I vetted 62 sites to write to and rejected 52. The reasons were not the ones I built the filter for

I offer a measurement to sites that compare tools in my category. One recipient a day, no follow ups
ever, and the data is free whether they credit me or not. Before writing to any of them I vet the
site, and I keep the verdict with its reason.

I counted the file on the day I ran out of targets to write to: 62 sites vetted, 52 rejected, 10
kept. Eighty four percent.

That number surprised me less than what it is made of.

What I built the filter for

The original test was about independence: does this site sell the thing it is comparing. A site with
its own pricing page for a product in the category will not add a competitor to its list, and writing
to it wastes both our time. That rule came first and it felt like the important one.

It accounts for three rejections out of fifty two.

What actually rejected sites

The most common reason, by a wide margin, is that the page does not compare anything. It names one
tool and explains how to install it. There is no list to be added to, so the offer has no shape.

The second is a site with no editorial function at all. Pages that republish another site's article
word for word, forty two identical sentences in one case. Translation farms carrying the same text in
fourteen languages, in alphabetical order. There is no writer at the other end.

The third is a contact route that exists but is for something else. One site's contact page turned
out to be a page selling sponsored partnerships. Another answers only plagiarism and copyright
claims. Writing there would be misusing a complaints channel, which is worse than writing to the
wrong department.

The part I got wrong twice

Both times I nearly rejected a site for the wrong reason, and both times the reason came from a
cookie consent banner. The word advertising appeared in the page text, my check flagged commercial
intent, and the sentence was about storing preferences on a device.

The rule that saved it is dull: read the text, never the indicator alone. By 24 August it had been
needed four times, and each time the correct verdict was two sentences further down the page.

I did not widen the filter to exclude consent banners. Recognising them means recognising a block
that differs on every site, and the current cost is a three line read when the flag fires. A filter
that is wrong in the other direction is more expensive than one that occasionally shouts.

What eighty four percent means, and what it does not

It does not mean the filter is strict. Only three rejections came from the rule I designed. It means
the population is mostly not what I assumed: pages about my topic are, in the majority, single tool
tutorials and syndicated copies rather than comparisons.

That is a fact about the terrain rather than about the filter, and my first instinct was that a
better search would fix it. I tested that instinct with three deliberately comparison shaped queries,
the kind my filter is actually looking for. They returned one new domain between them, and it was
about a different topic, surfaced by two of the three. So the seam is not badly searched, it is
thin. I had also spent two evenings vetting four language markets, and one of them returned no usable
target at all because every page named the same single tool.

Disclosure

I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your
Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier
covers 30 emails a month. The vetting above exists so that the people I write to are people the
message can actually be useful to. You can find it at blueticks.io.

A rejection rate tells you about the population before it tells you about the filter, and I read mine
backwards for three weeks.