A social media agency should collect account access, goals, audience, brand voice, usable assets, content boundaries, approval owners and reporting expectations before drafting posts. AI can turn the checked answers into a welcome pack, calendar brief and reminder list. A human account lead must verify every claim, permission and date, then get the client to approve the operating rules.
The first deliverable is an agreed working arrangement, not a full month's captions. If the client has not named a final approver, supplied image rights or decided who answers comments, a polished AI pack can disguise a real gap. Mark those items as pending, give each an owner and keep the first publication date provisional until the blockers clear.
Collect decisions before collecting creative files
Start with the signed scope: which channels, how many posts or stories, whether the agency handles comments and messages, and what is excluded. Extract these terms into a short intake record for the account lead to check. Do not ask AI to interpret an unclear contract clause as permission to do extra work. Resolve the ambiguity with the client. Agency AI clauses and approval terms deserve a separate discussion when the engagement does not already cover them.
Ask for one business goal per channel. “More engagement” is too vague to guide creative work. A café might want more weekday lunch enquiries; a photography studio might want qualified booking enquiries; a driving school might want a clearer explanation of its lesson process. Record the action a useful post should prompt, the audience it serves and how the client will recognise an enquiry. Do not promise that a number of posts will produce a particular number of sales.
Next map the real decision makers. Who can supply facts about products or services? Who checks compliance or sensitive claims? Who gives final creative approval? Who is the emergency contact if a scheduled post becomes unsuitable? Record a primary and a backup for each role. If one person wears all three hats, state that explicitly and agree a response deadline that fits their availability.
A compact intake record can use these fields: channel, handle or page reference, owner, agency permission, commercial goal, audience, evidence for claims, asset owner, approver, approval route, feedback deadline, publishing window, message-response owner, report recipient and exit process. The field labelled “evidence for claims” prevents an AI caption from turning an informal selling point into an unsupported guarantee.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowRequest access through roles and test it
Ask the client to grant the smallest role that lets the agency do the contracted job. Keep a private access register showing platform, permission, granting person, agency user, grant date and removal owner. Never put passwords, recovery codes or private tokens in an AI prompt. LinkedIn's help centre, for example, says there are no separate login credentials or passwords for Pages: admin roles are assigned to personal profiles, and sharing login credentials breaches its community policies. Its three Page admin roles separate full control (Super admin) from creating content (Content admin) and reading analytics (Analyst).
For an illustrative access error, the client adds an agency designer as an Analyst on its LinkedIn Page when the contract requires publishing. The designer can inspect figures but cannot carry out the intended content task. The account lead should request the appropriate role from the client's existing admin and record the change. Trying to work around the gap with somebody else's login creates poor accountability and can breach platform rules.
Test each channel before the production calendar is approved. Can the right person see the Page or account? Can a draft be prepared? Can the approver see a preview? Can the report owner get the agreed measurements? Keep the test harmless and unpublished. A role may appear granted in an invitation but remain unaccepted, or a connected scheduling tool may have expired permission. Set a specific retest date for any failed connection.
Give the client an exit promise in operational terms: when the engagement ends, the agency will hand over the content files and reporting record specified in the agreement, then remove its users and connected tools after the client confirms continuity. Avoid implying that a social platform account belongs to the agency because it created a scheduler workspace. The owner should retain control of the channel.
Ask brand questions that can actually guide a caption
A useful brand voice questionnaire asks for two approved examples and one rejected example, not five abstract adjectives. Ask how the business addresses customers, which claims need evidence, which words it avoids, how it handles complaints, whether it uses humour, and how direct a sales invitation can be. Ask for spelling and naming conventions, product names and the current offer sheet. If older posts conflict with the current answers, the new answers win.
For an illustrative café, the client writes: “Warm and concise; describe ingredients precisely; no claims that a dish is suitable for every allergy; use ‘book a table’ rather than ‘grab a spot’; never call a weekday offer permanent.” An AI draft of the voice guide might say “Playful, foodie and casual; highlight allergy-friendly choices.” The account lead should correct that output. “Allergy-friendly” was never approved, and “playful” might invite jokes the owner dislikes.
