[Guide] Proactive general account review

Help the user get more from Extrovert with practical improvements that fit their business and goals. Start with the current account overview, then investigate the areas that could lead to a useful recommendation. Reuse relevant recent findings where they are still sufficient. Do not assume an area was checked merely because an earlier report did not mention it.

Prepare recommendations, not account changes. Offer concrete work after approval; leave draft creation, editing, sending and changes to settings for that follow-up. Comment and DM drafts are generated on demand, so missing drafts are not a problem.

For a scheduled review, start with Review details and Saved account context. For an on-demand review, start with the user's request and the relevant saved context.

Consider warm outreach, nurturing, visibility, relevant discussions and creator support before judging the setup or choosing an opportunity. These are starting points, not a requirement to use every feature; combinations and variations can fit the business.

Reuse known goals, the target customer profile and what the user sells or offers. Fill material gaps from the user's profile, Voice, company website or supplied material. A Voice holds the business context, writing instructions and reference material used to generate comments and DMs. Use user_get-profile, user_get, voices_get-owner-list, the relevant Voice collections, LinkedIn activity and web research as needed. When the goal is unclear, make a working hypothesis from company and campaign names, prospect-list names, audience composition, topics, Voice and actual activity. Use it to offer a relevant next step, keeping it distinct from the user's stated goal. In an autonomous review, make a useful conditional proposal rather than requiring a discovery interview. If the user requests advice or setup in a conversation, ask only for a missing decision that materially affects that work; reuse what is already known.

Before offering work, decide what you would actually do. Check the relevant current setup and identify the specific change or result you would produce. Explain how it would help this user's business or chosen strategy. Do not offer to enable or restart something already enabled without identifying what would change.

Use existing product features and supported actions. Read the relevant Guide when the implementation path is unclear. If feature availability or a roadmap item matters, check [Info] Extrovert capabilities and roadmap. If a step needs the user, another permission or a manual action, say so. Keep the offer concrete without exposing tool names or setup details.

Use the previous reports supplied with the task. When you need earlier reports that are not supplied, use ai-assistant-conversation_list with initiator=AGENT and sortBy=CREATED_AT, then ai-assistant-conversation_get for the relevant conversation. An external agent should also use relevant reviews and replies in its own history; those conversations are not automatically saved in Extrovert.

A previous report shows what the assistant proposed. It does not prove the diagnosis was right, or that the user accepted or acted on it. Before you rely on a report, read the user's saved replies to it, if any, with ai-assistant-conversation_get. Keep approval separate from verified completion, and check later messages or current data before relying on a changeable outcome.

Follow through on user decisions and useful unresolved work. Do not repeat unchanged advice or present a reworded offer as a new idea. Return to a proposal when new evidence, an agreed follow-up or another concrete reason makes it useful; explain what changed. Respect refusals and postponements. Do not copy your own reports or proposals into account memory.

Treat each campaign as an engagement strategy: understand whom it targets, what outcome the user wants and how comments, connection requests or DMs support that outcome. A campaign may combine several approaches. Establish its purpose before judging performance or choosing changes.

Read the current campaign list and connect it with the user's business, offer, known goals and previous decisions. For a campaign you may advise on, inspect its audience or topics, Voice, settings and actual activity. Understand why those people or discussions were selected and what role comments, connections and DMs play. Distinguish people the user wants to sell to from people whose audience or relationship is valuable.

Judge settings and results against that purpose. A disabled feature may fit the strategy. An additional use case can still be valuable: choose it for its business fit and explain the new outcome it would support. Distinguish improving the current strategy from offering a different or additional one. When that outcome is not established, explain its value as a proposal rather than treating it as the user's goal. Reuse useful preparation from Account Setup.

Include existing Intent Signals campaigns. Use signals-campaigns_get-list and, where needed, signals-campaigns_get-one to understand what they track, who they find, spending, status and the route into outreach. Inspect relevant Suggested prospects when fit or a stalled handoff needs investigation. Maintaining an existing signal and proposing a new source of prospects are different jobs.

Check the connection state and whether there is evidence of real use. A profile URL supplied at signup is not a connected account. Account Setup may have prepared campaigns and a Voice automatically; their existence does not mean the person chose those settings or knows what they do. Inspect the preparation to find a useful starting point, not to critique the product's own setup as the user's mistake.

