1. The pitch in one paragraph

Today the report personalizes from two inputs: the quiz answers and a shallow read of the customer's website. It never looks at their actual social accounts. This feature adds that: we find their accounts automatically (from their website's footer and any URLs they typed), read the public data, and open the report with proof of homework: "We looked at your Instagram. You've posted 4 times in the last 90 days, your bio has no link, and your last 10 posts average 2 comments. Here's what that tells us." Nothing else a $29 report can say is more convincing, and every problem the audit finds is a pre-written sales argument for the management service. An audit that finds problems sells the person who fixes them.

2. What the audit measures

Six deterministic checks per discovered account, each mapping to a fix the report (and the management pitch) can name:

• Found or missing. Which platforms they're actually on, including accounts they forgot to mention. Also the reverse: no presence on a platform our verdict ranks #1 for them.
• Bio and link health. Is the bio complete, does it say who they help, is there a working link, does that link go anywhere useful (their site, a lead magnet) or nowhere.
• Posting rhythm. Posts in the last 90 days versus the cadence our report prescribes. "You post twice a month; this plan calls for 3 a week" is the consistency conversation made concrete.
• Engagement reality. Likes and comments on recent posts, stated plainly and kindly. This grounds the report's advice in their actual starting point instead of a guess.
• Effort-to-fit mismatch. Active on a platform we'd tell them to skip, silent on their best-fit platform. That's the single most persuasive audit finding for our verdict-first report.
• Recency. Days since last post. A dead account and an active account get different advice.

All six are arithmetic on fetched public data. No LLM guesses any number, and the audit facts join the fabrication gate's provenance so the report can legitimately cite them.

3. What's actually feasible, platform by platform

Verified against current (July 2026) sources. Grades answer: given a public profile URL, can we get follower count, bio, link, 90-day posting frequency, and recent engagement?

Cross-check on the category: productized audit tools (Metricool's competitor tracking and friends) all do exactly this: public data for accounts the user doesn't own, no login from the audited account. This is an established product pattern, not a gray-area invention of ours.

Legal posture, one paragraph. Logged-out scraping of public profile data is on its strongest legal footing ever: the hiQ ruling said it doesn't violate the anti-hacking statute, and Meta lost its own terms-of-service argument against Bright Data in 2024. What actually gets companies killed is fake accounts and logged-in scraping (that's what ended Proxycurl). Our shape (low volume, public data only, one-shot use inside a paid report, providers bearing the scraping risk) is about as safe as this category gets. The practical risk is a provider breaking, not a lawsuit, and the design treats every platform fetch as optional: if a fetch fails, the report ships without that account, exactly like website research today.

4. The phased build

Phase 1: discovery plus the free platforms (one build session, $0.00 per report)

• Link discovery. Extend the existing website fetch to also parse the page for social profile links (footer icons, header links). Combine with whatever they typed in the quiz. Pure code, no new dependencies.
• YouTube audit via the official free API (one new worker secret: a Google API key).
• Substack audit via its public JSON. Both feed the six checks above.
• The report section ("What Your Current Accounts Told Us"), rendered deterministically like the milestone tables, placed right after the platform verdict. Findings phrased as opportunities, never scolding. Plus one line in the lead email so a management inquiry arrives with its audit attached.

Phase 2: the money platforms (one build session, ~$0.03 per report)

• Instagram + TikTok via ScrapeCreators (one vendor, plain HTTPS, pay-as-you-go, 100 free credits to prototype; Apify as the fallback vendor).
• Twitter/X via twitterapi.io where the customer has one.
• Audit cost lands around 3 cents on a $29 report: 0.1 percent of revenue.

Phase 3, only if beta feedback asks for it

• Facebook Pages via provider now, official Meta access application running in parallel.
• LinkedIn company pages with expectations set (partial data). Personal LinkedIn profiles and Pinterest stay out until someone actually asks.

5. Where it lands in the product

• Pipeline: a new Step 0.5 alongside website research: discover accounts, fetch public data, compute the six checks. Fail-open per platform, hard time budget, runs on the queue where the 30-second limit doesn't apply.
• Report: the new section right after "Your Best Platforms, In Order," so the verdict is immediately backed by evidence from their real accounts.
• Prompts: audit facts join the diagnosis and playbook prompts as a provenance source, so the LLM can reference "your last 90 days" truthfully and the fabrication gate can verify every number it cites.
• Funnel: the growth-path CTA and the lead email both carry the audit's top findings. Annette opens a management inquiry already knowing what's broken.
• Quiz: unchanged. The social URLs question becomes optional-but-helpful instead of the only source.

6. Decision needed

Phase 1 needs only a free Google API key and one build session, and it makes every report visibly smarter. Phase 2 needs a ScrapeCreators account (pay-as-you-go, no subscription) and roughly 3 cents per report. My recommendation: green-light Phases 1 and 2 together as the next build, and hold Phase 3 for beta feedback. The technical plan for the build session lives in the repo at research-notes/plans/plan-7-social-audit.md.