What to measure when there's no attribution
Community work rarely shows up in your dashboard. The four proxies that do, how to read each one, and what not to put in a board deck.
Community work rarely appears in your attribution, and the usual response — buy a better attribution tool — solves nothing. People read a thread on their phone, forget about you for three weeks, and arrive by typing your name into Google. Your dashboard calls that direct traffic and gives the credit to nobody. The answer is not to chase the click. It is to measure four things that move before revenue does, and to be honest about what each one can carry.
Branded search is the cleanest proxy there is
If more people are typing your company's name into a search box this month than last, something upstream caused it. Nothing else you can measure is that close to "more people know who we are".
Read it in Google Search Console, under Performance, with a query filter for your brand name and its obvious misspellings. Two things to know before you trust the line:
- Search Console holds 16 months and no more. Export it monthly or you will lose the baseline you want to compare against next year.
- The last two to three days are always incomplete. A Monday report that shows the weekend collapsing is reading unfinished data, not a collapse.
Compare like for like: this month against the same month last year, or a 28-day window against the previous 28 days. Week-on-week on a small base is mostly noise.
Unpaid mentions are worth counting by hand
A mention is somebody naming your product in a thread that nobody was paid for. On a dev-first product the number is small enough to count manually, and counting it manually is the point — you read the context, which a tool cannot give you.
Keep one row per mention: the date, the link, the subreddit or account, and whether the person had used the product or only heard of it. That last column is the one that tells you whether awareness is turning into use.
Reply quality turns before the numbers do
The replies under a post are the earliest signal in this whole list, and the only one that is qualitative on purpose.
- Questions about how it works — somebody is evaluating you.
- Comparisons to a named alternative — you have entered a category in their head.
- Silence on a post with reach — the audience was wrong, or the post was an ad wearing a story.
- Dismissals about the company rather than the product — a reputation problem, not a reach problem.
Returning members, not member count
In an owned space — a subreddit, a Discord — the number that matters is the fraction of people who post in two consecutive weeks. A room that grows while that fraction falls is a room filling with people who will never come back.
Take the baseline before the work starts
Every number above is a comparison, and a comparison needs a before. Take it in the week you sign, not in the week somebody first asks how it is going — by then the only honest answer is that nobody wrote down where you started.
- 1Export 12 months of branded queries from Search Console: query, clicks, impressions, month.
- 2Count the unpaid mentions you can find from the last 90 days. Searching your product name on each platform takes an afternoon and gives you a real starting figure.
- 3Write down the size of every owned space and the posting-in-two-consecutive-weeks fraction, if you can get it.
- 4Note what else is happening in the next quarter — a launch, a funding announcement, a conference. These are the things that will be confused with your channel work later, in both directions.
That fourth one is the one everybody skips and the one that decides whether the report can be believed in six months.
What to distrust
- Impressions. Precise, freely available, and almost unrelated to outcome.
- Follower count. Buyable, and therefore evidence of nothing.
- Last-touch attribution on a long sales cycle. It will credit Google Ads with everything, every time, because the last click before a purchase is usually somebody searching for a name they already knew.
- Any single post's numbers. One thread that did well tells you about that thread.
How far to trust the proxies
Far enough to decide what to do more of. Not far enough to put in a board deck as revenue.
The honest sentence is: branded search is up over the quarter, unpaid mentions doubled, and this work is the largest thing that changed. That is a real claim, and it survives being questioned.
"Reddit generated $180k" is not a claim. It is a guess with a currency symbol in front of it.
We report against whichever of these were agreed at the start, monthly, next to what was spent. If a number moved for a reason that had nothing to do with us — a launch, a funding announcement, somebody else's viral post — that goes in the report too. A measurement you only trust when it flatters you is not a measurement.
If you want the other half of this, what we do about AI answers is measured differently again: there is no click to attribute at all, so the question becomes whether you are named, and by which engine.