Appmint Social Media Management: Publishing, AI Replies and Lead Follow-up

Social Accounts and Integration Setup

Cedar & Form needs one place to publish a product announcement, answer the questions it attracts, and find the conversations that need follow-up. We will keep the same example throughout: an announcement for the Arc desk lamp, a visitor asking about its dimensions, and a later request for help. The useful result is a post on the intended account and a reply in the customer's actual conversation. A connected-account badge is only the starting point.

Use a business account you administer and a separate participant for the demonstration. Keep the organisation in Appmint consistent throughout. Have the approved lamp image, its actual product information, and a provider account with permission to manage the intended page. The example text in this course is a draft; replace product facts with your own checked information before publication.

Open CRM → Social Media. The page heading is Social Media Command Center. On a wide screen, Dashboard, Conversations, Posts and Feed are the four primary tabs. More… contains Activity, Engagement, Listening, Scheduled, Accounts and Integrations. A narrower window can change the layout, so use the view names to stay oriented.

Choose More… → Integrations to open the integration setup surface. Select the provider for the account you intend to use and complete its connection flow with the business administrator's identity. Provider authorisation screens belong to that provider and can differ by account. Inspect the pages and permissions being granted before returning to Appmint. Do not infer that every platform named somewhere in the application supports the same publishing or messaging operations.

Next choose More… → Accounts. This opens Linked Accounts, with Facebook, Google, TikTok, Pinterest, LinkedIn and Twitter account sections. Expand the relevant section and check the returned account name and identifier. In the Facebook section, examine Page Details and any linked Instagram Account. An Instagram business account is displayed with its linked page, so selecting an unrelated personal account at authorisation will not establish the business-page route we need.

Each account has sync icons with tooltips: All, Messages, Posts, Comments, Reactions, Engagement, Notifications, Insights, Leads and Ads. Choose the data relevant to your operation. For this example, All makes the relationship between the post, its comments and its messages easy to inspect. When narrowing scope later, choose the individual categories deliberately; selecting an individual category replaces an All selection with specific categories. Click the Save icon at the top of Linked Accounts to persist the settings. The adjacent refresh icon reloads profiles; it does not save your selections. Both currently share a misleading “Save Setting” tooltip, so identify the disk-shaped save control by its icon and verify persistence by returning to the page.

For Facebook and its linked Instagram account, inspect Receiving DMs & comments. There are two requirements: an active App-level callback subscription for the current Callback URL, and a subscribed row for the page receiving the message. One green page does not repair an inactive app callback. Click Re-check to refresh the report. Where setup is missing, Set up requests subscriptions; once configured, the control becomes Re-subscribe. Read any incomplete-setup message and the individual page errors.

Leave Callback host override (optional) empty unless the installation's public callback host differs from the server's current host. If verification times out, use the host configured for this installation in the Meta app console, then Re-check and Set up. Do not paste the placeholder URL as a universal repair. Instagram DMs also require message access on the Instagram account and the instagram_manage_messages permission. A publishing connection alone does not establish inbound DM delivery.

Open Sync in the Social header. It offers All platforms and the connected platforms, with last-run health where available. Start a sync for the chosen provider and reopen the dropdown to inspect the result. “Syncing…” confirms that a pull was requested, while failed or partial status requires investigation. The main Refresh button refreshes account settings; it is not interchangeable with the provider sync action.

The connection is ready for our lesson when the intended page is present, its settings survive a reload, the required inbound subscriptions are healthy, and a controlled incoming message appears in the correct conversation. If Linked Accounts reports a loading failure, use its retry or refresh control. Do not reconnect accounts merely because the current request failed: first distinguish unavailable data from an actual disconnected identity.

Social Post Creation, Publishing and Feed Review

Return to Dashboard. Choose the 7d, 30d or 90d period appropriate to the work you are inspecting. Platform performance keeps each platform's own measures: a platform may report likes and shares while another reports views or reach. Compare the same account and time window before drawing a conclusion. Missing figures are not zero engagement, and adding unlike metrics across platforms will not produce a meaningful total. Use the top-post and trend information to locate the underlying content you want to inspect.

Click Create Post. Appmint opens New social post in place, so you can prepare content without leaving Social Media. Choose the social channel first, then select the connected destination in Post to account. For our worked example, choose the Cedar & Form test page. Select only that account for the first run; multiple selected accounts create additional destinations to verify.

