Research brief: Confidence Levels, Sources, and Dated Claims — why every statement on a credible site should be verifiable (piece 15 of 15)

Status: Closing piece of the 15-brief roadmap. Research material — not a finished article. Compiled May 22, 2026.

TL;DR

• In 2026, unsourced marketing copy is simultaneously a trust liability (Edelman 2025: 7 in 10 globally believe leaders deliberately mislead), an AI-visibility liability (Profound: ChatGPT pulls 7.8% of citations from Wikipedia; Perplexity API returns title+URL+date+snippet for every source), and a legal liability (FTC requires "reasonable basis" before any objective ad claim).
• The discipline that fixes this is not "add footnotes" but a borrowed editorial stack: BBC/Reuters/SPJ named-source rules, Wikipedia's "verifiability not truth," the US intelligence community's high/moderate/low confidence taxonomy, COPE's retraction guidelines, and Perma.cc/DOI-based link-rot mitigation — all operationalized as confidence labels, dated claims, and an inline citation pattern library.
• Candid Creative's default: every objective claim sourced (verbatim quote ≤25 words + URL + date + archive snapshot), every claim confidence-labeled (Verified / Industry-consensus / Single-source / Estimated / Author's view / Contested / Stale), every correction visibly logged, and a public Sources & Methodology page for marketing pages that derive from KB research.

The 15-piece roadmap closes here for a reason

Pieces 1-4 set the strategic frame. Pieces 5-7 set the infrastructure. Pieces 8-9 defend longevity. Pieces 10-11 establish data as the asset. Pieces 12-14 operationalize the asset. Piece 15 is the editorial layer that turns all of the above from a content-strategy story into a credibility story. Every piece in the roadmap makes claims; this piece is the one that says how those claims must be made.

It is piece 15 rather than piece 1 because it can only be specified once the rest of the system is described — but it underpins the rest in practice.

Honest caveats

• The AI-citation premium for sourced content is plausible but not yet rigorously proven at RCT level. Profound, Perplexity API design, Semrush AI Overview studies are strongly suggestive but largely vendor-produced. Confidence: Industry-consensus, not Verified.
• Confidence-labeling research is from intelligence/forecasting contexts (Tetlock, ICD 203) and translates only partially into marketing — the labels in CANDID REFERENCE: 7-label confidence taxonomy — Verified / Industry-consensus / Single-source / Estimated / Author's view / Contested / Stale are adapted, not directly imported.
• The "LLMs 34% more confident when wrong" circulating with "MIT research, January 2025" attribution: closest verifiable primary source is Simhi et al. (Technion/Oxford/Hebrew U), arXiv:2502.12964, February 2025. Treat the institutional attribution as Single-source/Contested; the core finding (LLMs hallucinate with high certainty) as Industry-consensus.
• ClaimReview's status is in flux: schema persists and remains valuable for AI ingestion; Google's June 2025 retirement of the rich result removes the visible incentive. Implement for non-Google AI engines; don't overinvest in Google-specific FactCheck markup.
• The Reuters Handbook of Journalism was last comprehensively revised in 2012; principles remain in force at Reuters but the document is older than this brief's default 2023 freshness threshold. Included because it is the primary editorial-standards document for Reuters and no successor has replaced it.
• The FTC's substantiation doctrine (1971/1984) is currently being challenged in Xlear v. FTC (filed late 2024, citing Loper Bright Enterprises v. Raimondo). Absent injunction, the doctrine remains in full force as of May 2026. Monitor for outcome.