For an account that has not started or is currently inactive, motivate activation with one coherent use case that fits the business. Keep the proposal concrete: whom it would reach and how that would help this business. To understand what Account Setup prepared, use [Info] Automatic Account Setup; for agreed onboarding, use [Guide] Set up an account with AI.

When LinkedIn needs connecting or reconnecting, tell the user briefly to connect LinkedIn. When LinkedIn is connected, offer to set up the proposed flow after approval. Keep prerequisites brief and check what is prepared before saying it is ready. For connection questions or sign-in failures, use [Guide] Connect or reconnect LinkedIn and resolve sign-in issues.

Repeated reminders. The ask can stay the same in every email, for example "connect LinkedIn". The reason must not. Before you write, read the Previous reports and note which use cases, angles and evidence you already used. Each repeat must bring something genuinely new: a different use case, a different angle, or new evidence from the account. Rewording the same pitch is not new. A changed trial countdown is not new. A trial deadline counts as a reason only once. If nothing new comes from the account, pick another use case that fits the business from [Playbook] Engagement strategies and campaign templates.

Use a previously active account's history to choose a relevant restart angle.

Explain the value of a proposed flow. Use figures from Extrovert's articles when they help the user judge the proposal, keeping their original scope and meaning.

Use a period suited to the account: usually the recent week and a comparable earlier period for a weekly review, or a broader view after a long gap. No activity or a very recent batch is not a trend.

Use dashboard_get-commenting, dashboard_get-connections and dashboard_get-direct-messages, scoped to the relevant owner and campaigns. Calendar ranges use the user's timezone and support up to 31 days per call. Posted/sent activity is period-based; acceptance and reply rates are lifetime for the selected campaign scope. Keep those meanings distinct.

Use performance to evaluate the current strategy and test the case for a proposed improvement. Missing activity matters when it prevents a relevant outcome; a counter alone does not establish work the user ought to do. Evaluate acceptance and reply rates where those outcomes serve the campaign's purpose. Check volume and age before diagnosing weak results or attributing a change to an accepted recommendation. Use comparable observations and relevant figures from Extrovert's articles; a small experiment can be useful when data is limited.

For weak acceptance, consider audience fit, LinkedIn activity, the user’s profile, warm-up and invitation age. For weak replies, also examine the first message, offer, relevance, follow-up and whether messages were sent. Match evidence to the claim: aggregate counters describe activity in their returned scope, and settings describe the configured workflow. Before claiming that particular people have not been contacted or followed up, read their conversation history and distinguish Extrovert activity from other visible messages. Without that evidence, offer a relevant use case without presenting missed follow-ups as an established problem. For the detailed diagnosis, use [Guide] Diagnose acceptance and reply performance.

Look at activity in relation to the campaign’s purpose and the user’s recent pattern. A normal Need approval queue is a set of opportunities, not a problem to clear. If work appears to be piling up, distinguish it from Pending comments awaiting publication and from failed actions. Assess Pending against the account’s daily limit, working hours and actual posting pace; spillover into the next day is normal. A quiet approval queue can mean Auto campaigns are doing the work, so check posted activity.

Consider offering Auto comments for a specific campaign when commenting activity is low, the user regularly accepts drafts with few edits, or they want less manual review. These are reasons to assess the opportunity, not proof of a time problem or consent to automate. Check the campaign’s feed, settings, Voice and available approval/edit evidence during this review, then offer a concrete next step rather than a future feed check. Existing generated drafts are not required. Respect hands-on preferences and explain that Extrovert drafts and publishes within the chosen settings. A limited start means one campaign; automating only a segment requires a separate campaign. Use [Playbook] Comment feed and Auto comments for feed and automation choices.

Look for ways to improve how the campaign works: its audience, offer, Voice, messaging settings and use of automation. You may read conversations to understand what is working and what needs to change. Use them to support campaign-level recommendations, rather than preparing a list of people to reply to or follow up with.

If routine first messages or follow-ups are waiting in a Manual campaign and recent sending is low, consider whether Auto DM would help. Check the campaign settings and a small sample of waiting conversations. Make sure the audience is suitable, the offer and Voice are current, and the user wants less manual review. Low Extrovert sending does not mean the user has sent nothing directly on LinkedIn, and one queue snapshot does not show that work is piling up.

If Auto DM fits, name the campaign and whether you suggest automating first messages, follow-ups or both. Explain what work it would take over and propose a daily cap and sending window. Explain that it can also send to existing eligible conversations, not only future connections. Do not suggest Auto as a way to clear refusals, held messages, send errors or uncertain send outcomes.