Turn approved voice rules into a short card a writer can use beside a draft: audience, tone, permitted claims, banned claims, naming, call to action and review owner. A card is easier to check than a long transcript. For deeper work on maintaining voice across an agency, see keeping AI content consistent with a client's brand.
Collect images and video with provenance. For each folder, note who made the asset, what it depicts, where it may be used, whether people shown have given the needed permission, and whether any expiry or restriction applies. “Use anything from our shared drive” is not a reliable rights instruction. A menu photograph may show an old price; a staff portrait may be out of date; a customer testimonial may be approved only for a specific campaign. Mark uncertain assets as unavailable until the client clears them.
Write content pillars as repeatable questions
Content pillars are themes that make planning less arbitrary. Keep three to five, each tied to a customer question and an approved source of truth. For the café illustration, a set might be “what is on the current lunch menu?”, “how is a dish prepared?”, “what happens at a booked group table?” and “what should a first-time visitor know?”. The client should also name topics to avoid, such as personal dietary advice or a supplier claim that has not been checked.
A weak pillar is “educational content”. A usable version says “explain one ingredient or preparation choice from the current menu, with the chef or menu sheet as the source and no health claim”. The latter tells a writer what to verify and what a client must approve. AI can suggest candidate topics, but it should return a source-needed flag for each factual statement rather than invent a backstory.
Suppose the first week's illustrative ideas are a lunch special, a short preparation clip and a group-table question. AI proposes a caption saying the lunch special is available every day. The menu sheet says weekdays only. The reviewer changes the line before it reaches the client. This is a small factual error with a clear source; the workflow should catch it without making the client inspect every sentence from scratch.
Agree the sign-off route before AI writes the welcome pack
Decide what counts as approval: a named person clicking an approval action, a written response to a specific version, or another agreed record. “Looks fine” in a chat thread may be ambiguous when several versions exist. Give each planned item a status: drafting, internal review, client review, changes requested, approved, scheduled or published. The person who writes a caption should not be able to mistake “client review” for “approved”.
Planable is one scheduler built around this step. Its approval-workflows page says posts can stay hidden from clients until the content is ready, approval rights can go to client stakeholders, occasional approvers can view and comment without an account, and posts can be scheduled automatically once they get the green light. If your agency uses it, check in its help centre which approval setting stops a post being scheduled before sign-off on your plan, and whether making a post visible to a client also sends that client an approval request. Confirm both before you describe them in your own client instructions.
A realistic failure is a draft moved from internal to client-visible, followed by silence. The agency assumes the client saw it; the client was waiting for a request. Record the request time and feedback deadline, and confirm that the correct approver received it. If the deadline passes, pause the publication or use the escalation rule agreed in the contract. Silence should not silently become consent.
For sensitive posts, add a second review category even if ordinary content needs only one approver. Examples include a price change, a customer photograph, a complaint response or a claim about a service outcome. A simple policy can say “Routine posts: account lead checks facts, client approver signs off. Prices, customer images and disputed claims: business owner also signs off.” This is a process choice to agree with the client, not a platform feature AI can infer.
A filled onboarding record for a café client
Here is a complete illustrative hand-over. A small café asks a two-person agency to manage one feed and one short-form channel for an initial four-week cycle. The contracted plan calls for eight feed posts and four short videos. The client supplies menu facts and photographs by each Monday at noon. The agency sends drafts for review on Wednesday. The café owner is the final approver, with a manager as backup after an agreed hand-off. Publication begins the following Monday only if access, assets and approvals are complete.
The business goal is to make weekday lunch options clearer to nearby regular customers. The action on a relevant post is a table booking or menu enquiry through the channel's existing route. The content pillars are current menu items, preparation detail, practical booking information and staff introductions using approved portraits. The voice card says warm, specific, brief and free of unsupported food claims. The client prohibits posts about unavailable dishes and asks the agency to check prices against the current menu sheet before every scheduled item.
