A claimed link is not a verified link

Almost every outreach tool ends the same way. A reply comes back, the editor says yes, and somewhere in the interface there is a checkbox marked Won. You tick it. The number on the dashboard goes up by one.

Nothing was checked. The number is your own memory of a conversation, rendered in a larger font.

This matters more than it sounds, because that number is the one that ends up in a client report. It is the number that decides whether a research gets renewed. And it is a number nobody — not you, not your client — can audit after the fact, because there is nothing behind it except the moment you clicked.

What we do instead

Marking a thread won asks for one thing: the URL of the page carrying your link.

We fetch that page. We look for a link whose destination is the target URL from the research — not a URL supplied in the form, which would let the claim define its own success condition. We record what we found: the anchor text, the HTTP status the page returned, the moment we looked. That record is append-only. A later check writes a new row; it never edits the old one.

The result is that “verified” means a specific thing, and the thing it means is checkable. If a client asks how you know, the answer is a date, a status code, and an anchor string.

We also keep two events instead of one. Claimed fires when you say you won. Verified fires only when a page we read had the link on it. The gap between those two numbers is information — it tells you how often an agreed placement quietly fails to appear — and merging them into a single “wins” figure is precisely what throws that information away.

And a weekly look back

A link that existed in March is not a link that exists today. Pages get rewritten, sections get pruned, sites get redesigned, and the link you were paid to earn goes with them. Almost nobody notices, because nobody looks.

We look once a week. If a link that was verified is no longer on the page, the placement moves to missing and you get an email. first_verified_at is written once and never moves, so the record still says the link was real on the day it was real — losing a link later does not retroactively erase that you earned it.

What we refuse to tell you

Here is the part most tools would quietly skip.

The provider endpoint we read pages through returns them as normalized markdown, not as raw HTML. Markdown has anchors and destinations. It has nowhere to put a rel attribute.

Which means we cannot see whether a link is nofollow, and we do not pretend to. The placement record carries a flag saying whether follow status was observed, and on this path it is always false. The interface says “not observed”. The report says “not observed”.

We could have shipped a dofollow badge that is right most of the time, because most editorial links are followed. It would look better. It would also be exactly the unevidenced claim this whole product exists to remove — a green tick asserting something nobody checked. If we would not accept that from a link vendor, we should not ship it ourselves.

Seeing rel for real needs a different kind of provider request: one that returns link attributes, crawls a whole site, and runs as a background task rather than a single page fetch. That is a real piece of work and it is on the list. Until it is done, the honest answer is that we do not know.

Why a blind spot is not a failure

There is one more distinction worth naming, because it is the difference between a useful system and a noisy one.

If we fetch a page and it returns nothing readable — a timeout, a block, a provider refusal — we do not record that the link is missing. We record that we could not see. A page that did not answer has not been shown to lack anything, and treating our own blind spot as evidence against you would manufacture losses out of network weather.

The same rule applies when a check would exceed the month’s research budget. The attempt is recorded, the state is unchanged, and a verified placement can never be demoted by us running out of money to look at it.

An instrument that reports “I could not measure” is more useful than one that reports zero.