In Content, enter: “Meet the Arc desk lamp. See the finish and dimensions on our product page. Ask us here if you need help choosing.” Add the real product-page URL once you have checked it. Attach the approved lamp image using the composer's file controls. The selected channel determines the available fields and limits. If you switch to a channel with a title field, such as Pin title or Video title, complete that field for the new channel instead of assuming the Facebook text supplies it.

Retained consultation-post draft, separate from the Arc lamp exercise. No Facebook account is connected and the preview asks for an account. This demonstrates content entry and the missing-destination warning, not publication readiness. Connect your controlled destination before the publishing steps.

Inspect the preview beside the message. It is useful for reviewing the selected account, text and media together. It cannot establish that the provider accepted the content. In particular, check the destination after changing channels: an account appropriate to one platform is not an appropriate sender for another.

Choose Save draft first. Close the composer and open Posts, select the Draft status filter and the relevant platform. Open the saved row. This tests that the body, image and destination survived saving and gives a colleague a reviewable item. The modern composer is the normal authoring surface; the Raw schema form is an alternative for users who understand the message fields. It is not required to write this post.

The same historical consultation text was saved and reopened. Delivery remains Draft, with no attempts yet. Saved content and delivery are different checkpoints; neither this frame nor its filled composer proves the provider received anything.

When the content and destination are approved for publication, use Send now in the composer. Read the returned result. Then return to Posts and inspect Sent or Failed, as appropriate. Open View on platform on the row when a destination link is available. In the provider's own page, confirm the page identity, the full text, the image and the public link. That destination check closes the publishing task; the presence of a saved or queued message alone does not.

Open Feed and choose the connected platform. Feed gives you the account's synced posts; Posts gives you Appmint's authoring records and statuses. A post created elsewhere may appear in Feed without having originated as one of your Appmint drafts. A newly sent post may take another sync to appear in the provider feed. Use Sync, wait for the actual pull to finish and refresh the view before creating another copy.

For a failed send, reopen the original post and inspect its destination, channel and attachments. Restore expired provider access through Integrations if the error identifies authorisation. Correct missing channel fields or invalid media if the provider rejects the payload. Check the destination before retrying an ambiguous timeout, because a response failure is not conclusive evidence that no post was created. Retry the corrected original rather than making a second draft with no connection to the first attempt.

Your independent check is to save a second draft for the same page, change one sentence, reopen it from Posts and confirm the changed sentence is preserved. Leave it as a draft unless it is intended for publication. You should now be able to explain which view proves saved authoring, which view shows synced platform content, and which destination proves the audience can see the post.

Social Activity and Conversation Management

A post is the outgoing content. Activity is what comes back: comments, mentions, reactions and follows. Conversations groups the exchanges you can read and answer. Use these views together when the participant asks, “Where can I find the Arc lamp dimensions?”

Choose More… → Activity and locate the incoming item by platform, author, text and time. A reaction does not carry the same reply context as a comment or DM. For a replyable item, use its Reply action to move to Conversations. Appmint attempts to open the matching conversation and reply box. Confirm the participant and original post after that handoff, especially when one person has commented on several posts.

You can also open Conversations directly. Select the specific account instead of All accounts when the organisation has several brands or pages. Use Reload list for the conversation list. The provider sync action pulls fresh activity; list reload reads the stored conversations. If the expected conversation is absent, widen the account filter and use the available further-page control before concluding that delivery failed.

Retained disconnected training state: All accounts (0), no conversations. Use this image to locate the account filter and list, not as evidence of an incoming message, AI answer or successful reply. The Google service-agreement banner is separate from configuring the intended Facebook or other social account.

Open the thread and read the customer's message in context. Customer messages, your outgoing messages and AI-generated outgoing messages have different presentations. Use the AI marker to identify automation; do not assume that every outbound bubble was typed by a colleague. Open any relevant attachment to understand the question before answering. Treat a photograph of a label, for example, as evidence about the customer's item rather than a reason to repeat the generic product description.

Choose Reply and enter: “The dimensions are listed on the Arc product page. Here is the link: [your verified product URL].” Replace the bracketed text with the real URL. Add a relevant approved attachment if the answer requires one. Submit the reply only in the intended demonstration conversation. Appmint ties the reply to the incoming activity, allowing the server to resolve the page, recipient and reply context. This is why replying inside the correct thread matters more than merely selecting the correct platform.

