.ai/blog/sales-content-personalisation-ai-copilot-wrong-layer/

Most AI-assisted sales emails now arrive with your company name in the subject line, a reference to your recent Series B, and a sentence about "challenges commonly faced by VP Operations in the logistics sector." The rep who sent it spent forty seconds on it. You can tell. Not because it mentions your job title, but because it says nothing that could only be true of you.

That is the personalisation problem teams are not talking about. They are so pleased the tool knows the industry vertical that they have stopped asking whether the email knows anything real.

## There are two layers. Copilots only reach one.

Surface personalisation is: company name, job title, funding round, recent hire, tech stack from a data enrichment source, maybe a LinkedIn post they made six months ago. This is genuinely useful information. It is also available to every other rep with the same tool and the same subscription. Which means it is not a signal of effort, and it is not a differentiator. It is table stakes, and table stakes have never closed a deal.

Situational personalisation is different. It is: what does this buyer privately believe is true about their problem? What do they think inaction costs them, and do they think it costs them enough to move? What objection have they already rehearsed before you showed up, the one they are going to deploy in the first ten minutes of any call?

No copilot knows that. Because the copilot was not on your last call with a similar buyer in this vertical who told you, unprompted, that they had tried this exact category of solution two years ago and it had burned them. That context lives in your head or in your CRM notes, and most of the time it lives nowhere at all.

## Why AI tooling has made this worse, not better

Before AI writing tools, personalising an email took time. That friction was a natural forcing function. A rep who was spending fifteen minutes on a single outbound email was, consciously or not, asking themselves whether they actually knew enough about this person to write anything worth reading.

Cheap, fast personalisation has removed that friction entirely. You can now produce fifty "personalised" emails in an hour, all of them mentioning the right vertical, all of them hollow. The volume feels like effort. The metrics look fine at send. The reply rates tell the real story, and most teams are too busy celebrating the productivity gain to read them correctly.

This is not the copilot's fault. The copilot is doing what it was built to do: take inputs and produce outputs. The problem is what teams are treating as an input. "They are in manufacturing and they are a CFO" is not a discovery insight. It is a filter criterion. Feeding it to an AI and calling the result personalised is a category error.

## What actually moves a buyer is a reflection, not a reference

The emails and decks and follow-ups that get responses are the ones where the buyer thinks: that is exactly the situation I am in. Not "they mentioned my industry." Exactly my situation.

That requires you to know something specific: that this CFO is personally accountable for a cost-reduction target she cannot hit with the current process, that she has already told her board she is on top of it, and that admitting she needs external help now is uncomfortable. You do not learn that from LinkedIn. You learn it from a conversation, or from a very close-in reference from someone who has had the conversation.

Once you know it, a copilot is genuinely useful. You have the raw material. The tool can shape it into a PAS structure, calibrate the tone, tighten the language. That is drafting. That is a legitimate use of the technology.

The mistake is running the sequence in reverse: generate the content first, hope the personalisation tokens do the job, and treat discovery as something that happens after the reply lands. By then, you are playing catch-up with a buyer who has already categorised you as one of fifty reps who mentioned their funding round this quarter.

## The measurement problem is upstream of the tooling problem

Most enablement teams measure personalisation by inputs. Did the email include a personalisation variable? Did it reference a trigger event? Did the deck have their logo on the cover? Tick, tick, tick. Good personalisation score.

None of those inputs measure whether the content reflected something the buyer privately believes. That is harder to score. It requires a human to read the email and ask: "Is there anything in here that this specific person could not have heard from the three other vendors in this process?" If the answer is no, the personalisation score is a lie.

Until teams measure the output rather than the inputs, better tooling just produces better-dressed mediocrity faster.

[Discovery Question Generator](https://www.peerlab.ai/library/tools/discovery-question-generator/) is worth running before you brief your copilot on anything, not after. If you cannot answer the situational questions it surfaces, you are not ready to write the email. You are ready to do more discovery.

## The fix is a sequencing problem, not a prompt problem

The teams getting this right are not using smarter prompts. They are enforcing a different order of operations:

1. **Discovery or research that surfaces a situational belief.** A call note, a reference customer conversation, a specific comment from the economic buyer in a previous meeting. Something that is true of this person, not this persona.
2. **A brief that includes that belief explicitly.** Not "CFO in logistics, 500 employees." Something like: "She has committed internally to reducing headcount costs without cutting headcount. She is three months from a board review and she has not solved it yet."
3. **The copilot as drafter.** Take that brief, apply a proven structure (the [Cold Email PAS Copywriting Framework](https://www.peerlab.ai/library/cold-email-pas-copywriting-framework/) works well here), produce the draft, edit for accuracy and tone.
4. **A human sense-check.** Does anything in this email require the buyer to think "that is exactly my situation"? If not, it goes back.

That fourth step is the one most teams skip because the tool has already produced something that looks finished. Looking finished is not the same as being true.

The hard news for anyone who has rolled out AI-assisted sequencing and measured it by volume metrics is that your reps are now producing shallow personalisation at industrial scale. The conversations that result, when they happen at all, will start from a buyer who has already decided you are one of many. Getting out of that frame takes more work than getting into it in the first place.

The copilot is not the problem. The problem is the assumption that inputs about a person are the same as insight about a person. They are not. One is a data point. The other requires a conversation.

Do the conversation first.

**The tool for this:** [AI Copilot Pre-Brief: Situational Personalisation Worksheet](https://www.peerlab.ai/library/ai-copilot-pre-brief-situational-personalisation-worksheet/), free and no signup.