skills/usertold-recruit-participants/references/channel-selection.md

# Channel selection

Choose the narrowest channel that reaches people with the target behavior. Name a contrasting channel to expose what the primary channel misses.

| Channel | Best fit | Common bias |
|-|-|-|
| In-product passive launcher | Spontaneous feedback and bug reports | Overweights active users who notice and choose the launcher |
| In-product contextual panel | Recent behavior on a relevant route | Misses people who abandoned before returning to that route |
| Direct email or CRM list | Churned, frustrated, or otherwise absent users | Overweights reachable contacts and people willing to reopen the relationship |
| Customer community | Engaged users and community/charity offers | Overweights enthusiasts and frequent community participants |
| Direct link shared by support or success | Known incidents or account workflows | Overweights escalated and managed accounts |
| Early-access cohort | New-feature and concept research | Overweights novelty-seeking, product-engaged users |

Write the bias note as a concrete limit: “This sample comes from customers who replied to a churn email, so it underrepresents unreachable customers and people unwilling to re-engage.” Do not weaken it to “results may be biased.”

Use the primary channel for the first recruitment wave. Use the contrasting channel to test whether findings depend on reachability or context, not to imply statistical representativeness.

## Source tracking

Assign channel-level tags before distribution:

```text
utm_source = product | lifecycle-email | support | community | early-access
utm_medium = launcher | contextual | email | direct-link | post
utm_campaign = stable-study-wave-label
```

Keep the opaque `ut_research` parameter generated by UserTold unchanged. Add only non-identifying campaign tags supported by the host flow; never put an email address, account ID, name, or qualification answer in a URL. The canonical generated UserTold URL guarantees only its opaque recruitment reference—do not claim the widget automatically records added UTM tags unless the live host submission path has been verified to do so. Otherwise keep the tags in the distribution ledger. Compare invited, started, qualified, and completed counts by source only when those counts are actually available, and retain the sampling-bias note in the readout.