After Reply sent, inspect the new outbound row and have the participant check the message in the original provider conversation. A public-comment reply and a private DM are different destinations. Match the reply to the same post or DM thread used for the question. Do not claim a public response was privately delivered, or a DM published beneath the post.

If sending fails, the conversation keeps an inline error rather than relying only on a disappearing notice. Read and retain that error while checking account access, the receiving page, the incoming activity and the provider's permitted reply context. Correct the identified cause, then retry from this thread. A sync can recover missing input; it does not necessarily repair an expired token or a provider restriction on replies.

If the list loads but the thread fails, reload the thread instead of answering from the list preview. A short preview may omit earlier context or attachments. If the Activity handoff opens no conversation, locate it directly in Conversations with the correct account and participant. The lesson's finish is an identifiable incoming item, a reviewed answer and an independently visible provider reply, all tied to the same exchange.

AI Social Replies and Human Takeover

The lamp question is a useful candidate for AI because the answer should come from known product information. A complaint requiring a staff decision is a different situation. We will configure that distinction where the assistant operates, then inspect its messages in Social Media.

Open AI, IVR, Automation → AI Assistant and create or edit the assistant intended for this business. In the editor, enter Name “Cedar social help” and Handle cedar-social-help. In What it is for, describe the actual task: “Answer questions about Cedar & Form products from approved information, and direct unresolved service requests to the support team.” Select a non-running status while preparing the configuration if your deployment offers it; use the available Status choices rather than inventing a separate test mode.

In How it behaves, set a concise Personality and add Rules it always follows such as “Use the approved product information. If a detail is missing, say that it needs checking.” Add a When… then… rule for a service complaint: when a customer needs a refund decision, explain that a person will review it. These instructions shape the answer, but a written rule is not a technical permission boundary.

Review Tools before enabling the assistant. Give it only the actions this example requires. The editor distinguishes no tools, every tool and a selected set. A product-information answer does not justify granting unrelated actions. What it is told it may do is instructional text; the allowed Tools determine which actions are available. Review the actual tools and their scope instead of relying on a reassuring sentence in the prompt.

Under Channels, choose the intended social channel explicitly. Under Accounts, add the real page/account identifier from Linked Accounts, not the friendly page name. An empty Channels selection means every channel; empty Accounts means all accounts. Those empty states broaden scope, so they are inappropriate for a narrowly scoped test assistant.

In What starts it, add the message-received event, message.received, and write instructions for handling incoming questions. This trigger is for conversational input. A like or follow is not automatically a customer question that should receive an answer. Under the knowledge-source controls, add the supported source containing the approved product facts: choose the actual collection where offered or supply the relevant reference URL for a URL source. Check the information itself; merely saving a knowledge reference does not guarantee it contains the dimensions the customer requested.

Expand Where it stops. Choose the actual staff member in Who it hands over to. Set After a person replies, stay quiet for to 30 minutes for this example. Zero keeps the assistant answering, so it would defeat our temporary human takeover. Leave the timeout appropriate to the installation and inspect it if replies repeatedly take too long. Save the configuration and set the assistant to the appropriate active status when the intended account is ready for the controlled exchange.

Have the participant submit a new question on that account. In Social Media → Conversations, locate the inbound question and inspect the AI-marked reply. Read its factual content against the configured source and check the receiving provider. A fluent answer is insufficient if it names a dimension that is absent from the product information. Correct the source or instructions and test a new question if the answer is wrong.

Now use Reply in the same thread as a person: “I am checking this with our team and will help you from here.” Submit another controlled incoming question within the 30-minute interval. The intended result is that the person's reply starts the quiet period and the assistant does not add a competing reply. Check the thread and receiver over the relevant interval; a momentary absence of an AI bubble does not establish the full handoff behavior.

The quiet period is temporary. It is different from a permanent per-conversation AI hold. The backend has a hold mechanism, but the reviewed Social Conversations screen does not expose a dedicated hold switch to teach here. Do not search for an invented “Pause AI” button. If the conversation must remain exclusively human beyond the configured interval, use the assistant's actual status or scope controls with awareness of their effect on other conversations, or use an exposed hold facility only when it is available and has been checked in your installation.

When the AI is silent unexpectedly, work outward from the input: did the incoming message reach this thread; does the assistant include this exact channel and account; is it active; is message.received configured; did a person recently reply; and did the reply fail? When it answers too broadly, inspect empty channel/account selections first. These checks distinguish a routing problem from an answer-quality problem without rewriting the entire assistant each time.

