Research Prompts

You are a senior analyst supporting the Undoctored Truth Alliance (UTA) steering committee. Produce a 30-day baseline of UK-relevant health misinformation across Twitter/X, TikTok, YouTube, and Facebook.

Scope
- Tier 1 topics: vaccines, cancer, mental health, nutrition.
- Tier 2 (note only): drugs, reproductive health, chronic disease, AMR.
- UK-relevant: posts authored in the UK, in English, or amplified by accounts with ≥10% UK audience share.

Deliverables
1. Prevalence: % of sampled health content that contains a verifiably false or unsupported claim, broken down by platform and by Tier 1 topic.
2. Top 20 narratives: short headline + 1-line summary + representative URL + estimated reach.
3. Coordination signals: clusters of accounts repeating the same claim within 24h, suspected platform-of-origin, and any cross-platform spillover.
4. Harm-likelihood scoring: 1–5 per narrative, based on (a) clinical severity, (b) actionability of the false claim, (c) audience vulnerability.

Output
- Structured JSON matching the schema in /docs/baseline-schema.json.
- A one-page executive summary in plain English for non-technical committee members.
- A list of caveats: sampling bias, platform API gaps, anything you could not verify.

Conduct a SWOT-style analysis of where UTA can credibly intervene in the UK health misinformation landscape over the next 18 months.

Compare UTA's potential positioning against existing actors:
- Full Fact (general fact-checking, established brand)
- Logically (commercial detection, government contracts)
- NHS communications (trusted but reactive, limited social-platform reach)
- Ofcom under the Online Safety Act (regulatory, slow cycle)
- Academic groups (King's, Cambridge, Imperial — research, not intervention)

For each actor: what they do well, what they do not cover, and the implication for UTA's positioning.

Then explicitly contrast two strategy archetypes:
A. Awareness-metric strategy — optimised for reach, impressions, and media coverage. Fast to demonstrate activity, slow to demonstrate harm reduction.
B. Outcome-metric strategy — optimised for measurable shifts in NHS-tracked behaviours and literacy indicators. Slower to start, defensible to funders.

Recommend 2–3 differentiated positions UTA could occupy, each with: target audience, primary measurable outcome, key risks, and 6-month proof point.

Draft a v0 technical spec for a live UK health misinformation monitoring pipeline that could replace the indicative data in this dashboard within 90 days.

Required sections
1. Data sources — Twitter/X API tier, TikTok Research API, YouTube Data API, public Facebook/Instagram via Meta Content Library, plus a Telegram/Reddit ingestion fallback. Note rate limits, auth, and known coverage gaps for each.
2. Ingestion cadence — near-real-time vs hourly batch; backfill strategy for the previous 12 months.
3. Classification approach — hybrid keyword/heuristic prefilter feeding an LLM classifier (Claude or GPT-4-class) with a structured rubric. Include the rubric (claim, evidence, verdict, harm-tier).
4. Human-in-the-loop review — sampling rate, reviewer interface requirements, inter-rater reliability target (κ ≥ 0.7), and escalation rules for high-harm items.
5. Dashboard surfaces — map outputs to the existing UTA Intelligence views: Baseline Overview KPIs, Topic Heatmap matrix, Channel Analysis volume + radar.
6. Evaluation harness — gold-standard set of ≥500 hand-labelled items, weekly precision/recall reporting, drift monitoring on classifier outputs.
7. Governance — data retention, PII handling, model card, and the editorial independence guardrails from the UTA Funding Principles.

Output as a markdown document suitable for circulation to a technical partner for scoping.