The agency's filled register shows two channel connections tested, one approval owner and one backup, 12 planned items, three approved photo folders and one pending portrait release. The portrait is excluded from the first calendar. The agency estimates 30 minutes for intake review, 45 for access tests, 45 for voice and pillar decisions, 60 for the first calendar and 30 for the welcome-pack review: 210 minutes of internal work in this illustration. These are planning assumptions, not industry averages or a service quote.
AI receives only the approved intake summary, with no passwords, private message history or personal contact details. It drafts a welcome email, a one-page operating guide and a missing-items list. The account lead checks all 12 planned-item counts, the two dates, named roles, asset restrictions and how approval works. The client corrects one assumption: the manager may approve only routine menu posts, not price changes. The account lead updates the record and the pack before the first draft is scheduled.
At the end of four weeks, the agency reports what it actually published and the measurements available from the agreed channels. If eight feed posts and four videos were planned but a portrait remained unapproved, the report should show the substitution or postponement. Do not let AI describe the full plan as delivered when the record shows fewer items. The next month's content decisions should come from the client's questions and the reviewed report, not a generic calendar generator.
Copy the welcome-pack prompt, then edit its output
The input to AI should be a compact, approved brief. Remove passwords, private customer information and unapproved asset links. The prompt below asks for a draft and a missing-items register. It does not authorise publication or assert that a role invitation has been accepted:
Draft a social media agency welcome pack from these approved facts only.
Client: a small café. Scope: eight feed posts and four short videos
in a four-week cycle. Goal: clarify weekday lunch options.
Voice: warm, brief, precise; no health claims. Owner approves all
price changes; owner or named backup approves routine menu posts.
Client provides current menu facts and cleared assets each Monday
at noon. Agency sends drafts Wednesday for review. First publication
is provisional until access and approval tests pass.
One staff portrait is pending permission and must not be used.
Write: a welcome email, responsibilities table in plain text,
approval steps, reporting outline, and missing-items list.
Mark any unknown date, name or permission as [to confirm].
Do not invent performance targets or claim access is complete.An illustrative draft output might say: “Welcome. We will prepare eight feed posts and four short videos over four weeks. Please send current menu facts and approved photos each Monday by noon. We will send drafts on Wednesday. Your owner will approve price changes; routine menu posts may be approved by the owner or named backup. The first publishing date is [to confirm] after access and review tests. We will not use the pending staff portrait.” That is the core of the pack, but it needs client-specific channel names, the agreed feedback deadline and a clear way to send corrections.
AI may add a cheerful line such as “We guarantee more lunch bookings” or quietly turn a provisional first date into a firm launch. Delete both. It may also treat the backup as authorised for every post, despite the price exception. Compare the output with the approved intake record line by line. Where the client has given no answer, leave “[to confirm]” visible rather than allowing the model to make a plausible guess.
Send a welcome email that assigns work clearly
A useful welcome email includes the scope, the first two dates, what the client must provide, who approves, how corrections are made and what happens if an item is late. It should be short enough to be read. The detailed guide can carry the full asset register, voice card and response rules. Use “[first name]” and “[your name]” in sample messages rather than inventing a client contact.
For example: “Hello [first name], thank you for confirming the four-week social plan. Please send the current menu sheet and cleared image folder by Monday noon. We will send the first drafts on Wednesday through the agreed review space. Please approve or request changes by Friday noon. The first publication date remains provisional until channel access and the approval test work. We will keep the staff portrait out of the plan until you confirm permission. Best, [your name].” Adjust every date to the real agreement before sending.
Contrast a vague reminder such as “Please send everything as soon as possible” with a precise one: “We still need the current menu PDF and permission for the group-table photograph. The menu holds three captions; the photograph holds one. If they are not cleared by Monday noon, we will use the already approved preparation clip and move the affected items to the next review.” The second message gives the client a decision and shows the effect of a delay without blame.
AI can produce the first version of both messages from the open-items register. The account lead still checks the dependency: perhaps the photograph is optional and should not hold any item. A sound welcome pack makes the consequences of missing inputs visible while leaving the client free to choose another cleared asset or a revised calendar.
