How to judge a site like this one
Sites in this category are cheap to produce and most are produced cheaply. Five checks separate the ones worth reading, and we run them on ourselves at the bottom.
Five checks, in the order they are worth doing
1. Do the numbers carry a date
A fee or a shipping price with no sampling date is not a fact, it is a souvenir. Rates in this industry move: one agent on our own tables went from +8.9% to +2.0% in about ten weeks. An undated figure cannot be checked and cannot be wrong, which is the problem.
2. Click three outbound links
Dead links are the fastest tell. A site that has not clicked its own links in a year has not been maintained in a year, and the advice on it is the same age. Three is enough to sample.
3. Look for anything the author could only know by doing it
A specific route name, a figure read off a screenshot, a clause quoted from a terms page, a date something changed. Pages assembled from other pages have none of this: every sentence is defensible and nothing is specific.
4. Check whether it ever says something unhelpful
Real testing produces inconvenient findings: this one is cheapest but takes two months, this agent is down today, this category has no good options. A site where every agent is excellent is a list of affiliate links with paragraphs around it.
5. See whether it admits what it cannot know
Nobody can tell you which agent will handle your problem well. A site that ranks agents on trustworthiness has invented a measurement. The honest move is to say so.
Why so many of these sites are thin
Because the economics favour it. These pages earn by referral, the content needed to rank is cheap to generate, and nobody checks a shipping table for accuracy before clicking through it. A site can be built in an afternoon, never touched again, and still earn for a year.
The tell is usually uniformity. Every agent gets a positive paragraph, the ordering looks like a ranking without explaining what it ranks, and the numbers are round. Round numbers are a warning on their own: real sampled figures come out at +12.2% and 129 to 139 days, not at "around 10%" and "about two weeks".
The second tell is that nothing is ever out of date, because nothing was ever dated.
The same five checks, applied to this site
Including the ones we do not pass cleanly.
1. Dates: yes, on every figure
Fee figures were sampled 2026-10-02, route figures 2026-10-07, and each one links back to the source table. Where a number came from our own testing, the date is the day we ran it.
2. Links: checked, and two of them are flagged rather than hidden
Every outbound link was requested on 2026-10-08. One agent returned HTTP 503 across its whole site and another had moved to a new domain eight days earlier. Both still appear on this site, labelled, because removing them would hide the most useful thing we found.
3. Specifics: the grid, the clauses, the route names
The 55 cell grid is computed from the sample rather than described. The compensation figures are quoted from terms pages. The slow lanes are named.
4. Unhelpful findings: several, deliberately kept
The cheapest line needs a month or more in a sizeable share of cells. One category in our own catalogue is mislabelled upstream and we say so on the page rather than quietly relabelling it.
5. What we cannot know, and do not claim to
We do not rank agents on trust, because there is no reproducible public data on it. We also cannot tell you whether a sampled rate is still correct today, which is why every figure links to its live source.
Where we are weakest: the sample is one third party snapshot
The freight and fee numbers come from a single external source rather than from us placing parcels with thirty agents. We read it, computed from it and checked the agents were reachable, but we did not independently weigh and ship. Treat it as one well documented snapshot, not as our own laboratory.
We also earn a commission on some of the agent links here, which is stated in the footer of every page. The ordering on the agent list comes from how many cells each agent won in the grid, a number you can recompute from the same public source. If you find a case where our ordering and the data disagree, that is a bug and we would like to hear about it.