Social Post Scheduling and Recurrence

The announcement can be prepared now and sent later. Start with one future occurrence so you can verify timing and destination before introducing recurrence. Click Create Post, prepare the content and select the intended account. Enable Schedule for later. Give the schedule a recognisable title, such as “Arc lamp announcement.”

In What it does, select the action appropriate to a message: Send. In When → Run at, choose a near-future date and time for your controlled demonstration. The field is a local date/time input, so check the computer's displayed timezone and compare the resulting next-run time after saving. Do not assume that a time typed on one computer will display identically on another timezone's computer.

Retained Instagram timing rehearsal deliberately left STOP and still showing Create schedule. Its old date is a field example, not a future occurrence to copy. A filled date does not activate a stopped schedule or prove a queue entry. Review the intended active state when scheduling a real controlled publication, then inspect the saved schedule, run and provider result.

Click Schedule and open More… → Scheduled. Use Upcoming, Today, 7-day or 30-day windows and the platform filter to find the item. Inspect the message preview, account, time and schedule status together. A future date with the wrong account is still a wrong schedule. An item under Past (not yet sent) needs investigation rather than another schedule layered on top of it.

Open the scheduled row in Communications Center to manage the message and its linked schedule. For an already saved message, the composer includes the full schedule manager, with records and run history. At the due time, inspect the run result, the message status and the provider destination. A row appearing in Scheduled establishes a saved scheduling record, not by itself a queued worker job. Check the active state and next occurrence, then the actual execution history and provider result.

To create a repeatable announcement, use the schedule editor's Repeat control. The available modes are Minutes, Daily, Weekly, Monthly, Yearly and Custom. Choose Weekly for a controlled recurring reminder, select the desired day and time, and inspect the resulting expression and next occurrence. Custom accepts a cron expression for installations and users that need that precision. Do not use a fast Minutes repeat against a real audience just to make a lesson appear active.

Ends on supplies an optional end date, with a separate When it ends action. Choose it only when you intend that end-time behavior; ending a schedule is not the same operation as automatically removing a provider post. The action picker is shared with other record types, so a label such as Unpublish or Delete is not evidence that a particular social provider can undo an already delivered message. Use Send for this worked message and verify any advanced action on the relevant record type separately.

After saving, inspect Next run, Last run where available and the run list. A recurring row carries its recurrence/cron indicator in Scheduled. Follow the same message to its actual destination after the controlled occurrence. Confirm that a later next-run time remains only if further occurrences are intended.

Use Cancel schedule on an unwanted queued item. The confirmation says it removes the schedule but keeps the draft message. Confirm cancellation, reload the list, and reopen the retained message. This gives you a safe way to stop future execution without losing the reviewed content. It does not remove something that has already been published.

For a practical check, create a future draft schedule, verify its account and next occurrence, then cancel it before the due time. The successful result is a retained draft and no remaining schedule for that intended occurrence. For an execution failure, inspect the recorded run before changing the time: a provider-access error needs account repair, whereas the wrong local time needs a timing correction.

Social Engagement Analytics and Listening Topics

Choose More… → Engagement and select the period you want to investigate. Top engagers groups the inbound interactions, Sentiment divides classified activity, and Hashtag performance examines tags on your posts. Use a specific comment as a check on any aggregate: a chart is useful when you can connect it to the actual activity behind it.

The Needs reply panel currently describes recent comments and mentions, oldest first, and states Reply tracking pending. Treat it as a place to find items to review. It is not a complete verified list of unanswered customers. Open the actual thread before replying so you can see whether a colleague or assistant already handled it. The panel's inbox/reply actions lead to Communications Center; this differs from Activity's direct handoff to Social Conversations.

Now choose More… → Listening. We want to find questions about dimensions across the activity already synced into Appmint. Create a New topic with Name “Arc dimensions” and Description “Questions about the Arc lamp's size.” In Include patterns, type dimensions and press Enter; add the phrase how tall the same way. Each entry becomes a pill. Click a pill to inspect or change its matching mode and case sensitivity.

A keyword is useful for a single term; a phrase is useful for words that belong together. Regex is available for a pattern requiring explicit variation, such as /height|dimensions/i, but use it only when you can explain what it matches. A complex expression that nobody can maintain is a poor replacement for two clear keywords.