Make the reporting promise measurable
Set the report date, recipient, channel measures and interpretation rule during onboarding. A sensible first report distinguishes activity from results: planned and published items, reasons for changes, audience interactions that the platform reports, enquiries the client can identify, and questions that should shape the next cycle. Avoid a promised uplift percentage unless it is part of a separate, evidence-backed agreement.
For an illustrative report, the agency planned 12 items, published 11, postponed one pending photograph clearance, recorded 18 relevant menu enquiries through the channel's inbox and received five table-booking messages that the client identified. Those figures are sample data, not a forecast. The report should also say how enquiries were classified and whether direct bookings outside that inbox are excluded. A plain statement of limits is more useful than a colourful AI paragraph claiming every interaction was a sale.
The onboarding record should tell AI which metrics may be used and where each comes from. If an export omits a period, mark the gap. If a scheduling tool and a native platform disagree, inspect definitions before averaging them. The fuller agency client-reporting workflow can help after the first reporting agreement is settled.
Check that the pack works before the first publication
Hold a short hand-over review with the client. Ask the approver to find one sample item, leave a correction and approve a harmless test or draft through the agreed route. Ask the account lead to show where a pending asset is marked and where a missed approval deadline appears. If either person has to search three inboxes to reconstruct the status, simplify the process before adding automation.
Use a seven-point release check: signed scope matches the pack; account roles work; brand voice is approved; each asset has a permitted use; the calendar has realistic facts and dates; the named approver can see and approve the intended version; and the report date and recipient are agreed. Record who checked each point and when. If one fails, keep the affected item unscheduled. The broader pre-publish check for AI drafted posts handles the content-level inspection after onboarding.
After the first cycle, ask which input caused the most back-and-forth. Perhaps menu changes arrived after drafts were prepared, or the backup approver did not receive requests. Update the welcome pack and intake questions around that real friction. A reusable template is useful precisely because it remembers these decisions for the next client; it should not force every client into the café's dates, channels or approval roles.
Questions clients ask during social media hand-over
Should a client send the agency its social media passwords?
Use each platform's role or partner access where available, with the minimum permissions needed for the agreed work. Record who grants access and test it with a harmless draft or view. Keep the client as an account owner and agree how access will be removed at the end of the engagement. Do not put passwords in an AI prompt or welcome pack.
Can AI write the brand voice guide from old posts alone?
Old posts are useful evidence, but they may include past campaigns, inconsistent tone or claims the client no longer makes. Ask the client for preferred wording, prohibited phrases and examples they approve. Let AI propose a short guide from those inputs, then have a named client reviewer correct it before anyone uses it for publishing.
When should the first post be scheduled?
Only after access works, the scope and content calendar are agreed, required assets and permissions are present, and the named approver has signed off the post through the agreed channel. A proposed start date in a welcome email is a planning target. If a dependency is missing, move the date and tell the client what is blocking it.
Further reads
- AI for Social Media Management: A Per-Client Workflow for Agencies — Keep each client's brief and approvals in its own workflow.
- How to Create On-Brand Social Media Graphics With AI — Build graphics only after asset and brand rules are agreed.
- Social Media Marketing AI Prompts for a Month of Client Posts — Draft the first month once the client brief is approved.
- AI Social Media Monitoring: A Monthly Insight Report for Clients — Add monitored audience questions to later reviews.
- Automate Client Onboarding With AI: Forms, Contracts, and Emails — Connect forms and folders after the manual hand-over works.
- AI Influencer Marketing for Agencies: Shortlist, Vet and Pitch — A scoring rubric, an eight-point fake-follower checklist, pitch prompts and templates, and the ad-label rules to settle before any creator fee.
- Agency Client Onboarding With AI: Questionnaire, Access and Kick-Off — One onboarding flow for an agency: an AI-tailored questionnaire, a safe access checklist and a kick-off agenda drafted from the answers.
- Nursery Welcome Pack: Letters and Settling-In Guide With AI — What goes in a nursery welcome pack, with prompts for plain-English letters, a settling-in guide, a key person introduction, forms and checked translations.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Planable approval-workflows feature page; LinkedIn Help, Understand admin access for LinkedIn Pages and LinkedIn Page admin roles. Checked 4 November 2026.