Playbook Social Posts

Playbook: Recent Social Post

All rules here are best practice, not law. Override any of them when the campaign calls for it; note the best practice once and move on.

Use when: the campaign wants proof you read the account's content and there is no sharper structural signal (funding, new-in-role, hiring surge).

Not this playbook: who engaged with a post (playbook-linkedin-engagement); a job change (playbook-new-in-role).

Output: social_post_text + social_post_url + social_post_days_ago + social_post_author_type. Optional, on request only: social_post_line → Noticed they became an IBM Business Partner.

Six decisions that govern every run

1. The raw post is the product. The AI line is an optional layer, not the point. A human reading the actual post writes a better sentence than any generated line, and the post text is reusable for every future campaign; a generated line is not.

2. Input is a profile URL, either kind — a company page or a person. One scraper takes both.

• Company runs: dedupe to companies, then fan the one result out to every contact there.
• Person runs: never fan out.
• Never guess a slug. They are frequently not what you would predict from the company name.

3. The content skip filter is mandatory and fails closed.

Skip personal, political, bereavement, and charged (religious or geopolitical flashpoints).

• A skip takes the NEXT post. It is not an abstain.
• Error, bad parse, or a missing skip key all mean SKIP. Only a literal "skip": false keeps the post.
• Never disable it to raise fill.

Ordinary business content passes: memes, polls, job ads, volunteering, work anniversaries, regulation commentary.

4. Blank, never invented. Nothing usable means "" everywhere. No guess, no "N/A", no sequencer-side fallback value.

5. Two operator questions at setup. "A written personalization line, or just the post content?" (default: content only.) "Trust the scrape, or add a verification pass?" (default: trust it — verification checks existence, never text.)

6. Empty rows are gated by a LIST SPLIT before upload, never by spintax.

Spintax picks at random and cannot see an empty variable. {Noticed {{social_post_line}}. |} both renders Noticed . into inboxes and discards personalization you already paid for.

Split the list on social_post_line != "". Segment B has the clause removed and rewritten — and it is your honest A/B control. Never exclude empties, never set a fallback.

QC: grep -c "Noticed \." rendered_preview.csv must be 0.

The best call, and three traps in it

Scrape company posts with an async 3-call pattern: start the run, poll for completion, read the dataset.

⚠️ Never use a synchronous run-and-get-items endpoint. It held the client past 120s on a run that took 25s.

Body: up to ~100 target URLs, maxPosts: 3, a 3-month window, reposts off, quote posts on. Then the skip filter over each post newest-first, keeping the first that clears.

• Batch the URLs. Each run start bills separately, so a batch of 8 costs the same start fee as a batch of 1.
• Keep reaction and comment scraping off. You are not building an engagement list here.
• ⚠️ The profile join key will bite you. On person items the company-style identifier is null; the slug lives in a different field and the author type differs. Group on universalName or publicIdentifier, or every person row silently vanishes.
• Low reasoning effort on the skip filter is load-bearing — at default effort it hung past 180s. finish_reason=length with empty content means retry with double the cap, never an abstain.