Create a BrandJet AI Mention Reply Agent: step-by-step screenshots
Follow the current BrandJet interface below. Each screenshot matches the numbered action directly beneath it.
Step 1: Open AI Agents and review the active automations.
Step 2: Select Create Agent to start a new workflow.
Step 3: Choose Mention Reply, then continue.
Step 4: Describe the exact public reply job and when the agent must not act.
Step 5: Add tone, facts-only, escalation, and human-approval guardrails.
Step 6: Review the full brief before selecting Generate Agent.
Create a Mention Reply Agent in BrandJet
Outcome: You will create a BrandJet AI Mention Reply Agent that watches eligible brand mentions, drafts or sends replies according to your configuration, and uses the linked social account you assign to it.
Prerequisites:
- Access to BrandJet AI Agents
- A social account linked to BrandJet that the agent is allowed to reply from
- A clear idea of which mentions should receive a reply
- A decision on whether replies should require human approval or can be sent automatically
BrandJet's current AI Reply Agent supports reply configuration around brand voice, connected accounts, platform and sentiment conditions, and approval versus automatic handling. The safest starting point is narrow conditions with human approval enabled.
1. Open AI Agents and create a Mention Reply Agent
Open BrandJet AI Agents in your authenticated workspace and start creating a new Mention Reply Agent.
Use a Mention Reply Agent when the job is specifically to respond to relevant social mentions. Keep the scope narrow. Do not try to make one agent handle every social, support, sales, and community task at once.
Before continuing, be able to describe the agent's job in one sentence.
For example:
Respond to relevant BrandJet mentions with useful, on-brand replies while sending uncertain or sensitive situations through human review.
That sentence becomes the standard against which you judge the rest of the setup.
2. Select the linked account the agent can reply from
Choose the social account the agent is authorized to use for replies.
This is an important boundary. The account you select determines which brand identity the agent can speak through, so do not treat account selection as a routine setup detail.
Before proceeding, confirm:
- the account belongs to the correct brand;
- your team actively manages it;
- the agent should genuinely be allowed to reply from it;
- a human knows who owns conversations that need escalation.
BrandJet's public AI Reply Agent documentation describes linking the social accounts the agent should reply from. (BrandJet AI Reply Agent)
If the required account is not connected yet, complete the account connection workflow before activating the agent.
Choose the account authorized to reply
Use one clearly authorized brand account first.
Add broader account coverage only after you have verified that the agent behaves correctly on the initial account.
3. Define the agent's role
Set the agent's role around one concrete responsibility.
The role should answer:
What is this agent responsible for doing?
Avoid vague instructions such as:
Manage social engagement.
A stronger role is:
Respond to relevant brand mentions, acknowledge the person's point, answer straightforward questions when the available context supports the answer, and route uncertain situations through review.
The role defines the job. It should not contain every tone preference or every safety rule.
Good role test
Ask three questions:
- Can another team member explain what this agent is for?
- Can you tell when the agent acted outside its job?
- Can you decide which mentions should not reach this agent?
If not, make the role narrower.
4. Define the persona
Use the persona to describe who the agent should sound like.
For example:
A knowledgeable BrandJet product team member who understands social listening and B2B outreach, answers clearly, and helps before selling.
Persona is about perspective and character.
Keep hard boundaries out of this field when those rules belong in instructions. "Helpful product expert" is persona. "Never invent a feature" is an instruction.
A useful persona should make replies feel consistent without forcing every response into the same wording.
5. Choose the tone
Set the tone for how replies should be delivered.
For a lean B2B brand, a practical starting tone might be:
Friendly, concise, calm, direct, and useful.
Keep this simple. Tone should shape delivery, not define business logic.
For example:
- "Concise" controls length.
- "Calm" helps with tense conversations.
- "Direct" discourages unnecessary filler.
- "Useful" keeps the response focused on the person's actual point.
BrandJet's public AI Reply Agent page describes configuring replies around communication style and brand voice. (BrandJet AI Reply Agent)
6. Write the operating instructions
Use instructions for rules the agent must follow.
This is where you turn your brand and safety requirements into explicit boundaries.
A practical starting instruction set could be:
- Address the person's actual point before mentioning BrandJet.
- Do not invent product capabilities, pricing, integrations, policies, or customer results.
- Do not argue with negative commenters.
- Keep replies proportionate to the original mention.
- Avoid aggressive promotion when a simple acknowledgement is more appropriate.
- Send complaints, uncertain factual questions, and sensitive situations through human review.
- If the available context is insufficient, prefer review over a confident answer.
The distinction between fields matters:
| Setting | What it controls |
|---|---|
| Role | What job the agent performs |
| Persona | Who the agent sounds like |
| Tone | How the response feels |
| Instructions | What the agent must or must not do |
If the agent later produces poor replies, this separation makes troubleshooting easier.
7. Set the platform conditions
Choose the platform conditions that determine where the agent should operate.
BrandJet publicly documents platform-based reply conditions for its AI Reply Agent. (BrandJet AI Reply Agent)
Use only the platform options that are actually available in your authenticated BrandJet setup.
Do not select every available platform simply to maximize coverage.
For each platform, ask:
- Does our team actively manage this account?
- Do we have enough context to reply responsibly?
- Do people expect brand participation there?
- Can someone take over when a conversation becomes sensitive or complex?
Match platform conditions to the reply surface
Enable the smallest platform set that covers the conversations you genuinely want the agent to handle.