In Exclude patterns, add giveaway only if you intend to exclude the entire activity when that term appears. Exclusions are not cosmetic highlights: any matching exclusion skips the row. Select the relevant Platforms and Activity types to keep the topic focused. Empty selections mean all, which is useful for a deliberate cross-platform topic but broader than our initial test.

Keep Enabled — match during sync selected. In Test against a sample, enter “How tall is the Arc lamp?” and click Test. Inspect the matching term and field. Then try a non-matching sample and an excluded sample. A message saying the test could not run is an unavailable test, not a successful “No match.” The save button needs a name and at least one include pattern.

Save the topic and open it from the topic list. Inspect Mentions, Unique authors, Platforms, Top term, the trend and recent matching items. Select a time range appropriate to the incoming activity. To apply the current patterns to older stored activity, use the historical re-scan control with the tooltip Re-scan historical activities with current patterns. Then inspect the actual matched rows. Historical re-scan does not retrieve the whole public internet; it reprocesses activity available to Appmint.

If a known comment is missing, first confirm it exists in Activity. Next check the topic's enabled state, platform/type filters, date range and exclusions. Run its exact text through Test. If you changed patterns after the comment arrived, re-scan historical activity. This sequence tells you whether you have an ingestion problem or a matching problem.

Use Edit when the topic is too broad; use Disable when you want to stop matching without removing its configuration. Delete is available for a topic that is no longer needed. For independent practice, make a second topic for “delivery time,” prove one included and one excluded sample, and inspect a real matching activity. The defensible conclusion is that the saved topic finds relevant synced conversations, not that it measures every mention of your brand across the web.

Social Automation and Failure Recovery

A question can be answered in a conversation; a service request may need a separate record that staff can work. Keep those outcomes distinct. We will prepare a follow-up workflow for a social activity and trace its execution before relying on it for unattended work.

Open Automation, choose New Workflow, and enter a Workflow Name such as “Cedar social follow-up.” Use a recognisable slug and description so the workflow is identifiable in run history. In the builder, choose Add Trigger → Record Created. Inspect Data Type and select the Social Activity collection if it is offered in this installation. This is the exposed generic record route. The current trigger picker does not list a dedicated Social Activity Created control, so do not substitute an imaginary social trigger in the walkthrough.

Use the available condition tools, such as Check Field Value or If/Then Branch, to restrict the incoming record to the intended provider/account and relevant content. Read the actual incoming record fields before mapping them. A social comment may carry content while a provider-native payload may have message, caption or text; using the wrong field can make a condition silently irrelevant. Check the field path and comparison in the condition editor against the example record.

Add the supported Create Ticket action when the service team should receive a ticket. Supply the action's required fields from the selected schema, using the incoming question and identity where supported. Keep the original activity identifier in the resulting work item when a suitable field is available, so someone investigating the ticket can return to the conversation. Save the workflow as a draft configuration until its trigger path and mapping have been tested. A form offering an action does not prove all its required business data has been supplied.

For the controlled execution, use a new incoming provider activity with recognisable wording, then open Runs & Logs. Locate the run by time and workflow, inspect the triggering record, inspect the condition result and inspect the action output. Open the resulting ticket itself and check its content and context. A successful run status without the intended business record is an incomplete check. Also submit a non-matching example and confirm it does not create unrelated work.

There is an installation boundary here: the backend contains a native social_activity_created event path, but the reviewed visual picker does not expose that trigger directly. Its specialised filter path also needs live verification. A generic Record Created configuration is a candidate route to test, not a guarantee that every connector's sync or webhook creates the same event. If a new activity appears in Social Media but no workflow run appears, inspect the event route and trigger registration with the administrator. Do not repeatedly send customer messages or remove all conditions to hide the mismatch.

When a run exists but the condition fails, compare the stored input against the field and value used in the condition. When the action fails, inspect its error and repair its missing mapping or permissions. When the ticket appears twice, investigate whether two workflows, two trigger routes or duplicate incoming events created it before enabling unattended retries. Keep manual follow-up available while the installation's event path is being validated.

Finish the example by following the same lamp post through its operational chain. The destination proves publication. Activity proves that the incoming interaction was ingested. Conversations proves which customer and staff/AI messages are associated with it. The independent provider view proves the reply arrived. Scheduled and its run history explain later sends. Engagement and Listening help find relevant stored activity. Automation runs and the resulting work record prove the follow-up. If one link is missing, repair that link; a healthy dashboard cannot substitute for a missing customer outcome.