ai-search-gaps-to-social-campaign Checked

### 2. Baseline AI visibility

• DATA_getAiSearchOverview(target, source, brand?) — capture brand_presence, link_presence, ai_opportunity_traffic, and average_position. **Read previous before quoting change:** if it's null, this is the first snapshot — report the current values as a baseline and do **not** present the change_percent of 100 as real growth.
• DATA_getAiSearchLeaderboard(primary{target,brand}, competitors[{target,brand}], source, engines[]) — share of voice for the brand vs competitors, per engine. Build a quick heatmap (rows = brands, columns = engines).
• **This endpoint is heavy and can return a 504 timeout** when you pass many competitors × many engines at once. Query **one engine at a time** (or keep it to ≤3 competitors per call), and retry once on timeout. If it still fails, fall back to calling DATA_getAiSearchOverview for each competitor and compare brand_presence / link_presence yourself.

### 3. Find the prompt gaps

For the target and each competitor, pull the prompts behind the presence:

• DATA_getAiSearchPromptsByBrand(brand, engine, source) — prompts mentioning the brand by name.
• DATA_getAiSearchPromptsByTarget(target, engine, source) — prompts where the domain is cited as a source.

Compare: cluster prompts by topic, then mark each cluster as **owned** (target appears), **contested** (target + competitors), or **missing** (competitors appear, target doesn't). The missing and contested clusters are the campaign targets.

• **AI prompts almost always have volume: 0** — they're conversational queries, not search keywords. That is expected and is **not** a signal of low value. Judge a cluster by topical relevance and by *which brands the LLM cites*, never by search volume.
• **Validate brand-name matches.** A brand can surface in loosely related answers ("best year planner", a person's name, etc.). Read the answer text and flag ambiguous matches rather than counting them as real presence.
• Note *where* the target sits when it does appear (e.g. cited 4th of 6 in "best X" answers) — moving up within contested prompts is as valuable as entering missing ones.

### 4. Turn gaps into content hypotheses

For each target cluster, write a hypothesis: *"If we publish clear, citable content asserting [brand] in [narrative], we should start appearing for prompts like [examples]."* Translate each into social angles that make the brand's position explicit and quotable — definitions, head-to-head comparisons, "X vs Y", myth-busting, FAQ-style answers. LLMs favour clear, structured, attributable claims, so write social copy that states the position plainly rather than burying it.

Present the clusters and hypotheses to the user before drafting.

### 5. Draft and create in Planable

Write platform-appropriate copy, then create drafts: create_post per page (per-platform copy) or create_grouped_post for synced content. Tag the batch with a label (via list_labels / create_label, e.g. "AI-visibility") so the campaign is easy to isolate when measuring.

**Scheduling — ask before creating.** Don't guess dates or leave everything undated by default. Ask how the user wants the batch dated and offer: **spread evenly** across a window (e.g. the next 7 days, one post per slot at a sensible hour), a **fixed cadence/interval** (e.g. every weekday at 10:00, laid out from a start date they give), **manual** dates per post, or **no dates yet** (undated drafts to place on the calendar later). Convert each chosen time to ISO 8601 and pass it as scheduledAt. Keep posts as **proposed drafts** — don't set publishAtScheduledDate — so nothing auto-publishes; only set it true if the user explicitly wants auto-publishing. Scheduled times are treated as **UTC**, so confirm the timezone or state that times are UTC.

### 6. Instrument before/after measurement

This is what makes the loop real:

• **Ongoing AI tracking (if a project exists):** create an AI Result Tracker engine with PROJECT_createLlmEngine, add the target prompts with PROJECT_addPrompts(site_id, llm_id, prompts[]), then read movement later with PROJECT_getPromptsRankings and PROJECT_getLlmStatistics. Because this writes to the user's live project (and consumes plan limits), confirm before creating engines/prompts.
• **Periodic re-checks:** re-run DATA_getAiSearchOverview and DATA_getAiSearchLeaderboard after the campaign has run and diff against the baseline from step 2.
• **Social side:** get_post_metrics_summary(workspaceId, pageIds, startDate, endDate) on the labelled campaign posts shows the engagement the content earned.

Content pointers: writing for keywords & AI visibility gaps

Keep these in mind when creating social content meant to target a specific keyword or close an AI-visibility gap:

• **Target one intent per post.** Pick a single keyword or question and answer that one thing clearly. Posts that try to cover everything rank and get cited for nothing.
• **Lead with the answer.** Put the takeaway in the first line, then support it. Skimmers and AI engines both extract the clearest, most self-contained statement — don't bury it.
• **Write the way people actually ask.** Phrase hooks, captions, and headers as real questions and plain-language answers. AI prompts are conversational, so natural phrasing beats keyword-stuffing.
• **Make claims quotable on their own.** AI tools lift snippets out of context, so each key sentence should stand alone — one idea, declarative, no "as mentioned above."
• **Be specific.** Numbers, concrete examples, named steps, clear definitions. Specificity is what gets cited and what sets you apart from generic content competitors already own.
• **Fill the gap, don't echo it.** If a competitor already owns a topic, find the sub-question or angle they're