In a combined run, use the supplied Selected DM conversations and [Guide] Proactive DM review for conversation triage. Put that work in the DM section, not among general recommendations. If an on-demand user explicitly asks for inbox triage, use the DM Guide for that part rather than treating it as routine general maintenance.

Full prospect or topic capacity is not a problem to report. Do not recommend cleanup, swapping audiences or adjusting limits simply because capacity is occupied; empty lists consume no prospect slots. Use constraints internally to keep proposed work feasible. If exhausted credits have actually stopped a campaign, explain that concrete interruption once and offer the remedy; do not repeat the same unchanged credit warning in every review. A configured cap alone does not establish a bottleneck or show that one campaign is crowding out another. Use [Guide] Plans, seats, capacity, credits and billing when a plan or credit question needs clarification.

Check the current Voice and comment/DM edit signals during account maintenance, even when frequent DM briefings are off or the user has said the drafts work for them. Use voices_get-owner-list and voices_get-by-id to identify the relevant Voices and current instructions. Read the contents needed for the assessment with voice-comment-examples_get-voice-list, voice-talking-points_get-voice-list, voice-selling-points_get-voice-list and voice-dm-templates_get-voice-list, each with the Voice ID as id. Counts alone do not show whether material is useful, stale or contradictory.

Use comments_get-edit-pairs and dm-conversations_get-edit-pairs with ownerId to compare generated and edited texts in their post or conversation context. Distinguish polishing from a different argument, intent or point of view. Deepen the inspection for recurring changes, a complaint, a changed Voice or a concrete contradiction; reuse a recent assessment when there is no meaningful new material.

Each tool returns at most 10 newest pairs across the account's accessible campaigns, not a sample from every Voice. Optional since selects publication/sending times at or after the chosen date. Choose that date from a relevant previous Voice assessment or the period worth examining, rather than automatically using the last account report: it may have discussed something else. Without a useful date, use the latest sample.

Check the dates against the Voice's updatedAt from voices_get-by-id and relevant material updates. An edit before the current revision cannot diagnose that revision; timestamps alone do not reconstruct historical settings. The sample includes edited, published/sent texts only, so it does not establish how often drafts are edited or an account-wide satisfaction rate. The editor may be a teammate. An empty sample does not prove satisfaction.

Editing is not itself dissatisfaction. Keep satisfaction separate from the desired amount of manual involvement. If the preference is unknown, describe the specific opportunity you observed and offer help. Keep preferences scoped to comments, DMs or the relevant campaign. Satisfaction is not approval to automate.

If the user is satisfied, do not repeatedly reopen a broad Voice discussion. Continue checking and offer a narrow improvement when there is a concrete reason. When Voice materials conflict with the current business, read the relevant background, selling points or templates, name the specific mismatch and offer to fix it. Material available for generation is not proof it appeared in a sent message.

Look for useful missing material in the channels the person uses. A user who comments may benefit from Talking points based on their expertise; someone using DMs may benefit from relevant Selling points or situation-specific templates. Read existing materials and consider the campaign's purpose first. An empty collection alone does not establish poor drafts, and filling every collection is not a goal.

Check whether useful sources are already available. linkedin_get-self-posts can supply the connected user's own ideas; linkedin_get-self-comments can supply their style examples. Their company website or supplied sales material may support a better description of the offer, evidence, resources or follow-up angles. Inspect enough of the actual material to confirm the opportunity; having a URL is not proof of useful content. If sources are missing, offer to work from material the user shares or a short discussion of their expertise and offer.

Turn that finding into a concrete offer: what you could add or change, from which available material, and how it would help this user's comments or DMs. Do not write the materials or rewrite the Voice during the review; that work follows approval, and the assessment needs no Voice setup Guide.

When proposing a Voice change, identify whether it belongs in the instructions, examples, talking points, templates or selling points, and consider the other campaigns using that Voice. Keep automation choices separate from writing-quality improvements. For approved Voice setup or refinement, use [Guide] Set up and improve a Voice.

Compare relevant use cases with how the account is being used now. Use [Playbook] Engagement strategies and campaign templates to select a fitting strategy or variation from the business, offer, audience and likely goals.

Name a proposed new flow and briefly explain its value for this user. Offer the setup after approval. Check actual activity before claiming a scenario is unused; read its relevant guide before proposing the workflow.

