Every claim in a draft is tied to something real

Specific claims in a draft are now bound to the evidence behind them at the moment the email is written. A sentence the writer cannot trace back to a source does not make it into the draft.

What changed

A personalised cold email lives or dies on one thing: whether the specific detail in it is true. "I saw you opened a second office in Manchester" is a good opening line if they opened a second office in Manchester. If they did not, it is worse than a generic email, because now the prospect knows the personalisation is fabricated and reads the rest of the message accordingly.

The writing step used to work from a pile of research and produce an email. Checking happened afterwards, by reading the finished draft and looking for claims that seemed unsupported. That is the wrong order. By the time you are checking a finished email you are trying to prove a negative about text that already reads convincingly.

Claims are now bound to evidence during generation. Each specific assertion about the prospect, their company, their situation, or their industry has to point at the piece of research it came from. A claim with nothing behind it does not survive into the draft.

The check also covers the softer cases. A sentence can be technically hedged and still assert something: "it looks like you have been expanding into new markets" is a claim about the prospect wearing a hedge. Those are now caught rather than passing on the strength of the hedge.

How to use it

You do not need to configure this. It applies to every draft.

What you will notice is that emails are more conservative about specifics. A draft written from strong research will mention concrete things about the prospect. A draft written from thin research will talk about what you offer, because there was nothing specific to say about them that could be supported.

That variance is the point. It reflects the research rather than covering for it. If you see a run of drafts that stay generic, the problem to solve is upstream in the research, not in the writing instructions.

Why it matters

The failure mode this prevents is the worst one an outreach tool has. A confidently wrong detail about someone's business does more damage than no detail at all. It costs you the prospect, and if it happens enough it costs you the reputation of the domain you sent from.

It is also the failure mode hardest to catch by reading. A fabricated specific reads exactly like a real one. There is nothing in the sentence itself that marks it as invented. The only reliable way to catch it is to require the source at the point the sentence is written, which is what now happens.

What counts as a claim

Anything specific about the prospect or their company. Their size, their location, their recent activity, their tooling, their customers, their hiring, their market.

Statements about your own offering are not claims about the prospect and are not checked the same way. You know what you sell. The check is about assertions concerning the person receiving the email.

Generic industry statements sit in between. "Agencies are dealing with tighter budgets this year" is a claim, and a weak one, because it is asserting something about their situation without knowing it. Those now need support or they get cut.

Banned phrasings

Alongside the evidence binding, a set of phrasings that consistently produce unsupportable claims are now blocked at the writing step.

These are the constructions that sound personalised while asserting nothing checkable, and that a reader recognises immediately as filler. Openers that guess at a pain point. Compliments about work the writer has not seen. Assertions about what the prospect is "probably" dealing with.

Blocking them at generation is more reliable than filtering afterwards, because these phrasings are the writer's natural fallback when research is thin. Removing the fallback forces the honest outcome, which is either a specific claim with a source or no claim at all.

What this costs

Some drafts are less personalised than they used to be. That is a real trade and worth stating plainly. An email that previously opened with a confident guess now opens with something duller and true.

The exchange is worth making. A guess that lands is worth a little; a guess that misses is worth a lot in the wrong direction, and you cannot tell in advance which one you are sending. Removing the guessing removes both.

Where you feel this most is on prospects with thin online presence. Those were the drafts most dependent on invention, and they are the ones that changed most.

Writing instructions that work with this

If your campaign instructions ask the writer to reference something specific about every prospect, you are asking for a guarantee the research cannot always make. On a prospect with a thin online presence, that instruction has no honest way to be satisfied, and the result is a lead that produces nothing rather than a lead that produces a plainer email.

Instructions written as preferences work better than instructions written as requirements. Asking the writer to lead with something specific when there is something specific to lead with gives it a legitimate path on both kinds of prospect. Asking it to always open with a detail about their recent work gives it a path only on prospects where such a detail exists.

The same applies to the shape of the email. Instructions that prescribe a structure depending on prospect detail will fall back awkwardly when the detail is absent. Instructions that describe what you want the email to accomplish leave room for the writer to reach that end with whatever it can actually support.

How this interacts with review

If you have review turned on, you will see the effect of this most directly, because you are reading the drafts rather than only their outcomes.

A draft that says less about the prospect than you expected is now more likely to be an honest reflection of thin research than a writing failure. Before deciding the email is weak, look at what was found. The two problems have different fixes and they look identical in the finished text.