Broader coverage should come after quality is proven.
8. Set the sentiment conditions
Configure the sentiment conditions that help determine which mentions qualify.
BrandJet's public feature page identifies positive, neutral, and negative sentiment as examples of reply conditions. (BrandJet AI Reply Agent)
Treat sentiment as a routing signal, not as permission to publish automatically.
A practical starting policy is:
| Mention type | Starting action |
|---|---|
| Positive praise | Draft a short acknowledgement |
| Straightforward neutral question | Draft and require review |
| Negative feedback | Require human review |
| Complaint | Require human review |
| Ambiguous or sarcastic mention | Require human review |
| Irrelevant keyword match | No reply |
Sentiment classification cannot replace judgment. A positive mention may still be inappropriate to automate, while a negative mention may contain a valid question worth answering carefully.
Start narrow
A conservative first configuration could use:
- one linked brand account;
- only the platforms your team actively manages;
- clearly positive or straightforward neutral mentions;
- human approval enabled.
That gives you a controlled sample before expanding the agent's scope.
9. Choose the approval policy
Decide whether the agent's replies require human approval or can be sent automatically.
BrandJet's current product materials describe both approval-based and automatic reply workflows. (BrandJet AI Reply Agent)
For a new agent, start with approval.
Use this rule:
Routine + repeatable + low downside = possible automation candidate
Ambiguous + negative + sensitive + high visibility + factual uncertainty = human approval
Keep human review for:
- customer complaints;
- negative feedback;
- sensitive topics;
- unclear intent;
- factual questions where the answer may depend on current product details;
- high-visibility conversations;
- anything with meaningful reputation risk.
Automatic sending should be earned through observed quality, not enabled simply because it is available.
10. Review the configuration before activation
Before activating the agent, check the complete setup in order:
- Correct linked account
- Narrow role
- Clear persona
- Appropriate tone
- Explicit operating instructions
- Correct platform conditions
- Correct sentiment conditions
- Appropriate approval policy
A good configuration should be easy to explain.
If your explanation sounds like "the agent handles social media," the scope is probably still too broad.
11. Activate the agent and verify its Activity
Activate the Mention Reply Agent, then use its Activity area to confirm what it is doing.
BrandJet's AI Agents overview documents per-agent activity as the place to review what an agent has done.
Start with a mention that clearly matches your configured conditions.
For each early example, compare:
Expected behavior: What should the agent have done?
Observed behavior: What action did it actually take?
Correction: Which configuration layer should change if the two do not match?
Check two things separately:
- Trigger quality: Should this mention have reached the agent at all?
- Reply quality: Was the resulting response acceptable?
This distinction is important. A well-written reply to an irrelevant mention is still a configuration problem.
12. Review a small live sample before expanding automation
Do not judge the configuration from a single good result.
Review a small range of real mentions with different wording and sentiment.
Look for recurring problems such as:
- irrelevant mentions qualifying;
- replies sounding too promotional;
- responses that are too long;
- incorrect assumptions;
- unsupported factual claims;
- negative mentions receiving overly casual responses;
- questions that should have been routed through approval.
If the same problem appears repeatedly, fix the relevant configuration layer before widening the agent's scope.
13. Pause the agent when the configuration needs correction
Use the agent's pause control when continuing with the current setup could create more poor replies.
Pause when:
- the agent repeatedly responds outside its intended scope;
- the tone is consistently wrong;
- irrelevant mentions keep qualifying;
- replies contain unsupported claims;
- negative conversations are being handled poorly;
- the linked account context changes;
- a sensitive public situation requires manual handling;
- you are making substantial changes to the configuration.
Change one variable at a time where possible, then resume with human approval still in place.
FAQ
The agent sounds wrong
Check persona and tone first.
Persona controls who the agent sounds like. Tone controls how the reply is delivered.
The replies are too promotional
Review the role and instructions.
Add a clear rule to help first and avoid forcing BrandJet into conversations where a simple acknowledgement is more appropriate.
The agent makes unsupported claims
Strengthen the instructions.
Explicitly prohibit invented product capabilities, pricing, integrations, policies, and results. Keep uncertain factual questions under human review.
Irrelevant mentions keep reaching the agent
Review the platform and sentiment conditions and your underlying mention-monitoring setup.
If the monitoring signal itself is too broad, fix that before relying on the agent.
See BrandJet Mentions help for the monitoring side of the workflow.
Negative mentions feel too risky
Keep those mentions under human approval.
Do not expand automatic sending to a category simply because the agent handles positive mentions well.
The agent produces good drafts, but you are not ready for automatic replies
Keep approval enabled.
Useful drafting and autonomous publishing are separate decisions. The agent can still reduce manual work while a person remains responsible for the final public response.
When should you allow more automation?
Expand the agent only when the evidence from Activity is consistently good.
Look for a repeatable pattern where:
- the right mentions qualify;
- the agent stays inside its role;
- replies follow the intended persona and tone;
- factual boundaries are respected;
- reviewers rarely need substantial changes;
- the category has low downside if a response is imperfect.
Then consider allowing more automation for that specific pattern while keeping higher-risk categories reviewed.
The operating principle is simple:
Start narrow, review real activity, correct the configuration, and automate only what has proved predictable.
If your brand monitoring still produces too much noise, improve the BrandJet Mentions setup first. If the agent is ready, return to BrandJet AI Agents, keep the first configuration tightly scoped, and expand only after the activity history shows that the agent is handling the right conversations safely.