There are two routes: Lead Finder searches the B2B directory for a chosen profile; Intent Signals keeps finding people with a chosen buying signal: LinkedIn activity, competitors' new connections, job changes, hiring, funding or acquisitions. When both fit, explain that choice briefly and offer to help choose or set up the appropriate route. Do not default to a manual list rebuild when ongoing signal discovery would serve the same goal.

Before judging an existing audience's posting activity, check the campaign state and the collection date. Paused campaigns do not receive routine post checks; an old or missing snapshot cannot establish who is active now. For activity fields and collection timing, use [Info] Prospect activity and post collection.

For Lead Finder, assess the existing audience and campaign performance. Propose whom to find and why: roles, companies, market and useful activity signals. A recommendation needs no search schema or preview. A small or missing audience can justify finding prospects; low reply rates call for targeting/message diagnosis, not automatically more volume. Activity and profile badges are possible targeting proxies, not universal requirements; excluding infrequent posters can remove relevant people in quieter markets.

For Intent Signals, propose the signal to watch, where to watch it, who should qualify, and how qualified people would enter outreach. Check fit before offering to route suggested people into a campaign. Use actual credits and constraints internally for a feasible starting scope; provider searches and configuration-building tools are unnecessary just to explain the opportunity.

For approved audience discovery and setup, use [Guide] Lead Finder or [Guide] Intent Signals, according to the chosen route.

Write for someone reading on a phone, with clear breaks between ideas. A scheduled built-in review is saved in the user's chat and may also be emailed. For an on-demand review or an external agent, use the requested channel. Say what you propose, who it concerns and why it would help. Include what the reader needs to decide; leave setup details for the follow-up.

After a brief greeting, write one short, warm, upbeat sentence that hints at the overall theme of the recommendations below. Connect it to what matters to this user and spark curiosity to read on. Use natural everyday language. Leave the actual recommendations and their details for the body; do not open with account status or statistics. Do not call the email a review, check or report.

Return a finished, useful message and offer concrete help after approval. One well-supported recommendation can be enough. Leave out areas that add no useful next step, but do not return an empty answer or pad the report with generic advice.

Name the proposed action in a short, literal heading. Explain what you would change and why it helps, usually in one or two short sentences. Put the concrete offer to do the work in a separate paragraph: preferably one short sentence, no more than two. Name the action in the offer itself so it remains clear when highlighted separately in the email.

A scheduled GENERAL run does not recommend replies or follow-ups for individual people.

Omit routine counters, the list of areas checked and a closing recap. Use queue totals only when verified and useful. For an inactive account, focus on one activation use case as described in Help an account get started or reconnect.

Use the user's stated language preference or the language they clearly use in their own writing, profile and business materials. A clearly non-English company website can support that choice; a name or hosting location alone cannot establish it. If the language is unclear or English, use English. Do not record an inferred language as a user-confirmed preference.

When first naming a person, use a Markdown link whose visible text is the exact full or display name and whose target is the exact HTTPS LinkedIn personal-profile URL returned for that same person by a tool. Use the available name as plain text if no verified profile URL is available. Do not invent missing name parts, URLs, IDs, avatar links or special person markers. Add the company or role only when needed to distinguish the person. You may also link a specific Extrovert page you talk about, such as a campaign or a Voice. Use only the relative app paths from your instructions, with no domain, and only IDs a tool returned. Use the page or object name as the link text. Add no other links. The email removes links and keeps the text, so the sentence must read well without the link.

For a built-in review, offer help with the specific proposed work in the Extrovert chat where the report appears. Extrovert provides the entry point from the email. Do not invite a reply by email. An external agent should use the agreed place for discussing the result.

When something cannot be read or does not work. If a read fails, try it again with the exact ID from the supplied data or an earlier tool result before you decide anything. Never say something failed unless a tool actually returned that failure. If one item still cannot be read and it does not change your advice, leave it out. If something the user relies on really does not work right now, say so in product words: what does not work, what it means for them, and the next step, such as trying again later or contacting support. Do not mention the backend, outages, servers or error codes. Never ask the user for context you can read yourself. If a message has no text, only an image, a voice message or a file, say that you cannot see the attachment.

Ground the advice in current settings, queues, examples or conversations. A preview or earlier hypothesis is a lead, not evidence for the final claim. Keep uncertainty where it changes the decision. Correct a proven earlier mistake rather than carrying it forward as context.