Guide
AI does not skip Meta’s rules.
An AI reply is still an API send. It needs an eligible comment private-reply path or an applicable open conversation window. Account permissions and credit limits also apply.
3 minute read
What DripDM adds
You set a goal, instructions, a knowledge note, a length cap, and a fallback line. The default length cap is 300 characters.
Emoji-only text and long repeated characters are skipped. The keyword’s fixed text can still send.
Separate the writing model from sending permission
AI composes text after an eligible rule matches. It does not grant a new audience, create a messaging window, or make a restricted recipient available. The same comment-private-reply and conversation rules still apply. A comment-led opening message needs a response from the person before ordinary follow-ups become eligible. The business remains responsible for what the generated text says, especially prices, policies, personal advice, and promises. Treat automation as assistance with a bounded reply, not an independent authority to negotiate.
A workflow to try
For a shipping question, provide the current service areas, expected fulfillment wording, and the support link in the knowledge note. Give the rule a goal to answer from those facts and direct missing-order disputes to a person. Test an ordinary question, an unclear question, and an instruction to invent a discount. Then inspect actual replies after launch. These checks do not prove hallucinations are impossible, but they are more useful than relying on a claim that a knowledge note guarantees correct answers.
Checks before you change the setup
Disclose automation where appropriate for your audience and keep a clear way to request human help. Pause or disable overlapping rules before a person handles a sensitive thread; this review did not verify a special automatic takeover timer. Free and Pro have no AI entitlement. Growth and Business still depend on credit availability and provider operation. Avoid guarantees of account safety, Meta endorsement, or perfect adherence to every written instruction.
State the assistance boundary to the customer
A concise automated answer can explain the supported facts and offer a route to a person when an exception is needed. Do not write a disclosure that suggests the model is a licensed advisor, a human teammate, or a platform-approved authority. Keep any required disclosure aligned with your business and applicable rules. Read the message with its surrounding conversation so a technically true sentence does not imply a completed action that never occurred. For example, sending a support URL is different from escalating a ticket, and sending a booking URL is different from confirming the appointment.
Questions
No. The person still has to contact you. AI only writes the reply.
No. Instructions affect wording, not platform permissions or dispatch checks. Even a perfectly written answer can be held or rejected when the send is ineligible. Use a fresh inbound interaction and the correct account connection before testing the model response.
Yes. Keep disputed terms, personal advice, and exceptions under responsible human review. Maintain knowledge and test difficult requests; the model’s fluency and official connection do not prove factual accuracy or legal suitability.