1. Every number opens into the thing it was counted from
If a cell says their entry price is a certain figure, the link goes to the product page that says so. A median opens into the catalogue it was calculated from, dearest first. A hook count opens into the creative it was counted across.
This is the whole product, really. A framework you cannot check is a decoration, and a number you cannot trace is a rumour with a decimal point in it. Everything else on this list follows from taking that one rule seriously enough to be inconvenienced by it.
2. Estimates are labelled, and never enter a score
Some things can be measured from outside: their price, their review count, what their site says today, how many pages they have indexed. Some things can only be modelled: traffic, ad spend, revenue.
Both are useful and they never look alike on the page. Modelled figures carry the estimate label in the data and in every row they appear in, and they are kept out of anything that gets summed or ranked. One unlabelled estimate inside a total makes the whole total unarguable, and the reader has no way to tell which number did it. Where a figure genuinely cannot be known, we publish a band with its assumptions instead of a confident number.
3. Both sides get collected the same way, or the difference is not reported
The most common way a competitor report goes wrong is invisible: your side read one way, their side read another, and the gap between them is partly the gap between the two methods.
So the rule here is that a difference is only reportable when both sides were collected identically, in the same run. Where a source samples rather than counts, we can say how many but not which, and we say so. A diff of a sample measures our own collection, not the brand, and a tool that hides that is measuring itself and charging you for it.
4. What is missing is written down
Every report has holes: a source that failed this week, a tracker that has no data for this brand, a measure that does not apply. The tempting thing is to let the section quietly not appear.
We name them instead. A missing source is stated with the reason, a first measurement is announced as a first measurement rather than dressed up as "no change", and a section that cannot be supported says why it is absent. A gap you can see is information. A gap that has been tidied away is a lie with good manners.
5. Seventy-four watchdogs whose job is to stop us
Every run is checked by seventy-four automated tests before anything is published, and they exist because plausible output is more dangerous than a crash.
A crash is obvious. A report that looks measured, reads well and is quietly wrong is not, and it will be believed. So there are checks for a brand whose collection has silently frozen, for a scale that cannot be compared across brands being compared anyway, for a figure on the front page that disagrees with the report it describes, for a name-matching source that catches the wrong company. One of them caught a brand whose data had been frozen for twelve days while every page still looked perfectly healthy.
They block the publish. That is the point of them.
6. The advice is not for sale
Where a recommendation names a tool, four rules apply and they are enforced by a check rather than by good intentions: the free way to do it comes first, only one tool per recommendation, any relationship is disclosed in the sentence itself rather than in a footnote, and a tool is never named beside numbers in a report section.
There are no affiliate links. Not because it would be illegal, but because the moment there is money behind a recommendation, the doubt spreads to everything else on the page, including the figures we just went to the trouble of making clickable.
7. It happens every week whether anyone remembers or not
A competitive analysis done once has a shelf life of about six weeks, and the dangerous part is that it still sounds right afterwards. Prices move, tiers appear without an announcement, messaging drifts, an ad starts and stops.
Running weekly turns that from a document into a line, and the line is the asset. Week one is nearly worthless and month six is worth a lot, which is the only honest argument for starting now rather than later. Nothing is deleted, so the archive only grows.
Now the parenthesis
We do not have enough customers to know that any of this is why they stay. Presenting a wall of quotes at this stage would break the first rule on the list, and we would rather keep the rule than have the wall.
What we do have is a list of things we were told we had got wrong, and this list is mostly the result. Rule six exists because four people on a tester panel read a tool suggestion as hidden advertising when there was no relationship at all to disclose. Rule four exists because a first measurement was once presented as though nothing had changed. Rule five grew by one every time something looked measured and was not.
So treat this as a statement of what we are optimising for, and hold us to it. If a number on your report does not open into its source, that is a bug and we would like to hear about it. The version of this page worth reading is the one written by customers, and it does not exist yet.
Pick one competitor. See what we find.
Your first benchmark is free. One domain, measured against you, no card and no call.