We audited our own index. 91% of movement claims did not survive.
Every keyword tool in legal marketing hands you numbers that look exact. They are not. Before publishing anything from our own data, we went looking for how wrong it might be. We found six problems — four of them in our own work. Here they are, along with the rules we adopted so they cannot happen again.
1. Google rounds its numbers off. Heavily.
We had 44,280 rows of search data. In all of it there were only 35 different numbers.
That is because Google does not tell you a keyword got 143 searches. It rounds to the nearest step on a fixed ladder — 10, 20, 30, and so on up through 1,600, 2,400, 3,600. Down at the low end those steps are 20 to 30% apart.
So when a small keyword “jumps 25%,” it may have simply moved one step up the ladder while nothing changed in the real world. If a tool is showing you month-to-month swings on low-volume keywords, a lot of what you are seeing is the rounding, not your market.
2. We were counting the same keyword twice
“Car accident lawyer” and “lawyer for car accident” came back with identical numbers in all 205 markets, every month for a year. Not similar. Identical, to the digit.
Google treats those two phrasings as the same search and reports one number under both names. We had been adding both together — which inflated our own totals by 13%.
We removed one of them. Now, before any keyword joins a total, we check whether some other keyword has been shadowing it with the exact same twelve months of numbers.
3. The 68% surge that never happened
One month, our national numbers jumped 68%. It looked like a genuine spike in people needing lawyers.
We took it apart. Three keywords accounted for the entire jump. Meanwhile “personal injury lawyer near me” dropped 28% in the very same month.
That cannot be real. The same people, in the same month, typing almost the same words, cannot triple one phrase and cut the other by a third. It was three keywords stepping up Google's rounding ladder at once.
So we made a rule: we never publish raw search volumes, and we never publish month-to-month change on generic keywords. Generic data only appears as a seasonal shape averaged over ten years, where one bad month cannot fool us.
4. Ten years of history we had to throw away
Google Trends does not give you real numbers. It scores everything from 0 to 100, where 100 is the busiest point in whatever time range you asked for.
We asked for ten years at once. That made the busiest recent month a 100 and squashed everything older down near zero — one keyword averaged 1 for the whole of 2018. When you do maths on numbers that small, you get nonsense: our first seasonal chart had a month scoring 533.
The fix was to ask Google for one year at a time, so each year gets the full 0-to-100 range to itself. We also switched from averages to medians, so one strange year cannot drag a whole month off course.
This is also why our seasonal chart cannot tell you if injury search is growing. Asking year by year is what reveals the monthly shape — and it is exactly what destroys the ability to compare one year against another.
5. We were about to publish 853 “rising” and “falling” claims. Only 72 held up.
This is the one that stung.
Our index was set up to show a little green arrow when a firm was rising and a red one when it was falling. 853 firms qualified for an arrow.
Then we asked a simple question about each one: could this movement just be Google's rounding? For 777 of the 853, the answer was yes. Nine out of ten arrows were meaningless.
We deleted them. Today a firm only gets an arrow if it averages more than 500 name searches a month — 72 firms clear that. Between 150 and 500 we show your ranking but no arrow. Below 150 we tell you plainly that you are under our measuring floor, because at that size “your brand grew 25%” is a fact about Google's rounding, not about your firm.
6. Three smaller things, including one still open
Duplicate firms. Thirteen firms were in our system twice under slightly different names, splitting their own numbers in half. Merged, with the history kept.
A February that was not really slow. In our twelve-month data February looked like a sharp drop, until someone pointed out February has 28 days. Fewer days, fewer searches. Divide by days and the drop disappeared. (The ten-year data still shows February genuinely quiet — but for real reasons, not calendar maths.)
Still open, and we would rather say so. Some firms are named after ordinary words or local nicknames. A firm called Sears picks up people looking for the department store. A firm named after a city's nickname picks up people searching for the city. We have not solved this yet, so where a firm's name is an everyday word, treat its ranking as approximate until we do.
The rules we now follow
No rising or falling arrows for firms under 500 name searches a month. No month-to-month change on generic keywords, ever. No claims about demand growing or shrinking until we have two full years. When we delete data, we archive it with the reason. Every number we publish says which months it covers.
And the one that matters most: where our tools cannot tell a real change from rounding, we say so instead of printing a number anyway.
Why publish our own mistakes? Because an index that wants to be quoted should show what it got wrong before it shows what it found. BrandTerritory shares ownership with Mass Tort Ad Agency, which sells mass tort advertising — not brand advertising. A firm can pay to be included in the index. No firm can pay to change where it ranks. Methodology · The concentration study.