AI SDR
In short
Why is the AI ignoring a messaging rule I created?
Reach decides which rule wins. An organization rule outranks a workspace rule, a workspace rule outranks one set for your account, and an account rule outranks one limited to named sequences, regardless of whether a rule was written by you or by the AI, and a rule that loses is marked as overridden, naming the rule that beat it. Two rules with the same reach that say opposite things cancel each other out, and neither is applied until you choose between them.
Quick fix
- Open the rule. If it is marked as overridden, the label names the rule that is winning; change that one instead of re-saving yours.
- Confirm the rule is switched on. A rule can be disabled without being deleted, and a disabled rule looks normal in every other way.
- Look for an unresolved flag on the rule and on its opposite. Where two rules at the same reach disagree, neither is used until you pick one.
- If you created that pair in the last few minutes, come back to it shortly; conflicts are worked out in the background and the flag takes a little while to appear.
- If the rule is limited to named sequences, check that the sequence you are watching is one of the ones listed on it.
Didn't work? Contact Reply support; the team can check your account directly.
Symptom
The rule sits in the list, worded the way you meant it, and the AI writes as though it were not there. Nothing reports an error. Sometimes the wording is no longer quite what you typed.
Where rules come from, and how far they reach
Rules reach the list two ways. You write them yourself, or Reply writes them for you from what you have been approving, rejecting and editing; the AI turns that behaviour into standing instructions so you do not have to correct the same thing twice.
Each rule carries a reach, which is both who it applies to and how much authority it has:
| Reach | Applies to | Rank |
|---|---|---|
| Organization | Everyone in the organization | Highest – a hand-written one beats all others |
| Workspace | Everyone in the workspace | Above account |
| Account | Only you | Above sequence |
| Sequence | Only the sequences named on the rule | Lowest |
Two rules never merge. When they meet, one of them is set aside, and which one depends entirely on this ranking.
The six reasons a rule has no effect
| What happened | How it shows up |
|---|---|
| A rule with wider reach outranks it | Marked overridden, with the winning rule named |
| Two rules at the same reach contradict each other | Both flagged unresolved; neither is used |
| The contradiction has not been noticed yet | A new pair looks healthy for a few minutes |
| The rule is switched off | Still listed, no longer in effect |
| The automatic writer replaced your wording | The text is not what you typed |
| The rule names sequences you are not looking at | Works in one sequence, absent in another |
1. Something with wider reach outranks it
This is the usual answer. A rule set for the whole organization silences the equivalent workspace rule, a workspace rule silences yours, and yours silences a sequence-scoped one. Hand-written organization rules sit above everything, including anything written automatically.
How to recognize it: the rule carries an overridden marker, and the page tells you which rule is winning.
What to do: work on the winning rule rather than the losing one. Either soften the wider rule so it no longer covers your case, or accept it; a narrower rule cannot out-argue it.
2. Two rules at the same reach say opposite things
When two rules of equal reach pull in different directions, Reply does not guess. Both are flagged as unresolved and neither is applied: so the behaviour you get is the behaviour you would have had with no rule at all.
How to recognize it: two rules are flagged, not one, and the writing follows neither.
What to do: open the flagged pair and keep one. Editing the survivor to cover both cases usually works better than deleting one outright.
3. The contradiction has not been noticed yet
Conflicts are worked out in the background rather than the moment you press save. A freshly created pair can look perfectly healthy for several minutes before either is flagged.
How to recognize it: you created or edited the rules very recently, and nothing is marked.
What to do: give it a few minutes and reload. If the flags appear, treat it as the case above. If they do not and the rule still has no effect, the reason is somewhere else on this page.
4. The rule is switched off
A rule can be disabled and kept. Disabled rules stay in the list at their normal reach and read exactly like active ones.
How to recognize it: the rule's switch is off. There is no other signal.
What to do: switch it back on. If you are working through a list of rules that are not taking effect, check this first; it is the quickest to rule out.
5. The rule is stored but was never switched on, because the budget was full
Each reach has a ceiling on how many rules can be enabled at once, and the ceilings are separate: one for your organization, one for your workspace, one for your account, and one covering all of your sequence-scoped rules together. That last one catches people out: sequence rules do not each get their own allowance, they share a single pool across every sequence you have scoped a rule to.
When a ceiling is full, a new rule is still saved. It is not in force.
How to recognize it: the rule exists and looks correct, but is not switched on, and switching it on fails or silently reverts.
What to do: disable a rule you no longer need at the same reach, then enable the new one. Disabled rules keep their text, so nothing is lost by turning one off.
> Your own wording is safe. The automatic refresh only ever replaces rules it wrote itself and that are > currently switched on. A rule you typed is never touched by it, and editing a rule the AI wrote makes it > yours from then on, so it stops being a candidate for replacement too. If the refresh returns nothing at > all for a category, the rules already there are kept rather than cleared.
6. The rule is aimed at sequences it does not cover
A sequence-scoped rule applies only inside the sequences named on it. It has no effect anywhere else, and nothing marks it as inactive in the sequences it does not cover.
How to recognize it: the rule behaves correctly in one sequence and appears to be ignored in another.
What to do: add the other sequences to the rule, or raise its reach to your whole account if you want it everywhere.
Who can set a rule for everyone, and what the limits are
Setting a rule for a whole workspace or organization needs permission. If those reaches are not offered to you, your admin has not granted it. Your own rules and sequence-scoped rules are unaffected.
The interface enforces a maximum on the length of a rule's text, and on how many sequences a single sequence-scoped rule can name. Both are shown where you type them. If you are pushing against either, splitting one long rule into two focused ones reads better anyway.
A rule created by someone who has since left the company can only be removed by that person. This is current behaviour, and it means an inherited rule can outrank yours with no way to clear it from your side. Contact support and they can deal with it on the account.
FAQ
Why does my rule say overridden?
Because another rule covering the same ground has a wider reach. The marker is informational, not an error; the rule is intact and would apply again if the wider one changed. The page names the rule that is currently winning so you know where to go.
Who can create a rule for the whole team?
Anyone your admin has granted that permission. Without it you can still create rules for yourself and for individual sequences, and those work normally; they are outranked whenever a broader rule covers the same point.
Why did my rule change on its own?
Reply refreshes rules from your recent approvals, rejections and edits on a schedule, and that refresh can rewrite a rule you set by hand. If a rule matters, check it again after a round of approvals.
Still stuck? Contact Reply support
If these steps didn't solve the problem, the Reply support team can look at your account directly. Open the Reply Help Center and send the team a message.
To get an answer faster, include:
- the exact wording of the rule and the reach you set on it
- whether it is marked as overridden or flagged as unresolved
- an example message that should have followed it and did not
- whether the rule was written by you or generated from your approvals
Don't send passwords, API keys, or access tokens.
Related
- Why did my generated playbook disappear? – the other place your messaging is defined
- Why does the email not look right?
- Why can a teammate not see a sequence or mailbox? – permissions across a team
Build with Reply
- REST API: AI SDR learnings – list, create, change and remove messaging rules, including their reach
- MCP: docs.reply.io/mcp – no messaging-rule actions today; use the API