Evidence is a page, not a domain

Open any link prospecting tool and you get a list of domains with numbers beside them. Authority scores. Traffic estimates. Spam scores. Sort descending, export, start emailing.

The numbers are usually accurate. They are also answering a question nobody asked.

A domain metric tells you that a site is, in aggregate, worth being linked from. It cannot tell you that this page, written by this person, has a paragraph your work would improve. And that second thing is the only thing an editor cares about, because it is the only thing that makes saying yes rational for them.

The email that gets ignored

You know the shape of it, because you have received a hundred:

Hi there — I came across your excellent article on [topic] and thought your readers might find our guide useful. Would you consider adding a link?

Every word of that is compatible with never having read the article. The sender knows the domain score. They know the URL. They do not know what is on the page, and the editor can tell within one sentence.

The reply rate is what it is not because the pitch is badly written. It is because the pitch contains no evidence that the sender did the work, and the editor’s inbox is full of people who also did not do the work.

What “evidence” has to survive

We built the research step around a rule: a page does not exist until we can point at the passage that makes the case.

That means, for every candidate page, we hold:

• the owned target — the specific page of yours the link would point to, which you confirmed you own before anything ran;
• the exact candidate page, not the domain it sits on;
• a quoted passage from that page, pulled from its actual content, where the connection is visible.

If the passage does not exist, the page does not get shown to you. A candidate that survives only as a domain score is a candidate we could not justify, and showing it anyway would just be moving the guesswork downstream into your inbox.

This is deliberately expensive. Reading pages costs more than reading a metrics API, and the plan’s research budget is finite because of it. But the constraint is doing real work: it forces the question “why this page?” to be answered before anyone’s name goes on an email.

The part that changes how outreach feels

A brief built this way does something to the draft that follows it.

When the evidence is a quoted paragraph, the first line of the email writes itself, and it is a line that could only have been written by someone who read the piece. Not “your excellent article”, which is a compliment about a URL. Something closer to: here is the sentence where you said X, and here is the thing we published that goes one step further on exactly that.

The editor’s decision changes shape too. They are no longer deciding whether to do a stranger a favor. They are deciding whether a specific paragraph would be better with a specific reference. That is a normal editorial judgment, and people make normal editorial judgments all day.

What this rules out

Being honest about the method means being honest about what it cannot do.

It rules out volume. If a page needs a real passage from a real page, you cannot have ten thousand of them, and a tool built on this rule will always return fewer candidates than one built on a domain export. That is a genuine trade and you should make it with your eyes open.

It rules out the sites that have nothing to say about your subject. Some perfectly authoritative domains will never appear in your list, because no page on them is about the thing you work on. This looks like a gap in coverage. It is the filter working.

And it rules out sending without reading. Every draft here is built from the evidence and then shown to a person before it can leave. Not because approval is a feature — because a system that can send an unread email will eventually send one, and the passage you quoted is only worth anything if it is true.

Domain metrics are a fine way to decide which sites are worth your time. They are not a reason for anyone to link to you. Only a page is that.